Method and apparatus for indexing, searching and displaying data

Abstract

A computer research tool for indexing, searching and displaying data is disclosed. Specifically, a computer research tool for performing computerized research of data including textual objects in a database or a network and for providing a user interface that significantly enhances data presentation is described. Textual objects and other data in a database or network is indexed by creating a numerical representation of the data. The indexing technique called proximity indexing generates a quick-reference of the relations, patterns and similarity found among the data in the database. Proximity indexing indexes the data by using statistical techniques and empirically developed algorithms. Using this proximity index, an efficient search for pools of data having a particular relation, pattern or characteristic can be effectuated. The Computer Search program, called the Computer Search Program for Data represented in Matrices (CSPDM), provides efficient computer search methods. The CSPDM rank orders data in accordance with the data's relationship to time, a paradigm datum, or any similar reference. An alternative embodiment of the invention employs a cluster link generation algorithm which uses links and nodes to index and search a database or network. The algorithm searches for direct and indirect links to a search node and retrieves the nodes which are most closely related to the search node. The user interface program, called the Graphical User Interface (GUI), provides a user friendly method of interacting with the CSPDM program and prepares and presents a visual graphical display. The graphical display provides the user with a two or three dimensional spatial orientation of the data.

Claims

What is claimed is:

1. A method for using active links within the data of an object stored in a database of a computer so that a user may jump from viewing the data of the object in the database to a position outside the object in the database and outside the computer, comprising:

storing one or more links within data of the object in the database to positions outside of the computer, wherein the stored links are active links;

displaying the data of the object within the database, wherein one or more active links are displayed with the data from the object in the database, wherein positions are nodes in a network that may be accessed, the active links including hyperjump links between nodes in the network and the objects, and the step of displaying comprises:

generating a source map, wherein the source map represents hyperjump links that identify a chosen node as a destination of a link, and

wherein the method further comprises activating a link represented on the source map, wherein a user may hyperjump to a node represented as a node of the link;

selecting one of the displayed active links from those displayed with the displayed data; and

jumping to the position outside the object in the database.

2. The method of claim 1, wherein the active links are embedded icons and wherein the step of selecting comprises activating an embedded icon.

3. The method of claim 1, wherein the active links are embedded text and wherein the step of selecting comprises activating the embedded text.

4. The method of claim 1, wherein computer software is used, further comprising:

generating an active link, wherein the active link can be used to jump from a location in the database to another database.

5. A method for displaying information about a network that has hyperjump data, comprising:

choosing a node;

accessing the hyperjump data;

identifying hyperjump data from within the accessed hyperjump data that has a direct reference to the chosen node;

determining hyperjump data from within the accessed hyperjump data that has an indirect reference to the chosen node using the identified hyperjump data, wherein the step of determining comprises proximity analyzing the identified hyperjump data; and

displaying one or more determined hyperjump data.

6. The method of claim 5, wherein the hyperjump data includes pointers and wherein the direct reference is a pointer pointing to the chosen node or from the chosen node, and the step of determining comprises analyzing the pointers.

7. The method of claim 5, wherein the node represents a topic, the determined hyperjump data has a relationship to the topic, and the step of displaying displays determined hyperjump data that has a relationship to the topic.

8. The method of claim 5, wherein the node is a web page in the network, the accessed hypejump data are Universal Resource Locators of linked pages, and the step of determining hyperjump data comprises analyzing the identified hyperjump data.

9. The method of claim 5, wherein the node is a document in the network and the determined hyperjump data has a relationship to the document, the step of displaying comprising the step of listing the hyperjump data that has a relationship to the document.

10. The method of claim 5, wherein the step of displaying comprises generating a graphical user display, and wherein information is displayed on a graphical display visually representing more than one coordinate plane.

11. The method of claim 5, wherein the nodes are nodes in the network that may be accessed, the hyperjump data includes hyperjump links between nodes in the network, and the step of displaying comprises:

generating a source map using one or more of the determined hyperjump data, wherein the source map represents hyperjump links that identify the chosen node as a destination of a link; and

wherein the method further comprises activating a link represented on the source map, wherein a user may hyperjump to a node represented as a node of the link.

12. A method for visually displaying data related to a web having identifiable web pages and Universal Resource Locators with pointers, comprising:

choosing an identifiable web page;

identifying Universal Resource Locators for the web pages, wherein the identified Universal Resource Locators either point to or point away from the chosen web page;

analyzing Universal Resource Locators, including the identified Universal Resource Locators, wherein Universal Resource Locators which have an indirect relationship to the chosen web page are located, wherein the step of analyzing further comprises cluster analyzing the Universal Resource Locators for indirect relationships; and

displaying identities of web pages, wherein the located Universal Resource Locators are used to identify web pages.

13. The method of claim 12, further comprising selecting a web page using the displayed identities of web pages.

14. The method of claim 12, further comprising hyperjumping to the selected web page.

15. The method of claim 12, wherein the step of displaying the identities of web pages comprises generating a graphical user display wherein information within the Universal Resource Locators is parsed and used to generate the graphical user display.

16. A method for navigating documents on the world wide web, comprising: choosing a document;

identifying documents that have a direct relationship to the chosen document;

locating documents that have an indirect relationship to the chosen document identifying Universal Resource Locators for the documents, wherein the identified Universal Resource Locators either point to or point away from the chosen document;

analyzing Universal Resource Locators, including the identified Universal Resource Locators, wherein Universal Resource Locators which have an indirect relationship to the chosen document are located, wherein the step of analyzing further comprises cluster analyzing the Universal Resource Locators for indirect relationships; and

displaying a located document.

17. The method of claim 16, wherein pages and their respective Universal Resource Locators are used and the step of locating documents comprises analyzing the pages and their respective Universal Resource Locators.

18. The method of claim 17, wherein the step of analyzing pages comprises cluster analyzing the pages.

19. The method of claim 16, wherein the step of displaying a located document comprises:

generating a screen display of identities of one or more located documents; and

selecting one or more of the located documents.

20. The method of claim 19, wherein the step of generating a screen display comprises generating a graphical display.

21. A method for displaying information about a network that has hyperjump data, comprising:

choosing a node;

accessing the hyperjump data;

identifying hyperjump data from within the accessed hyperjump data that has a direct reference to the chosen node;

determining hyperjump data from within the accessed hyperjump data that has an indirect reference to the chosen node using the identified hyperjump data, wherein the step of determining comprises cluster analyzing the hyperjump data; and

displaying one or more determined hyperjump data.

22. A method for displaying information about a network that has hyperjump data, comprising:

choosing a node;

accessing the hyperjump data;

identifying hyperjump data from within the accessed hyperjump data that has a direct reference to the chosen node;

determining hyperjump data from within the accessed hyperjump data that has an indirect reference to the chosen node using the identified hyperjump data; and

displaying one or more determined hyperjump data, wherein the nodes are nodes in the network that may be accessed, the hypejump data includes hyperjump links between nodes in the network, and the step of displaying comprises:

generating a source map using one or more of the determined hyperjump data, wherein the source map represents hyperjump links that identify the chosen node as a destination of a link, and wherein the method further comprises activating a link represented on the source map, wherein a user may hyperjump to a node represented as a node of the link.

Description

TECHNICAL FIELD

This invention pertains to computerized research tools. More particularly, it relates to computerized research on databases. Specifically, the invention indexes data, searches data, and graphically displays search results with a user interface.

BACKGROUND

Two manuals containing background materials are hereby incorporated by reference "V-Search Integration Tool Kit For Folio VIEWS", containing thirty-six (36) pages, "V-Search Publisher's Tool Kit User's Manual", containing one hundred sixty (160) pages.

Our society is in the information age. Computers maintaining databases of information have become an everyday part of our lives. The ability to efficiently perform computer research has become increasingly more important. Recent efforts in the art of computer research have been aimed at reducing the time required to accomplish research. Computer research on non-textual objects is very limited. Current computer search programs use a text-by-text analysis procedure (Boolean Search) to scan a database and retrieve items from a database. The user must input a string of text, and the computer evaluates this string of text. Then the computer retrieves items from the database that match the string of text. The two popular systems for computerized searching of data used in the legal profession are Westlaw.TM., a service sold by West Publishing Company, 50 W. Kellogg Blvd., P.O. Box 64526, St. Paul, Minn. 55164-0526, and Lexis.TM., a service sold by Mead Data Central, P.O. Box 933, Dayton, Ohio 45401.

However, Boolean searches of textual material are not very efficient. Boolean searches only retrieve exactly what the computer interprets the attorney to have requested. If the attorney does not phrase his or her request in the exact manner in which the database represents the textual object, the Boolean search will not retrieve the desired textual object. Therefore, the researcher may effectively by denied access to significant textual objects that may be crucial to the project on which the researcher is working. A second problem encountered with Boolean searches is that the search retrieves a significant amount of irrelevant textual objects. (It should be noted that in the context of research, a textual object could be any type of written material. The term textual object is used to stress the fact that the present invention applies to all types of databases. The only requirement that a textual object must satisfy in order to be selected by a Boolean search program is that part of the textual object match the particular request of the researcher. Since the researcher cannot possibly know all of the groupings of text within all the textual objects in the database, the researcher is unable to phrase his request to only retrieve the textual objects that are relevant.

Aside from the inefficiency of Boolean searches, the present systems for computerized searching of data are inadequate to serve the needs of a researcher for several other reasons. Even if one assumes that all the textual objects retrieved from a Boolean search are relevant, the listing of the textual objects as done by any currently available systems does not convey some important and necessary information to the researcher. The researcher does not know which textual objects are the most significant (i.e., which textual object is referred to the most by another textual object) or which textual objects are considered essential precedent (i.e., which textual objects describe an important doctrine).

In the legal research field, both Westlaw.TM. and Lexis.TM. have a Shepardizing.TM. feature that enables the researcher to view a list of textual objects that mention a particular textual object. The Shepardizing feature does not indicate how many times a listed textual object mentions the particular textual object. Although the Shepardizing feature uses letter codes to indicate the importance of a listed textual object (e.g., an "f" beside a listed textual object indicates that the legal rule contained in particular textual object was followed in the listed textual object), data on whether a listed textual object followed the rule of a particular textual object is entered manually by employees of Shepard's.TM./McGraw Hill, Inc., Div. of McGraw-Hill Book Co., 420 N. Cascade Ave., Colorado Springs, Colo. 80901, toll free 1-800-525-2474. Such a process is subjective and is prone to error.

Another legal research system that is available is the Westlaw.TM. key number system. The Westlaw.TM. key number system has problems similar to the shepardizing feature on the Lexis.TM. and Westlaw.TM. systems.

The video displays of both the West.TM. and Lexis.TM. systems are difficult to use. The simple text displays of these systems do not provide a researcher with all the information that is available in the database.

Computerized research tools for legal opinions and related documents are probably the most sophisticated computer research tools available and therefore form the background for this invention. However, the same or similar computer research tools are used in many other areas. For example, computer research tools are used for locating prior art for a patent application. The same problems of inefficiency discussed above exist for computer research tools in many areas of our society.

What is needed is a system for computerized searching of data that is faster than the available systems of research.

What is needed is a system for computerized searching of data that enables researchers to research in a manner in which they are familiar.

What is needed is a computerized research tool that will reorganize, re-index or reformat the data into a more efficient format for searching.

What is needed are more sophisticated methods to search data.

What is needed is a system for computerized searching of data that will significantly reduce the number of irrelevant textual objects it retrieves.

What is needed is a user friendly computerized research tool.

What is needed is a visual user interface which can convey information to a user conveniently.

What is needed is a system for computerized searching of data that easily enables the researcher to classify the object according to his or her own judgment.

What is needed is a system for computerized searching of data that provides a visual representation of "lead" objects and "lines" of objects, permitting a broad overview of the shape of the relevant "landscape."

What is needed is a system for computerized searching of data that provides an easily-grasped picture or map of vast amounts of discrete information, permitting researchers to "zero in" on the most relevant material.

What is needed is a system for computer searching of data that provides a high degree of virtual orientation and tracking, the vital sense of where one has been and where one is going, and that prevent researchers from becoming confused while assimilating a large amount of research materials.

Accordingly, there is an unanswered need for a user friendly computerized research tool. There is a need for "intelligent" research technology that emulates human methods of research. There is a need in the marketplace for a more efficient and intelligent computerized research tool.

The present invention is designed to address these needs.

SUMMARY OF THE INVENTION

This invention is a system for computerized searching of data. Specifically, the present invention significantly aids a researcher in performing computerized research on a database or a network. The invention simplifies the research task by improving upon methods of searching for data including textual objects, and by implementing a user interface that significantly enhances the presentation of the data.

The invention can be used with an existing database by indexing the data and creating a numerical representation of the data. This indexing technique called proximity indexing generates a quick-reference of the relations, patterns, and similarity found among the data in the database. Using this proximity index, an efficient search for pools of data having a particular relation, pattern or characteristic can be effectuated. This relationship can then be graphically displayed.

There are three main components to the invention: a data indexing applications program, a Computer Search Program for Data Represented by Matrices ("CSPDM"), and a user interface. Each component may be used individually. Various indexing application programs, CSPDMs, and user interface programs can be used in combination to achieve the desired results. The data indexing program indexes data into a more useful format. The CSPDM provides efficient computer search methods. The preferred CSPDM includes multiple search subroutines. The user interface provides a user friendly method of interacting with the indexing and CSPDM programs. The preferred user interface program allows for easy entry of commands and visual display of data via a graphical user interface.

The method which the invention uses to index textual objects in a database is called Proximity Indexing. This method can also be used to index objects located on a network. The application of this method to network domains is discussed in greater detail later in this specification. Proximity Indexing is a method of preparing data in a database for subsequent searching by advanced data searching programs. Proximity Indexing indexes the data by using statistical techniques and empirically developed algorithms. The resulting search by an advanced data searching program of the Proximity Indexed data is significantly more efficient and accurate than a simple Boolean search.

The Proximity Indexing Application Program indexes (or represents) the database in a more useful format to enable the Computer Search Program for Data Represented by Matrices (CSPDM) to efficiently search the database. The Proximity Indexing Application Program may include one or more of the following subroutines: an Extractor: a Patterner: and a Weaver. The Proximity Indexing Application Program indexes (or represents) data in a locally located database or remotely located database. The database can contain any type of data, including text, alphanumerics, or graphical information.

In one embodiment, the database is located remotely from the Computer Processor and contains some data in the form of textual objects. The Proximity Indexing Application Program indexes the textual objects by determining how each full textual object (e.g., whole judicial opinion, statute, etc.) relates to every other full textual object by using empirical data and statistical techniques. Once each full textual object is related to each other full textual object, the Proximity Indexing Application Program compares each paragraph of each full textual object with every other full textual object as described above. The Proximity Indexing Application Program then clusters related contiguous paragraphs into sections. Subsequently, the Proximity Indexing Application Program indexes each section and the CSPDM evaluates the indexed sections to determine which sections to retrieve from the database. Such organization and classification of all of the textual objects in the database before any given search commences significantly limits the number of irrelevant textual objects that the CSPDM program retrieves during the subsequent search and allows retrieval of material based on its degree of relevancy.

In a preferred embodiment, the Proximity Indexing Application Program includes a link generation subroutine wherein direct and indirect relationships between or among data is used to generate a representation of the data. Generally, direct and indirect relationships in the database are identified as links and placed in a table.

Again, this method of computerized research can be used for nearly any database including those containing non-textual material, graphical material, newspapers material, data on personal identification, data concerning police records, etc.

The remaining two programs in the present invention are the CSPDM and the GUI Program. The CSPDM has seven subroutines that each search for different pools of objects. The GUI Program also has seven subroutines. Each CSPDM subroutine performs a different type of search. Each of the subroutines of the GUI uses the results of the corresponding subroutine of the CSPDM to create the proper display on the display.

After the Proximity Indexing Application Program indexes a database, the CSPDM application program is used to search the indexed database. For example, the CSPDM program can either be located in memory that is remote from the Computer Processor or local to the Computer Processor. In addition, the CSPDM program can either be remote or local in relation to the database.

The subroutines of the CSPDM utilize the coefficients and other data created by the Proximity Indexing Application Program to facilitate its search. However, if the researcher does not have the particular object citation available, the researcher can perform a Boolean search to retrieve and organize a pool of objects. Alternatively, the researcher can subsequently search for related objects by using the Pool-Similarity Subroutine, the Pool-Paradigm Subroutine, the Pool-Importance Subroutine or the Pool-Paradigm-Similarity Subroutine as defined below.

If the researcher already has the citation of a particular object available, the researcher can search for related objects by utilizing the Cases-In Subroutine, Cases-After Subroutine or Similar-Cases Subroutine. The Cases-In Subroutine retrieves all of the objects from the database to which a selected object refers. In addition, the subroutine determines the number of times the selected object refers to each retrieved object and other characteristics of each object, including its importance, and degree of relatedness to the selected object.

The Cases-After Subroutine retrieves all of the objects from the database that refer to the selected object. Also, the subroutine determines the number of times each retrieved object refers to the selected object and other characteristics of each object, including its importance and degree of relatedness to the particular object to which it refers.

The Similar-Cases Subroutine determines the degree of similarity between the retrieved objects and the selected object. Similarity may be defined, in the context of legal cases, as the extent to which the two objects lie in the same lines of precedent or discuss the same legal topic or concept. Numerous other relationships may be used to define similarity.

In addition, for a textual, object, if the researcher does not know of a particular textual object on which to base his or her search, the researcher may execute a Boolean word search. After a standard Boolean word search has been run, the researcher may run the Pool-Similarity Subroutine to retrieve information containing the degree of similarity between each textual object in the pool and a particular textual object selected by the user. Similarly, the Pool-Importance Subroutine can be used to determine the degree of importance (i.e., whether a judicial opinion is a Supreme Court opinion or a District Court opinion) and other characteristics of each textual object retrieved using the Boolean word search.

The Pool-Paradigm Subroutine calculates the geographic center in vector space of the pool of textual objects retrieved by the Boolean word search or other pool generating method. It then orders the retrieved textual objects by their degree of similarity to that center or "paradigm." The researcher can then evaluate this "typical textual object" and utilize it to help him or her find other relevant textual objects. In addition, the researcher can scan through neighboring "typical textual objects" to evaluate legal subjects that are closely related to the subject of the researcher's search.

The Pool-Paradigm-Similarity Subroutine similarly creates a paradigm textual object from the retrieved textual objects. However, the subroutine calculates the similarity of all textual objects in the database to the paradigm textual object in addition to the similarity of the retrieved textual objects to the paradigm textual object.

After the CSPDM has retrieved the desired objects, the Graphical User Interface (GUI) Program may be used to display the results of the search on the display. In one embodiment, the GUI is a user interface program. The GUI Program contains three main subroutines: Cases-In Display Subroutine (CIDS), Cases-After Display Subroutine (CADS) and Similar-Cases Display Subroutine (SCDS). The main subroutines receive information from the corresponding subroutines Cases-In, Cases-After and Similar-Cases of the CSPDM. The GUI Program also contains four secondary subroutines: Pool-Similarity Display Subroutine ("PSDS"), Pool-Paradigm Display Subroutine ("PPDS"), Pool-Importance Display Subroutine ("PIDS"), and the Pool-Paradigm-Similarity Subroutine (PPSDS). The secondary subroutines also receive information from the corresponding subroutines Pool-Similarity Subroutine, Pool-Paradigm Subroutine, Pool-Importance Subroutine and the Pool-Paradigm Similarity Subroutine of the CSPDM.

The CIDS subroutine receives information gathered from the Cases-In Subroutine of the CSPDM. The CIDS subroutine displays user friendly active boxes and windows on the display which represent the textual objects retrieved from the database represented in Euclidean space. It can also use the boxes to represent objects retrieved from a network. Various active box formats and arranging of information within the boxes may be utilized. The display depicts the appropriate location of textual objects in Euclidean space on a coordinate means. An algorithm may be used to determine the appropriate location of the boxes. The coordinate means may have one or more axis. In one embodiment, the horizontal axis of the coordinate means may represent the time of textual object creation; the vertical axis could represent a weighted combination of the number of sections in which that particular retrieved text is cited or discussed, its degree of importance, and its degree of similarity to the host textual object and the depth axis (Z-axis) represents the existence of data and length of the textual data or object.

The invention can also alter the background color of the window itself to communicate additional information graphically to the user. For example, if the horizontal axis represented time, then the invention could display the portion of the window containing objects occurring previous to the search object in one color and the portion containing the objects occurring after in another. Thus, the researcher can understand at a glance the relative position of his search target in relation to all the other objects related to it.

CIDS also enables the researcher to open up various active boxes on the display by entering a command into the computer processor with the input means. After entering the proper command, the active box transforms into a window displaying additional information about the selected textual object. These windows can be moved about the display and stacked on top or placed beside each other via the input means to facilitate viewing of multiple windows of information simultaneously. In one embodiment, the windows are automatically arranged by the computer system. Since the number of textual objects retrieved in a single search may exceed the amount which could be displayed simultaneously, the GUI Program enables the researcher to "zoom in" or "zoom out" to different scales of measurement on both the horizontal and vertical axis.

The CADS receives information gathered by the Cases-After Subroutine of the CSPDM. The CADS creates a display similar to the CIDS display. However, the active boxes representing the retrieved textual objects indicate which textual objects in the database refer to a selected textual object as opposed to which textual objects a selected textual object refers.

The SCDS receives information gathered by the Similar-Cases Subroutine of the CSPDM. The SCDS causes a similar display on the display as the CIDS and the CADS except that the vertical axis indicates the degree of similarity between the retrieved textual objects and the selected textual object.

The GUI Program contains four secondary subroutines: Pool-Search Display Subroutine (PSDS), Pool-Paradigm Display Subroutine (PPDS), Pool-Importance Display Subroutine (PIDS) and the Pool-Paradigm-Similarity Display Subroutine (PPSDS). The PSDS receives the results gathered by the Pool-Search Subroutine of the CSPDM. The PPDS receives the results gathered by the Pool-Paradigm Subroutine of the CSPDM. The PIDS receives the results gathered by the Pool-Importance Subroutine of the CSPDM. The PPSDS receives the results gathered by the Pool-Paradigm-Similarity Subroutine of the CSPDM. The results of the PSDS, PPDS, PIDS and PPSDS are then displayed in a user friendly graphical manner similar to the results of the CIDS, CADS and SCDS. A researcher can access the PSDS, PIDS, PSDS or PPSDS from any of the three main or four secondary subroutines of the GUI to gather information corresponding to the active boxes that represent the pool of textual objects retrieved by the corresponding subroutine of the CSPDM.

By using the graphical display, the researcher can view immediately a visual representation of trends in the data (for example, trends developing in the law and current and past legal doctrines). In addition, the researcher can immediately identify important data or important precedent and which object serving as the precedent is most important to the project on which the researcher is working. This visual representation is a vast improvement over the current computerized research tools. Furthermore, the researcher using the present invention does not have to rely on the interpretation of another person to categorize different textual objects because the researcher can immediately visualize the legal trends and categories of law. In addition, new topic areas can be recognized without direct human intervention. The current research programs require a researcher to view objects in a database or to read through the actual text of a number of objects in order to determine which objects are important, interrelated, or most closely related to the topic at hand and which ones are not.

It is an object of this invention to create an efficient and intelligent system for computerized searching of data that is faster than available systems of research.

It is an object of the invention to integrate the system of computerized searching into the techniques to which researchers are already accustomed.

It is an object of the invention to utilize statistical techniques along with empirically generated algorithms to reorganize, re-index and reformat data in a database into a more efficient model for searching.

It is an object of the invention to utilize statistical techniques along with empirically generated methods to increase the efficiency of a computerized research tool.

It is an object of the invention to create a system of computerized searching of data that significantly reduces the number of irrelevant objects retrieved.

It is an object of this invention to create a user friendly interface for computer search tools which can convey a significant amount of information quickly.

It is an object of the invention to enable the researcher to easily and immediately classify retrieved database objects according to the researcher's own judgment.

It is an object of the invention to provide a visual representation of "lead" objects and "lines" of objects, permitting a broad overview of the shape of the relevant "landscape."

It is an object of the invention to provide an easily-grasped picture or map of vast amounts of discrete information, permitting researchers to "zero in" on the most relevant material.

It is an object of the invention to provide a high degree of virtual orientation and tracking that enables a researcher to keep track of exactly what information the researcher has already researched and what information the researcher needs to research.

These and other objects and advantages of the invention will become obvious to those skilled in the art upon review of the description of a preferred embodiment, and the appended drawings and claims.

DESCRIPTION OF THE DRAWINGS

FIG. 1 is a high level diagram of the hardware for the system for computerized searching of data.

FIG. 2 is high level diagram of the software for the system for computerized searching of data. The three main programs are the Proximity Indexing Application Program, the Computer Search Program for Data Represented by Matrices (CSPDM) Application Program and the Graphical User Interface (GUI) Program.

FIG. 3A is a flow chart illustrating a possible sequence of procedures that are executed during the Proximity Indexing Application Program.

FIG. 3B is a flow chart illustrating a possible sequence of the specific subroutines that are executed during one stage of the Proximity Indexing Application Program. The subroutines are the Initial Extractor Subroutine, Opinion Patterner Subroutine, the Opinion Weaver Subroutine, the Paragraph Patterner Subroutine (Optional), the Paragraph Weaver Subroutine and the Section Comparison Subroutine.

FIG. 3C is flow chart illustrating a possible sequence of subroutines that are executed after the Section Comparison Subroutine. The Section Comparison Subroutine may comprise the Sectioner-Geographic Subroutine and the Section-Topical Subroutine (Optional). The sequence of subroutines executed after the Section Comparison Subroutine are the Section Extractor Subroutine, the Section Patterner Subroutine and the Section Weaver Subroutine.

FIG. 3D is a high level flow chart illustrating a possible sequence of subroutines that comprise the Boolean Indexing Subroutine which are executed during another stage of the Proximity Indexing Application Program. The first two subroutines, Initialize Core English Words and Create p.times.w Boolean Matrix, are executed by the Initial Extractor Subroutine. The results are then run through the Pool-Patterner Subroutine, the Pool-Weaver Subroutine, the Pool-Sectioner Subroutine, the Section-Extractor Subroutine, the Section-Patterner Subroutine and the Section Weaver Subroutine.

FIG. 3E is a chart illustrating the database format. The figure shows the types of structures contained within the database, links, link types, link subtypes, nodes, node types, node subtypes, and visual styles and also shows the various types of information that can be assigned to the links and nodes, including weights, identifications, names, comments, icons, and attributes.

FIG. 3F is a high level diagram showing a sequence of nodes, N.sub.o -N.sub.3, connected by direct links which have weights W.sub.1 -W.sub.3.

FIG. 3G is a high level diagram showing a sequence of nodes, N.sub.1 -N.sub.3, connected by direct and indirect links. The set of cluster links are also shown in the figure as functions of the weights associated with the direct links and the weight of the previous cluster link.

FIG. 3H is a flow chart which depicts the Cluster Link Generation Algorithm.

FIG. 4A is a high level diagram illustrating the flow of various search routines depending on the type of search initiated by the user by inputing commands to the Computer Processor via the input means. The diagram further illustrates the interaction between the CSPDM and the GUI Program.

FIG. 4B is a high level flow chart illustrating the sequence of subroutines in the CSPDM program and user interactions with the subroutines.

FIG. 4C is a high level flow chart for the Cases-In Subroutine.

FIG. 4D is a high level flow chart for the Cases-After Subroutine.

FIG. 4E is a high level flow chart for the Similar-Cases Subroutine.

FIG. 4F is a high level flow chart for the Pool-Similarity Subroutine.

FIG. 4G is a high level flow chart for the Pool-Paradigm Subroutine.

FIG. 4H is a high level flow chart for the Pool-Importance Subroutine.

FIG. 4I is a high level flow chart showing two possible alternate Pool-Paradigm-Similarity Subroutines.

FIG. 5A is a high level diagram illustrating the interaction between respective subroutines of the CSPDM and of the GUI Program. The diagram further illustrates the interaction between the GUI Program and the display.

FIG. 5B is an example of the display once the Cases-After Display Subroutine (CADS) is executed.

FIG. 5C is an example of the display after a user selects an active box representing a textual object retrieved by the Cases-After Subroutine and chooses to open the "full text" window relating to the icon.

FIG. 5D is an example of the display once the Cases-In Display Subroutine (CIDS) is executed.

FIG. 5E is an example of the display once the Similar-Cases Display Subroutine (SCDS) is executed.

FIG. 5F is an example of the display after a user chooses to execute the Similar Cases Subroutine for a textual object retrieved by the Similar-Cases Subroutine represented in FIG. 5E.

FIG. 5G is an example of the display after a user chooses to execute the Similar Cases Subroutine for one of the cases retrieved by the Similar-Cases Subroutine represented in FIG. 5F.

FIG. 5H depicts an Executive Search Window.

FIG. 6 depicts a schematic representation of eighteen patterns.

FIG. 7 is a high level diagram of the Layout of Boxes Algorithm.

FIG. 8 is a diagram of a screen showing execution of a show usage command.

FIG. 9 is a diagram of the Internal Box Layout Algorithm.

FIG. 10A is a diagram of a screen showing an Influence Map, which is a screen used in one embodiment of this invention.

FIG. 10B is a diagram of a screen showing a Source Map, which is a screen used in one embodiment of this invention.

FIG. 10C is a diagram of a screen showing a Cluster Map, which is a screen used in one embodiment of the invention.

FIG. 11 depicts a Look-Up Table for Bitmaps.

FIG. 12 is a software flow chart for the auto arranging window feature.

FIG. 13A is a depiction of a display with vertically tiled windows.

FIG. 13B is a depiction of a display with horizontally tiled windows.

FIG. 14A is a high level diagram of a method for searching, indexing, and displaying data stored in a network.

FIG. 14B is a high level diagram of a method for searching, indexing, and displaying data stored in a network using the cluster generation algorithm.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT

Referring now to the drawings, the preferred embodiment of the present invention will be described.

FIG. 1 is an overview of the preferred embodiment of the hardware system 26 for computerized searching of data. The hardware system 26 comprises a Computer Processor 30, a database 54 for storing data, input means, display 38, and RAM 34.

The Computer Processor 30 can be a processor that is typically found in Macintosh computers, IBM computers, portable PCs, clones of such PC computers (e.g. Dell computers), any other type of PC, or a processor in a more advanced or more primitive computing device. Parallel processing techniques may also be utilized with this invention.

The database 54 is connected to the Computer Processor 30 and can be any device which will hold data. For example, the database 54 can consist of any type of magnetic or optical storing device for a computer. The database 54 can be located either remotely from the Computer Processor 30 or locally to the Computer Processor 30. The preferred embodiment shows a database 54 located remotely from the Computer Processor 30 that communicates with the personal computer 28 via modem or leased line. In this manner, the database 54 is capable of supporting multiple remote Computer Processors 30. The preferred connection 48 between the database 54 and the Computer Processor 30 is a network type connection over a leased line. It is obvious to one skilled in the art that the database 54 and the Computer Processor 30 may be electronically connected in a variety of ways. In the preferred embodiment the database 54 provides the large storage capacity necessary to maintain the many records of textual objects.

The input means is connected to the Computer Processor 30. The user enters input commands into the Computer Processor 30 through the input means. The input means could consist of a keyboard 46, a mouse 42, or both working in tandem. Alternatively, the input means could comprise any device used to transfer information or commands from the user to the Computer Processor 30.

The display 38 is connected to the Computer Processor 30 and operates to display information to the user. The display 38 could consist of a computer monitor, television, LCD, LED, or any other means to convey information to the user.

The Random Access Memory (RAM 34) is also connected to the Computer Processor 30. The software system 60 for computerized searching of data may reside in the RAM 34, which can be accessed by the Computer Processor 30 to retrieve information from the software routines. A Read Only Memory (ROM), Erasable Programmable Read Only Memory (EPROM), disk drives, or any other magnetic storage device could be used in place of the RAM 34. Furthermore, the RAM 34 may be located within the structure of the Computer Processor 30 or external to the structure.

The hardware system 26 for computerized searching of data shown in FIG. 1 supports any one, or any combination, of the software programs contained in the software system 60 for computerized searching of data. The software system 60 for the computerized searching of data comprises one or more of the following programs: the Proximity Indexing Application Program 62, the Computer Search Program for Data Represented by Matrices (CSPDM 66) and the Graphical User Interface (GUI 70) Program. The Proximity Indexing Application Program 62 could reside in RAM 34 or in separate memory 58 connected to the database 54. The Computer Processor 30 or a separate computer processor 50 attached to the database 54 could execute the Proximity Indexing Application Program 62. In the preferred embodiment the Proximity Indexing Application Program 62 resides in separate memory 58 that is accessible to the database 54, and a separate computer processor 50 attached to the database 54 executes the Proximity Indexing Application Program 62.

The CSPDM 66 could reside in the RAM 34 connected to the Computer Processor 30 or in the separate memory connected to the database 54. In the preferred embodiment, the CSPDM 66 is located in the RAM 34 connected to the Computer Processor 30. This is also the preferred embodiment for the application of this method to network searching. For network application, a separate database 54 storing information to be analyzed is remotely connected to the computer processor 30. The CSPDM 66 may use the display 38 to depict input screens for user entry of information.

The GUI Program 70 could likewise reside in the RAM 34 connected to the Computer Processor 30 or in separate memory 58 connected to the database 54. In the preferred embodiment, the GUI Program 70 is located in the RAM 34 connected to the Computer Processor 30. The GUl Program 70 also communicates with the display 38 to enhance the manner in which the display 38 depicts information.

FIG. 2 is an overview of the preferred embodiment of the software system 60 for computerized searching of data. The software system 60 for computerized searching of data comprises at least one or more of the following programs: the Proximity Indexing Application Program 62, the Computer Search Program for Data Represented by Matrices (CSPDM 66) and the Graphical User Interface (GUI 70) Program. Proximity Indexing is a method of identifying relevant data by using statistical techniques and empirically developed algorithms. (See Appendix #2) The Proximity Indexing Application Program 62 is an application program which represents or indexes the database 54 to a proper format to enable the Computer Search Program for Data Represented by Matrices (CSPDM 66) to properly search the database 54. The Proximity Indexing Application Program 62 can index data in a local database 54 or a remote database 54. The Proximity Indexing Application Program 62 is shown in more detail in FIGS. 3A to 3H.

After the Proximity Indexing Application Program 62 indexes the database 54, the CSPDM 66 application program can adequately search the database 54. The CSPDM 66 program searches the database 54 for objects according to instructions that the user enters into the Computer Processor 30 via the input means. The CSPDM 66 then retrieves the requested objects. The CSPDM 66 either relays the objects and other information to the GUI Program 70 in order for the GUI Program 70 to display this information on the display 38, or the CSPDM 66 sends display commands directly to the Computer Processor 30 for display of this information. However, in the preferred embodiment, the CSPDM 66 relays the objects and other commands to the GUI Program 70. The CSPDM 66 is described in more detail in FIGS. 4A to 4I.

After the CSPDM 66 has retrieved the objects, the Graphical User Interface (GUI 70) Program, which is a user interface program, causes the results of the search to be depicted on the display 38. The GUI Program 70 enhances the display of the results of the search conducted by the CSPDM 66. The GUI Program 70, its method and operation, can be applied to other computer systems besides a system for computerized searching of data. The GUI Program 70 is described in more detail in FIGS. 5A to 5H.

FIGS. 3A to 3D depict examples of the procedures and subroutines of a Proximity Indexing Application Program 62, and possible interactions among the subroutines. FIG. 3A depicts a sequence of procedures followed by the Proximity Indexing Application Program 62 to index textual objects for searching by the CSPDM 66. FIG. 3B depicts specific subroutines that the Proximity Indexing Application Program 62 executes to partition full textual objects into smaller sections. FIG. 3C depicts subroutines executed by the Section Comparison Routine of FIG. 3B and subsequent possible subroutines to format and index the sections. FIG. 3D depicts a sequence of subroutines of the Proximity Indexing Application Program 62 which first sections and then indexes these sections of "core english words" 140 contained in the database 54. "Core english words" 140 are words that are uncommon enough to somewhat distinguish one textual object from another. The word searches of the CSPDM 66 search these sections of core English words to determine which textual objects to retrieve.

FIGS. 3E-3H show a preferred embodiment for representing the data in a database 54 or documents in a network in accordance with the present invention. The application of this method for representing documents on a network is described in greater detail later in this specification.

FIG. 3E shows a method for representing the data using the present invention. Specifically, FIG. 3E shows a method in which links 2004 and nodes 2008 can be used along with link types 2012, link subtypes 2020, node types 2016 and node subtypes 2024 to represent the data.

A node 2008 is any entity that can be represented by a box on a display 38 such as a GUI 70. A node 2008 might be, for example, an object in a database 54, a portion of an object in a database 54, a document, a section of a document, a World Wide Web page, or an idea or concept, such as a topic name. A node 2008 need not represent any physical entity such as an actual document. It is preferred that a node 2008 have links 2004, specifically, it is preferred that a node 2008 have links to other nodes 2008 (for example source links (a source link is a link 2004), or influence links (an influence link is a link 2004)). A node 2008 can represent any idea or concept that has links to other ideas or concepts. For example, two nodes 2008 can exist such as a node 2008 called Modem Architecture (not shown) and a node 2008 called Classical Architecture (not shown) and the links would show that Classical Architecture is a source for Modern Architecture and that Modem Architecture is influenced by Classical Architecture. In this example, a source link 2004 and an influence link 2004 would exist between the two nodes 2008. (Many times, links 2004 represent inverse relationships such as source links 2004 and influence links 2004, and one type of link may be derived or generated from analysis of another link.)

More specifically, in the preferred embodiment, the software defines a node 2010 as something that has a unique node 2008 identification, a node type 2016, a node subtype 2024, and an associated date (or plot date 2011). Node types 2016 or subtypes 2024 may have names 2021 or identifications, title descriptors 2026 and external attributes 2018. A node 2008 may have a corresponding numerical representation assigned, a vector, a matrix, or a table. In the preferred embodiment a table format is used for the nodes.

Referring to FIGS. 3E, 3F, and 3G, a link 2004 is another name or identification for a relationship between two nodes 2008. The relationship may be semantical, non-semantical, stated, implied, direct 2032, indirect 2036, actual, statistical and/or theoretical. A link 2004 can be represented by a vector or an entry on a table and contain information for example, a from-node identification 2010 (ID), a to-node ID 2010, a link type 2012, and a weight 2034. A group of links 2004 may be represented by a series of vectors or entries in a table, a link table. Link subtypes 2020 may be used, named and assigned comments.

In addition, to better integrate the GUI 70 and the data representation, visual styles 2028 may be assigned for example to nodes 2008, links 2004, link types 2012, and link subtypes 2020 to assist in the visual displays 38.

In the preferred embodiment, three types of links 2004 are used: source links 2004, influence links 2004 and cluster links 2004. Source links 2004 generally link a first node 2008 to second node 2008 that represents information or documentation specifically cited or referred to by the first node 2008. Influence links 2004 are generally the inverse of a source link 2004. The relationships represented by these links 2004 may be explicit or implied.

Links 2004 and nodes 2008 may be manually entered by a user or automatically generated by a computer 30. It is preferred that duster links 2004 be generated automatically by a processor. A cluster link 2004 is a relationship between two nodes 2008, for example, two nodes 2008 both directly linked to the same intermediate nodes 2008, may be indirectly linked through many paths and therefore have a cluster link 2004 between them. The cluster links 2004 may be determined using the specific or general methods described later for finding relationships in a database 54. However, the preferred method is through using a Proximity Indexing Application Program 62.

"Proximity indexing" is a method of indexing that uses statistical techniques and empirically generated algorithms to organize and categorize data stored in databases or on a network. The Proximity Indexing Application Program 62 applies the Proximity indexing method to a database 54. One embodiment of the present invention uses the Proximity Indexing Application Program 62 to Proximity index textual objects used for legal research by indexing objects based on their degree of relatedness--in terms of precedent and topic--to one another.

Applying the method to legal research, the "Proximity indexing" system treats any discrete text as a "textual object." Textual objects may contain "citations," which are explicit references to other textual objects. Any legal textual object may have a number of different designations of labels. For example, 392 U.S. 1, 102 S.Ct 415, 58 U.S.L.W. 1103, etc. may all refer to the same textual object.

Cases are full textual objects that are not subsets of other textual objects. Subjects of a full textual object include words, phrases, paragraphs, or portions of other full textual objects that are referred to in a certain full textual object. (The system does not treat textual objects as subsets of themselves.)

Every case, or "full" textual object, is assigned a counting-number "name" --designated by a letter of the alphabet in this description--corresponding to its chronological order in the database 54. Obviously, textual objects may contain citations only to textual objects that precede them. In other words, for full textual objects, if "B cites A," (i.e. "A is an element of B" or "the set `B` contains the name `A`"), textual object A came before B, or symbolically, A<B. Every textual object B contains a quantity of citations to full textual objects, expressed as Q(B), greater than or equal to zero, such that Q(B)<B.

Textual objects other than full textual objects may be subsets of full textual objects and of each other. For example, a section, page, or paragraph of text taken from a longer text may be treated as a textual object. Phrases and words are treated as a special kind of textual object, where Q(w)=0. Sections, pages, and paragraphs are generally subsets of only one full textual object, and may be organized chronologically under the numerical "name" of that full textual object. For purposes of chronology, phrases and words are treated as textual objects that precede every full textual object, and can generally be treated as members of a set with name "0," or be assigned arbitrary negative numbers.

Any two textual objects may be related to each other through a myriad of "patterns." Empirical research demonstrates that eighteen patterns capture most of the useful relational information in a cross-referenced database 54. A list of these eighteen patterns, in order of importance, follows:

Given that:

a, b, c<A;

A<d, e, f<B; and

B<g, h, i.

Patterns Between A and B Include

1. B cites A.

2. A cites c, and B cites c.

3. g cites A, and g cites B.

4. B cites f, and f cites A.

5. B cites f, f cites e, and e cites A.

6. B cites f, f cites e, e cites d, and d cites A.

7. g cites A, h cites B. g cites a, and h cites a.

8. i cites B, i cites f [or g], and f [or g] cites A.

9. i cites g, i cites A, and g cites B.

10. i cites g [or d], i cites h, g [or d] cites A, and h cites B.

11. i cites a, i cites B, and A cites a.

12. i cites A, i cites e, B cites e.

13. g cites A, g cites a, A cites a, h cites B, and h cites a.

14. A cites a, B cites d, i cites a, and i cites d.

15. i cites B, i cites d, A cites a, and d cites a.

16. A cites b, B cites d [or c], and d [or c] cites b.

17. A cites b, B cites d, b cites a, and d cites a.

18. A cites a, B cites b, d [or c] cites a, and d [or c] cites b.

These 18 patterns are shown schematically in FIG. 6.

(For a discussion on probability theory and statistics, see Wilkinson, Leland; SYSTAT: The System for Statistics; Evanston, Ill.: SYSTAT Inc., 1989 incorporated herein by reference.) Some patterns occur only between two full textual objects, and others between any two textual objects; this distinction is explained below.

Semantical patterning is only run on patterns number one and number two, shown above.

For purposes of explaining how patterns are used to generate the Proximity Index, only the two simplest patterns are illustrated.

The simplest, Pattern #1, is "B cites A." See FIG. 6. In the notation developed, this can be diagramed: a b c A d e f B g h i, where the letters designate textual objects in chronological order, the most recent being on the right, arrows above the text designate citations to A or B, and -arrows below the text designate all other citations. The next simplest pattern between A and B, Pattern #2, is "A cites c, and B cites c," which can also be expressed as "there exists c, such that c is an element of (A intersect B)." See Appendix #1. This can be diagramed: a b c A d e f B g h i. For every textual object c from 0 to (A-1), the existence of Pattern #2 on A and B is signified by 1, its absence by 0. This function is represented as P#2AB(c)=1 or P#2AB(c)=0. The complete results of P#1AB and P#2AB can be represented by an (A).times.(1) citation vector designated X.

The functions of some Patterns require an (n).times.(1) matrix, a pattern vector. Therefore it is simplest to conceive of every Pattern function generating an (n).times.(1) vector for every ordered pair of full textual objects in the database 54, with "missing" arrays filled in by 0s. Pattern Vectors can be created for Pattern #1 through Pattern #4 by just using the relationships among textual object A and the other textual objects in the database 54 and among textual object B and the other textual objects in the database 54. Pattern Vectors for Patterns # 5 through # 18 can only be created if the relationship of every textual object to every other textual object is known. In other words, Pattern Vectors for Patterns # 1 through # 4, can be created from only the rows A and B to the Citation Matrix but Pattern Vectors for Patterns #5 through #18 can only be created from the whole Citation Matrix.

(total textual objects c)/(theoretical maximum textual objects c) [(x)(x).sup.T /TMax],

(total textual objects c)/(actual maximum textual objects c) [(x)(x).sup.T /AMax]

frequency of object c per year [f], and

the derivative of the frequency [f'].

In pattern # 2, given that A<B, the theoretical maximum ("TMax") number Q(A intersect B)=A minus 1. The actual maximum possible ("AMax"), given A and B, is the lesser of Q(A) and Q(B). The ratios "X(X).sup.T /TMax" and "X(X).sup.T /AMax," as well as the frequency of occurrence of textual objects c per year, f2(A, B), and the first derivative f'2(A, B), which gives the instantaneous rate of change in the frequency of "hits," are all defined as "numerical factors" generated from patterns #1 and #2. These are the raw numbers that are used in the weighing algorithm.

For Pattern #2, the total number of possible textual objects c subject to analysis, i.e., TMax, is A-1, one only for the years at issue which are those up to the year in which A occurred. However, a relationship may remain "open," that is, it may require recalculation of f(x) and f'(x) as each new textual object is added to the database 54, (for a total of n cases subject to analysis).

The "numerical factors" for all eighteen patterns are assigned various weights in a weighing algorithm used to generate a scalar F(A, B). The function F generates a scalar derived from a weighted combination of the factors from all eighteen patterns. The patterns are of course also weighted by "importance," allowing Supreme Court full textual objects to impose more influence on the final scalar than District Court full textual objects, for example. The weighing of the more than 100 factors is determined by empirical research to give results closest to what an expert human researcher would achieve. The weighing will vary depending upon the type of material that is being compared and the type of data in the database 54. (See Thurstone. The Vectors of Mind, Chicago, Ill.: University of Chicago Press, 1935, for a description of factor loading and manipulating empirical data incorporated herein by reference.) In a commercial "Proximity Indexer" it will be possible to reset the algorithm to suit various types of databases.

A scalar F(A, B) is generated for every ordered pair of full cases in the database 54, from F(1, 2) to F(n-1, n). F(z,z) is defined as equal to 0.

The full results of F(A,B) are arranged in an (n).times.(n) matrix designated F. Note that F(B, A) is defined as equal to F(A, B), and arrays that remain empty are designated by 0. For every possible pairing of cases (A,B), a Euclidean distance D(A,B) is calculated by subtracting the Bth row of Matrix F from the Ath row of Matrix F. In other words:

D(A,B)=[(F(1, A)-F(1, B)).sup.2 +(F(2, A)-F(2, B)).sup.2 + . . . +(F(n, A)-F(n, B)).sup.2 ].sup.1/2.

A function designated D(A,B) generates a scalar for every ordered pair (A,B), and hence for every ordered pair of textual objects (A,B) in the database 54. The calculations D(A,B) for every ordered pair from D(1,1) to D(n,n) are then arranged in an (n).times.(n) "proximity matrix" D. Every column vector in D represents the relationship between a given case A and every other case in the database 54. Comparing the column vectors from column A (representing textual object A) and column B (representing textual object B) allows one to identify their comparative positions in n-dimensional vector space, and generate a coefficient of similarity, S(A,B), from 0-100%, which is more precise and sophisticated than F(A,B) or D(A,B) alone. A similarity subroutine can run directly on F(A,B). However, the real power of the Proximity Matrix D is that it allows one to identify "groups" or "clusters" of interrelated cases.

Through factor loading algorithms, the relationships represented by D for "n" cases can be re-represented in a vector space containing fewer than "n" orthogonal vectors. This knowledge can be reflected in S(A,B).

The Proximity Indexing Application Program 62 is an application program that applies the above techniques and algorithms to index and format data to be searched by the CSPDM 66.

FIG. 3A describes the overall procedure of the Proximity Indexing Application Program 62. The first stage initializes the data 74 in the database 54. The second stage determines the relationships between full textual objects 78. The third stage determines the relationships between paragraphs of each textual object and each full textual object 80. The fourth stage clusters related paragraphs using factor loading and empirical data and then groups the paragraphs into sections based on such data 84. The fifth stage determines the relationships between the sections 88. In the final stage, the sectioned textual objects are not further processed until commands are received from the CSPDM Routine 92.

The following description of FIG. 3B and FIG. 3C elaborates on this general procedure by describing specific subroutines of a Proximity Indexing Application Program 62. The following is a step by step description of the operation of the Proximity Indexing Application Program 62.

Section A Initial Extractor Subroutine 96

FIG. 3B describes subroutines for the first portion of the preferred Proximity Indexing Application Program 62. The first subroutine of the Proximity Indexing Applications Program is the Initial Extractor Subroutine 96. The Initial Extractor Subroutine 96 performs three primary functions: Creation of the Opinion Citation Matrix, creation of the Paragraph Citation Matrix, and creation of Boolean Word Index.

The following steps are performed by the Initial extractor subroutine 96.

1. Number all full textual objects chronologically with arabic numbers from 1 through n.

2. Number all paragraphs in all the full textual objects using arabic numbers from 1 through p.

3. Identify the page number upon which each paragraph numbered in step two above begins.

4. Create Opinion Citation Vectors (X). By comparing each full textual object in the data base to every other full textual object in the data base that occurred earlier in time.

5. Combine Opinion Citation Vectors to create the bottom left half portion of the n.times.n Opinion citation matrix.

6. Create a mirror image of the bottom left half portion of the Opinion citation matrix in the top right half portion of the same matrix, to complete the matrix. In this manner only n.sup.2 /2 comparisons need to be conducted. The other 1/2 of the comparisons are eliminated.

7. Create the p.times.n Paragraph Citation Vectors by comparing each paragraph to each full textual object that occurred at an earlier time. This will require (n/2)p searches.

8. Create a Paragraph Citation Matrix by combining Paragraph Citation Vectors to create the bottom left half portion of the matrix.

9. Complete the creation of the Paragraph Citation Matrix by copying a mirror image of the bottom left half portion of the matrix into the top right half portion of the matrix.

10. Initialize the Initial Extractor Subroutine 96 with a defined set of core English words 140.

11. Assign identification numbers to the core English words 140. In the preferred embodiment 50,000 English words are used and they are assigned for identification the numbers from -50,000 to -1.

12. Create a Boolean Index Matrix 144 with respect to the core English words by searching the database 54 for the particular word and assigning the paragraph number of each location of the particular word to each particular word. This procedure is described in greater detail in FIG. 3D.

Section B Opinion Patterner Subroutine 100

The Opinion Patterner Subroutine 100 performs three primary functions: Pattern analysis 6n matrices, calculation of the numerical factors and weighing the numerical factors to reach resultant numbers.

13. Process the Opinion Citation Matrix through each of the pattern algorithms described above and in FIG. 6 for each ordered pair of full textual objects to create opinion pattern vectors for each pattern and for each pair of full textual objects. The pattern algorithms determine relationships which exist between the ordered pair of textual objects. The first four pattern algorithms can be run utilizing just the Opinion Citation Vector for the two subject full textual objects. Each pattern algorithm produces a opinion pattern vector as a result. The fifth through eighteenth pattern algorithms require the whole Opinion Citation Matrix to be run through the Opinion Patterner Subroutine 100.

14. Calculate total hits (citation) for each pattern algorithm. This can be done by taking the resultant opinion pattern vector (OPV) and multiplying it by the transposed opinion pattern vector (OPV).sup.T to obtain a scalar number representing the total hits.

15. Calculate the theoretical maximum number of hits. For example, in the second pattern, the theoretical maximum is all of the full textual objects that occur prior in time to case A (A-1).

16. Calculate the actual maximum number of hits. For example, in the second pattern, the actual maximum possible number of hits is the lesser of the number of citations in full textual object Q(A) or full textual object Q(B).

17. Calculate the total number of hits (citations) per year. This is labeled f(A,B).

18. Calculate the derivative of the total change in hits per year. This is the rate of change in total hits per year and is labeled f' (A,B).

19. Calculate the ratio of total hits divided by theoretical max [((OFV)(OVP).sup.t/TMAX).

20. Calculate the ratio of the total hits divided by the actual maximum [(OPV)(OPV).sup.t.sub./AMAX ].

21. Calculate a weighted number F(A,B) which represents the relationship between full textual object A and full textual object B. The weighted number is calculated using the four raw data numbers, two ratios and one derivative calculated above in steps 14 through 20 for each of the 18 patterns. The weighing algorithm uses empirical data or loading factors to calculate the resulting weighted number.

22. The Opinion Patterner Subroutine 100 sequence for the Opinion Citation Matrix is repeated n-1 times to compare each of the ordered pairs of full textual objects. Therefore, during the process, the program repeats steps 13 through 21, n-1 times.

23. Compile the Opinion Pattern Matrix by entering the appropriate resulting numbers from the weighing algorithm into the appropriate cell locations to form an n.times.n Opinion Pattern Matrix.

Section C The Opinion Weaver Subroutine 104

The Opinion Weaver Subroutine 104 shown in FIG. 3B, performs two primary tasks: calculation of the Opinion Proximity Matrix and calculation of the Opinion Similarity Matrix. The Opinion Proximity Matrix D is generated by calculating the Euclidean Distance between each row A and B of the Opinion Pattern Matrix (D(A,B)) for each cell DC(A,B). The Opinion Similarity Matrix is generated by calculating the similarity coefficient from 0 to 100 between each row A and B of the Opinion Proximity Matrix (S(A,B)) in each cell SC(A,B) in matrix S.

24. Calculate the n.times.n Opinion Proximity Matrix. To calculate D(A,B) the program takes the absolute Euclidian distance between column A and column B of the n.times.n Opinion Pattern Matrix. The formula for calculating such a distance is the square root of the sum of the squares of the distances between the columns in each dimension, or:

D(A,B)=[(F(1,A)-F(1,B)).sup.2 +(F(2,A)-F(2,B)).sup.2 + . . . +(F(N,A)F(N,B)).sup.2 ].sup.1/2

The Opinion Proximity Matrix created will be an n.times.n matrix. The smaller the numbers in the Opinion Proximity Matrix the closer the relationship between full textual object A and full textual object B.

25. Create n.times.n Opinion Similarity Matrix. To calculate the Opinion Similarity Matrix each scalar number in the Opinion Proximity Matrix is processed through a coefficient of similarity subroutine which assigns it a number between 0 and 100. By taking the coefficient of similarity, the program is able to eliminate full textual objects which have Euclidian distances that are great. (For example, a Euclidean distance that is very large and is run through the coefficient of similarity would result in a very low coefficient of similarity. Euclidean distances resulting in similarities below four are eliminated in the preferred embodiment).

Section D Paragraph Patterner Subroutine 108 (Optional)

26. Obtain the p.times.n Paragraph Citation Matrix calculated by the Initial Extractor Subroutine 96.

27. Run each ordered pair of rows of the p.times.n Paragraph Citation Matrix for an individual full textual object i through the pattern algorithms number one and two and determine the resultant Paragraph Pattern Vector.

28. Calculate the various numerical factors (AMax, TMax, etc.) by evaluating the values in the Paragraph Pattern Vector.

29. Run the Paragraph Pattern Vector and the numerical factors through the weighing algorithm to determine the appropriate value for each cell of the c.sub.i.times.n Partial Paragraph Pattern Matrix where c.sub.i is the number of paragraphs in full textual object i.

30. Repeat steps 27 through 29 for each full textual object i where i=1 to n, to create the p.times.n Paragraph Pattern Matrix.

Section E Paragraph Weaver Subroutine 112

31. Calculate the Euclidean distance of each ordered pair of rows of either the p.times.n Paragraph Citation Matrix or the p.times.n Paragraph Pattern Matrix for a single full textual object i.

32. Place the resultant Euclidean distance values in the appropriate cell of the c.sub.i.times.c.sub.i Paragraph Proximity Matrix where c.sub.i is the number of paragraphs in full textual object i, where 0<i<n+1.

33. Repeat steps 31 through 32 n times in order to calculate n different Paragraph Proximity Matrices (one for each full textual object i).

34. The Section Comparison Subroutine 116 clusters all p paragraphs in the database 54 into sections. Then the sections are compared and indexed in the database 54. This procedure is described in greater detail in FIG. 3C.

FIG. 3C depicts possible subroutines that the Section Comparison Subroutine 116 comprises. The subroutines are the Sectioner Geographical Subroutine 120, the Sectioner Topical Subroutine 124 (Optional), the Section Extractor Subroutine 128, the Section Patterner Subroutine 132 and the Section Weaver Subroutine 136. Section F Sectioner Geographical Subroutine 120

35. For each full textual object i, the Sectioner Geographical Subroutine 120 uses the corresponding c.sub.i.times.c.sub.i Paragraph Proximity Matrix and a contiguity factor for each paragraph to determine which paragraphs may be clustered into sections. Sections are made up of continuous paragraphs that are combined based upon weighing their Euclidean distances and contiguity.

36. Repeat step 35 for all n full textual objects until all p paragraphs are grouped into q sections.

Section H Sectioner Topical Subroutine 124 (Optional)

37. The Sectioner Topical Subroutine 124 provides additional assistance to the Sectioner Geographical Subroutine 120 by considering the factor of topical references to determine the q sections.

38. For the total number of discrete references "z" to each full textual object in a particular full textual object, a z.times.z Citation Proximity Matrix is formed by comparing the Euclidean distances between each reference to a full textual object contained in each paragraph and calculating the topical weight given to each paragraph.

Section I Section Extractor Subroutine 128

39. The Section Extractor Subroutine 128 numbers each section created by the Sectioner Geographical Subroutine 120 and Sectioner Topical Subroutine 124 Subroutines from 1 to q.

40. The Sectioner Extractor Subroutine 128 creates a q.times.q Section Citation Matrix by determining which sections refer to every other section.

Section J Section Patterner Subroutine 132 (shown in FIG. 3C)

41. The Section Patterner Subroutine 132 then calculates 18 Section Pattern Vectors corresponding to each row of the q.times.q Section Citation Matrix using the 18 pattern algorithms.

42. From the Section Pattern Vectors, the numerical factors (AMax, TMax, etc.) are calculated.

43. The weighing algorithm evaluates the numerical factors and the Section Pattern Vectors and determines the values for each cell of the q.times.q Section Pattern Matrix.

Section K Section Weaver Subroutine 136

44. The Section Weaver Subroutine 136 calculates the Euclidean distances between each row of the q.times.q Section Pattern Matrix and creates a q.times.q Section Proximity Matrix.

45. The Section Weaver Subroutine 136 then creates a q.times.q Section Similarity Matrix with coefficients 0 to 100 using the values of the Section Proximity Matrix and empirical data and factor loading.

Section L Semantical Clustering of a Boolean Index Routine 138

FIG. 3D depicts a possible Semantical Clustering of a Boolean Index Routine 138. (See Hartigan, J. A. Clustering Algorithms. New York: John Wiley & Sons, Inc., 1975, for detailed description of clustering algorithms incorporated herein by reference.) The Semantical Clustering routine of a Boolean Index 138 indexes the textual objects according to the similarity of phrases and words contained within each textual object in a database 54. The routine comprises seven possible subroutines: the Initial Extractor Subroutine 96, the Pool Patterner Subroutine 152, the Pool Weaver Subroutine 96 the Pool Sectioner Subroutine 160, the Section Extractor Subroutine 128, the Section Patterner Subroutine 132 and the Section Weaver Subroutine 136, In fact, it is quite possible, using only semantical statistical techniques, to "Proximity-index" documents that do not refer to one another at all based on there Boolean indices.

Section M Initial Extractor Subroutine 96

46. As described in steps 10 and 11, the Initial Extractor Subroutine 96 initializes a set of core English words 140 and assigns each word a number. The preferred embodiment uses 50,000 discrete core English words and assigns each discrete core English word a number from -50,000 to -1.

47. The Initial Extractor Subroutine 96 then converts the core English words into a p.times.w matrix. The number of columns (w) represents the number of discrete core English words in the database 54 and the number of rows (p) represents the number of paragraphs in the database 54.

48. The Initial Extractor Subroutine 96 fills the p.times.w matrix by inserting a "1" in the matrix cell where a certain paragraph contains a certain word.

Section N Pool Patterner Subroutine 152

49. The Pool Patterner Subroutine 152 creates two pattern algorithm vectors for only the first two patterns and determines values for the total number of hits, the theoretical maximum number of hits, the actual maximum number of hits, the total number of hits per year and the derivative of the total number of hits per year.

50. The weighing algorithm of the Pool Patterner Subroutine 152 uses empirical data and factor loading to determine values to enter into a p.times.w Paragraph/Word Pattern Matrix.

51. The Pool Weaver Subroutine 156 creates a p.times.w Paragraph/Word Pattern Matrix by filling the appropriate cell of the Matrix with the appropriate value calculated by the weighing algorithm.

52. The Pool Patterner Subroutine 152 creates a p.times.w Paragraph/Word Proximity Matrix taking the Euclidean distance between the rows of the Paragraph/Word Pattern Matrix.

Section O Pool Sectioner Subroutine 160

53. The Pool Sectioner Subroutine 160 evaluates the Euclidean distances in the Paragraph/Word Proximity Matrix and the contiguity factor of each paragraph to cluster the paragraphs (p) into a group of (v) sections and create a v.times.w Preliminary Cluster Word Matrix.

Section P Section Extractor Subroutine 128

54. The-Section Extractor Subroutine 128 numbers each section chronologically and creates a v.times.v Section Word Citation Matrix.

Section Q Section Patterner Subroutine 132

55. The Section Patterner Subroutine 132 evaluates the v.times.v Section Word Citation Matrix to create two word pattern vectors for only the first two patterns algorithms (described above and shown in FIG. 6) and determines numerical factors for the total number of hits, the theoretical maximum number of hits, the actual maximum number of hits, the total number of hits per year and the derivative of the total number of hits per year.

56. The Weighing algorithm uses empirical data and factor loading to weigh the numerical factors created from the word pattern vectors and uses the numerical factors and the word pattern vectors to determine values t o enter into a v.times.v Section Word Pattern Matrix.

Section R Section Weaver Subroutine 136

57. The Section Weaver Subroutine 136 creates a v.times.v Section Word Proximity Matrix by taking the Euclidean distance between the rows of the Section Word Pattern Matrix and placing the appropriate Euclidean distance value in the appropriate cell of the Section Word Proximity Matrix.

58. The Section Weaver Subroutine 136 create a v.times.v Section Word Similarity Matrix by evaluating the Euclidean distances from the Section Word Proximity Matrix and empirical data, and calculating the similarity coefficient for each order ed pair of sections, and places the value in the appropriate cell of the Section Word Similarity Matrix.

59. The Pool Searches of the CSPDM 66 evaluate the Section Word Similarity Matrix as well as other matrices to determine whether or not to retrieve a full textual object.

The following describes a preferred cluster link generator 2044 which implements a specific type of patterer or clustering system for use alone or in conjunction with other proximity indexing subroutines, and prior to searching. The cluster link generator 2044 analyzes a set of numerical representations of a database 54 and generates a second set of numerical representations of the database 54. This second set is stored in the RAM 34. This second set of numerical data can represent indirect 2036, direct 2032, or a combination of both direct 2032 and indirect 2036 relationships in the database 54. Preferably, the second set of numerical representations accounts for indirect 2036 relationships in the database 54. It is preferred that the first and second set of numerical data be in a table format and that-the first set represent direct 2032 relationships or links and the second set represent cluster links 2004.

Referring to FIG. 3H, the cluster link generation algorithm 2044 analyzes links to generate a set of cluster links 2004. More specifically, the cluster link generation algorithm 2044 generates a set of cluster links 2004 by analyzing direct 2032 and/or indirect relationships 2036 between nodes 2008 or between objects in a database 54 and generates a set of cluster links 2004.

In the preferred embodiment, the cluster link generator 2044 analyzes direct links 2004 (for example source links 2004 and influence links 2004). These direct links 2032 may be represented by a table or series of vectors. The cluster link generator 2044 then locates indirect relationships 2036 between nodes 2008 or objects in a database 54. The indirect relationships 2036 are preferably made up of direct links 2032. The indirect relationship 2036 paths are preferably made up of direct links 2004. The cluster link generator 2040 then generates a set of cluster links 2004 based upon both the direct links 2032 and on the indirect relationships. The set of cluster links 2004 may be represented by a table or a series of vectors. Another embodiment of this invention uses candidate cluster links 2004 to provide a more efficient search. Candidate cluster links are the set of all possible cluster links 2004 between a search node 2008 and a target node 2004. In this embodiment, only a subset of the candidate cluster links 2004, the actual cluster links 2004, which meet a certain criteria are used to locate nodes 2008 for display.

Consider a set of nodes 2008 N.sub.0 . . . N.sub.3 connected by a sequence of direct links 2032 whose weights 2034 are given by W.sub.1 . . . W.sub.3, as shown in FIGS. 3F.

Node 2008 N.sub.1 is reachable from N.sub.0 through a path P.sub.1 of length 1 (that is, N.sub.0.fwdarw.N.sub.1); node 2008 N.sub.2 is reachable through a path P.sub.2 of length 2 (N.sub.0.fwdarw.N.sub.1.fwdarw.N.sub.2); and so on.

Each path P provides some evidence that the start node 2008 (N.sub.0) and destination node 2008 (N.sub.1, N.sub.2, or N.sub.3) are related to some extent. The strength of the indirect relationship 2036 depends on a length L of the path P and on the weights 2034 of the individual direct links 2032 along that path P.

In FIG. 3G, the indirect relationship 2036 from N.sub.0 to N.sub.1, N.sub.2, and N.sub.3 are shown as arcs.

The weight C.sub.1 . . . C.sub.3 of each implied relationship, is a function of the weight 2034 from the path to the previous node 2008 and the weight 2034 of the last direct link 2032.

The individual functions F1 . . . F3 describe how to combine the weights 2034 of the direct links 2004 to determine the weight c of an indirect link 2036. Selecting appropriate functions is the key to making cluster link generation work well. A preferred definition of F.sub.N is as follows:

C.sub.N =F.sub.N (C.sub.N-1, W.sub.N)=min(C.sub.N-1, D.sub.N * W.sub.N),

where D.sub.N is a damping factor that decreases rapidly as N increases.

The cluster link algorithm 2044 determines the set of all paths P from a given start node 2008 N.sub.0 that have a length less than or equal to a given length L. Each path is rated using the method described above. The paths are then grouped by destination node 2008; the candidate cluster link 2004 C(N.sub.0, N.sub.N) between N.sub.0 and a given destination node 2008 N.sub.N has a weight C.sub.N equal to the sum of the weights 2034 of all paths P.sub.N leading to N.sub.N.

The set of all candidate cluster links 2004 is then sorted by weight 2034. A subset of the candidate links 2004 is chosen as actual cluster links 2004. The number of cluster links 2032 chosen may vary, depending on the number of direct links 2004 from N.sub.0, and on the total number of candidate cluster links 2004 available to choose from.

Performance considerations and efficiency are more important with large databases than for small databases. For large databases, finding the set of all paths P from a given node 2008 N.sub.0 that have a length less than or equal to a given length L may be impractical, since the number of unique paths may number in the tens of millions.

One embodiment of this invention uses candidate cluster links 2004 to provide a more efficient search. Candidate cluster links 2004 are the set of all possible cluster links 2004 between a start node (2008) and a destination node (2008).

Clearly, it is not necessary to examine millions of paths when the goal is to select the top or strongest duster links 2004 for each start node 2008 N.sub.0 (for example, the top 20 to 25 cluster links 2004). The great majority of paths have an insignificant effect on the final results. What is needed is an implementation of the cluster link algorithm 2044 where the total number of paths examined is bounded, independent of the size of the database 54, without a loss in effectiveness. To this end, we have an implementation of the algorithm 2044 such that a cluster link 2004 is defined recursively.

We define C.sub.L (N.sub.0, N.sub.N), the order-L cluster link 2004 from start node 2008 to destination node 2008, as the cluster link 2004 between N.sub.0 and N.sub.N, considering only paths of length less than or equal to L. Then, we can derive C.sub.L+1 (N.sub.0, N.sub.N) from C.sub.L (N.sub.0, N.sub.N) and C.sub.1 (N.sub.0, N.sub.N).

The assumption is that most of the paths P.sub.L (N.sub.0, N.sub.N) of length L (or greater) from N.sub.0 to N.sub.N will not have a significant impact on cluster link generation. Therefore, we an use a set of candidate duster links 2004 C.sub.L (N.sub.0, N.sub.N) as a summary of that path information for the purpose of determining C.sub.L+1 (N.sub.0, N.sub.N). This assumption has a significant impact on the performance of the algorithm 2044 in this implementation, since the search space is significantly reduced at each step. The computer processing "cost" of generating cluster links 2004 is bounded by the size of the candidate cluster link 2004 sets generated at the intermediate steps, rather than by the total number of relevant paths in the database 54.

The size of the candidate duster link 2004 set generated at each intermediate step affects the speed of the algorithm 2044 in this implementation. If too many candidate cluster links 2004 are generated at each intermediate step, the algorithm 2044 is too slow. On the other hand, if too few candidate cluster links 2004 are generated, and too many paths are pruned, then C.sub.L (N.sub.0, N.sub.N) is no longer an accurate summary of P.sub.L (N.sub.0, N.sub.N).

Finally, since the weights 2034 of the individual candidate cluster links 2004 in C.sub.L (N.sub.0, N.sub.N) are generally much greater than the weights 2034 of the individual paths in P.sub.L (N.sub.0, N.sub.N), the damping factors D.sub.N used to derive the combined weights 2034 at each step must be decreased accordingly in this implementation.

The specifics for the basic generator algorithm 2044 of this implementation, for determining the set of order N cluster links 2004 from a given start node 2008 N.sub.0, are shown in FIG. 3H. The generator algorithm 2044 works for any value of N greater than zero. If N=1, the set of candidate cluster links 2004 generated is simple. The processing cost of determining the candidate cluster links 2004 increases with N. In practice, N=3 appears to yield the best results.

The generator algorithm 2044 starts by initializing the candidate cluster link 2004 set 2048 and creating a loop for i=0 to N 2052. The generator algorithm 2044 then performs a series of steps for each path P 2056. First, it selects the destination node 2008 as the node to analyze and retrieves the set of direct links 2032 (L) from the selected node 2008 to any other node 2008 in the database 54, N.sub.i+1. Second, for each direct link 2032 L, the generator algorithm 2044 performs a series of steps.

The generator algorithm 2044 creates a new path P' of length i+1 consisting of the path P plus the direct link 2064 L from the selected node 2008 to the node 2008 N.sub.i+1 2056. The algorithm 2044 then determines the combined weight 2034 WC.sub.i+1 from WC.sub.i, the weight 2034 of the path P, and W.sub.i+1, the weight 2034 of Link 2004 L 2064, using the following preferred formula:

WC.sub.i+1 =min(WC.sub.i, D.sub.i+1 *W.sub.i+1).

Following these computations, the generator algorithm 2044 decides whether there already is a paths P in the cluster link 2004 from N.sub.0 to N.sub.i+1 2068. If there is a not already a path, the algorithm 2044 adds P' to C.sub.i+1 2072. If there already is a path, the algorithm 2044 adds WC.sub.i+1 to the weight 2034 of the existing path in C.sub.i+1 2076. These steps are then repeated as necessary.

Once the candidate cluster link 2004 set has been generated, deriving the actual cluster links 2004 is a simple matter of selecting or choosing the T top rated candidate links 2004, and eliminating the rest. In practice, the following formula has yielded good results:

T=min(constant, 4*d),

where d is the number of direct links 2004 from N.sub.0. Setting the constant equal to twenty has yielded good results. More than T cluster links 2004 may be generated if there are ties in the ratings. After each iteration, the candidate cluster link 2004 set C.sub.i may be pruned so that it contains only the top candidate cluster links 2004 (for example, the top 200).

FIGS. 4A and 4B are high level flow charts that illustrate the general flow of the subroutines of the CSPDM 66. FIG. 4A illustrates that the flow of various search routines depend on the type of search initiated by the researcher. The diagram further illustrates the interaction between the CSPDM 66 and the GUI Program 70. FIG. 4B illustrates the sequence of subroutines in the CSPDM 66 program and the user interactions with the subroutines. FIG. 4B further shows that the researcher can access the different search subroutines and use information that the researcher has already received to find new information.

FIG. 4B provides a high level flow chart illustrating the sequence of subroutines in the CSPDM 66 program and the researcher's interactions with the subroutines. Assuming that the database 54 the researcher desires to access has been proximity indexed, the researcher must log on 260 to the database 54. By entering the appropriate information into the Computer Processor 30 via the input means, the researcher electronically accesses 264 the database 54 and enables the CSPDM 66 to search 200 the database 54.

FIGS. 4A and 4B both show the preliminary options that the researcher can choose from before selecting one of the searching subroutines of the CSPDM 66. The CSPDM 66 questions the researcher on whether the researcher has identified a pool of textual objects 204. If the researcher has selected a pool of textual objects 204, then the researcher is able to choose one of the pool search 208 subroutines 212. If the researcher has not selected a pool of textual objects, the CSPDM 66 questions the researcher on whether the researcher has selected a single textual object 216. If the researcher has selected a single textual object 216, then the researcher is able to choose one 220 of the textual object searches 224. If the researcher has not selected either a pool of textual objects 204 or a single textual object 216, then the researcher must execute a Boolean Word Search or alternate Pool-Generation Method 228 to retrieve textual objects 268, 272.

After CSPDM 66 subroutine has executed a particular search, the CSPDM 66 retrieves the appropriate data from the database 54, analyzes the data, and sends the data to the GUI Program 70 in order for the GUI Program 70 to display the results of the search on the display 38.

FIG. 4B illustrates that after the CSPDM 66 has completed the above procedure, the researcher has the option to exit the CSPDM 66 by logging off, executing a search based on the results of a previous search, or executing a new search.

FIGS. 4A and 4B also depict the seven subroutines of the CSPDM 66. There are three textual object search subroutines 224 and four pool search subroutines 212. The three textual object search subroutines 224 are: the Cases-In Subroutine 232, the Cases-After Subroutine 236 and the Similar Cases Subroutine 240. The four pool search subroutines 212 are the Pool-Similarity Subroutine 244, the Pool-Paradigm Subroutine 248, the Pool-Importance Subroutine 252, and the Pool-Paradigm-Similarity Subroutine 256. Each of these subroutines are described in more detail in FIGS. 4C to 4I. The following is a step by step description of the subroutines 224, 212 of the CSPDM 66.

Section A Cases-In Subroutine 232

FIG. 4C is a high level flow chart for the Cases-In Subroutine 232.

1. The researcher must select a single textual object 400.

2. The researcher selects the Cases-In Subroutine 232 option.

3. The Cases-In Subroutine 232 examines the n.times.n Opinion Citation Matrix and other matrices 404 created by the Proximity Indexing Application Program 62 and retrieves the textual objects to which the selected textual object refers 408, data relating to the number of times the selected textual object refers to the retrieved textual objects, data relating to the importance of each textual object, and other relevant data.

Section B Cases-After Subroutine 236

FIG. 4D is a high level flow chart for the Cases-After Subroutine 236.

4. The researcher must select a single textual object 400.

5. The researcher selects the Cases-After Subroutine 236 option.

6. The Cases-After Subroutine 236 examines the n.times.n Opinion Citation Matrix and other matrices 412 created by the Proximity Indexing Application Program 62 and retrieves the textual objects that refer to the selected textual object 416, data relating to the number of times the retrieved textual objects refer to the selected textual object, data relating to the importance of each textual object, and other relevant data.

Section C Similar-Cases Subroutine 240

FIG. 4E is a high level flow chart for the Similar-Cases Subroutine 240.

7. The researcher must select a single textual object 400.

8. The researcher selects the Similar-Cases Subroutine 240 option,

9. The Similar-Cases Subroutine examines the q.times.q Section Similarity Matrix and other matrices 420 created by the Proximity Indexing Application Program 62 and retrieves the textual objects that are similar to the selected textual object 424, data relating to the degree of similarity between the selected textual object and the retrieved textual objects, data relating to the importance of each textual object, and other relevant data. In order to be retrieved, a textual object must have a similarity coefficient with respect to the selected textual object of at least a minimum value. The preferred embodiment sets the minimum similarity coefficient of four percent (4%).

Section D Pool-Similarity Subroutine 244

FIG. 4F is a high level flow chart for the Pool-Similarity Subroutine 244.

10. The researcher must select a pool of full textual objects 428.

11. The researcher must then select a single full textual object 400 to which in compare the pool of full textual objects. It should be noted that the researcher can select the single textual object from the selected pool of textual objects, or the researcher can select a textual object from outside of the pool 432.

12. The Pool-Similarity Subroutine 244 examines the n.times.n Opinion Similarity Matrix and other matrices 436 and values created by the Proximity Indexing Application Program 62 for the selected full textual object and the pool of full textual objects.

13. The Pool-Similarity Subroutine 244 determines the degree of similarity of other full textual objects in the pool to the selected full textual object 440.

Section E Pool-Paradigm

FIG. 4G is a high level flow chart for the Pool-Paradigm Subroutine 248.

14. The researcher must select a pool of full textual objects 428.

15. The Pool-Paradigm Subroutine 248 examines the n.times.n Opinion Proximity Matrix, the n.times.n Opinion Similarity Matrix and other matrices and values created by the Proximity Indexing Application Program 62 for the pool of full textual objects 448.

16. The Pool-Paradigm Subroutine 248 determines the Paradigm full textual object by calculating the mean of the Euclidean distances of all the textual objects in the pool 452.

17. The Pool-Paradigm Subroutine 248 determines the similarity of the other full textual objects in the pool to the Paradigm full textual object 456.

Section F Pool-Importance Subroutine 252

FIG. 4H is a high level flow chart for the Pool-Importance Subroutine 252.

18. The researcher must select a pool of full textual objects 428.

19. The Pool-Importance Subroutine 252 examines 448 the n.times.n Opinion Citation Matrix, the n.times.n Opinion Similarity Matrix, numerical factors and other matrices and values created by the Proximity Indexing Application Program 62 for the pool of full textual objects 460.

20. The Pool-Importance Subroutine 252 then ranks the importance of each of the full textual objects in the pool 464.

FIG. 4I is a high level flow chart showing two possible alternate Pool-Paradigm-Similarity Subroutines 256.

Section G Pool-Paradigm-Similarity Subroutine 256 (Option 1) 256

21. The researcher must select a pool of k full textual objects where k equals the number of full textual objects in the pool 428.

22. For each of the k full textual objects, the Pool-Paradigm-Similarity Subroutine 256 selects a n.times.1 vector from the corresponding column of the n.times.n 468.

23. The Pool-Paradigm-Similarity Subroutine 256 creates an n.times.k matrix by grouping the n.times.1 vector representing each of the k full textual objects beside each other.

24. The Pool-Paradigm-Similarity Subroutine 256 calculates the mean of each row of the n.times.k matrix and enters the mean in the corresponding row of an n.times.1 Paradigm Proximity Vector 472.

25. The Pool-Paradigm-Similarity Subroutine 256 combines the n.times.1 Paradigm Proximity Vector with the n.times.n Opinion Proximity Matrix to create an (n+1).times.(n+1) Paradigm Proximity Matrix 476.

26. From the (n+1).times.(n+1) Paradigm Proximity Matrix, the Pool-Paradigm-Similarity Subroutine 256 evaluates the Euclidian distances and empirical data to create an (n+1).times.(n+1) Paradigm Similarity Matrix 480.

27. The Pool-Paradigm Similarity Subroutine 256 searches the row in the (n+1).times.(n+1) Paradigm Similarity Matrix that corresponds to the Paradigm full textual object and retrieves the full textual objects that have a maximum degree of similarity with the Paradigm full textual object 500.

Section H Pool-Paradigm Similarity Subroutine 256 (Option 2)

28. The researcher must select a pool of k full textual objects where k equals the number of full textual objects in the pool 428.

29. For each of the k full textual objects, the Pool-Paradigm-Similarity Subroutine 256 selects an n.times.1 vector from the corresponding column of the n.times.n 484.

30. The Pool-Paradigm-Similarity Subroutine 256 creates an n.times.k matrix by grouping the n.times.1 vector for each of the k full textual objects beside each other.

31. The Pool-Paradigm-Similarity Subroutine 256 calculates the mean of each row of the n.times.k matrix and enters the mean in the corresponding row of an n.times.1 Paradigm Pattern Vector PF 488.

32. The Pool-Paradigm-Similarity Subroutine 256 combines the n.times.1 Paradigm Pattern Vector PF with the n.times.n Opinion Pattern Matrix to create a (n+1).times.(n+1) Paradigm Pattern Matrix 492.

33. From the (n+1).times.(n+1) Paradigm Pattern Matrix, the Pool-Paradigm-Similarity Subroutine 256 evaluates the Euclidean distances between the rows of the Paradigm Pattern Matrix and creates an (n+1).times.(n+1) Paradigm Proximity Matrix 496.

34. From the (n+1).times.(n+1) Proximity Matrix, the Pool-Paradigm-Similarity Subroutine 256 evaluates the Euclidean distances between the rows of the (n.times.1).times.(n.times.1) Paradigm Proximity Matrix and empirical data to create an (n+1).times.(n+1) Paradigm Similarity Matrix 480.

35. The Pool-Paradigm Similarity Subroutine 256 searches the row in the (n+1).times.(n+1) Paradigm Similarity Matrix that corresponds to the Paradigm full textual object and retrieves the full textual objects that have a minimum degree of similarity with the Paradigm full textual object 500.

Application of the Proximity Indexing Technique

The above Proximity Indexing Application Program 62 and CSPDM 66 have a number of different applications and versions. Three of the most useful applications are described below.

The first type of Proximity Indexing Application Programs 62 is for use on very large databases. The matrices generated by this type of Proximity Indexer are "attached" to the database 54, along with certain clustering information, so that the database 54 can be searched and accessed using the Cases-In Subroutine 232, Cases-After Subroutine 236, Similar Cases Subroutine 240, Pool-Similarity Subroutine 244, Pool-Paradigm Subroutine 248, Pool-Importance Subroutine 252 and Pool-Paradigm-Similarity Subroutine 256 of the CSPDM 66.

The second type of Proximity Indexing Application Program 62 is a Proximity Indexer that law firms, businesses, government agencies, etc. can use to Proximity Index their own documents in their own databases 54. The researcher can navigate through the small business's preexisting database 54 using the Cases-In Subroutine 232, Cases-After Subroutine 236, Similar Cases Subroutine 240, Pool-Similarity Subroutine 244, Pool-Paradigm Subroutine 248, Pool-Importance Subroutine 252 and Pool-Paradigm-Similarity Subroutine 256 of the CSPDM 66. In addition, this type of Proximity Indexer Application Program 62 will be designed to be compatible with the commercial third-party databases 54 which are Proximity Indexed using the first type of program. In other words, the researcher in a small business may "weave" in-house documents into a commercial database 54 provided by a third party, so that searches in the large database 54 will automatically bring up any relevant in-house documents, and vice versa.

The third type of Proximity Indexing Application Program 62 involves the capacity to do Proximity indexing of shapes. Each image or diagram will be treated as a "textual object." The various matrix coefficients can be generated purely from topological analysis of the object itself, or from accompanying textual information about the object, or from a weighted combination of the two. The text is analyzed using the Proximity Indexing Application Program 62 as explained above. Shapes are analyzed according to a coordinate mapping procedure similar to that used in Optical Character Recognition ("OCR"). The numerical "maps" resulting from scanning the images are treated as "textual objects" that can be compared through an analogous weighing algorithm to generate a proximity matrix for every ordered pair of "textual objects" in the database 54. A similarity matrix can then be generated for each ordered pair, and the results organized analogous to a database 54 totally comprised of actual text.

This third type of Proximity Indexing Applications Program 62 can provide "Proximity Indexed" organization access to many different types of objects. For example, it can be used to search patent diagrams, or compare line drawings of known pottery to a newly discovered archeological find. It can be used to scan through and compare police composite drawings, while simultaneously scanning for similar partial descriptions of suspects. It can be used to locate diagrams of molecular structures, appraise furniture by comparing a new item to a database 54 of past sales, identify biological specimens, etc., etc.

FIG. 5A is a high level drawing that depicts one embodiment of the GUI Program 70 and its interaction with both the CSPDM 66 and the display 38. The GUI Program 70 has one or more display subroutines. One embodiment contains seven display subroutines. The seven subroutines comprise three textual object display subroutines 504 and four pool display subroutines 508. The three textual object display subroutines 504 are the Cases-In Display Subroutine (CIDS) 512, the Cases-After Display Subroutine (CADS) 516 and the Similar-Cases Display Subroutine (SCDS) 520. The four pool display subroutines 508 are the Pool-Similarity Display Subroutine (PSDS) 524, the Pool-Paradigm Display Subroutine (PPDS) 528, the Pool-Importance Display Subroutine (PIDS) 532 and the Pool-Paradigm-Similarity Display Subroutine (PPSDS) 536. The three textual object display subroutines 504 receive data from the corresponding textual object search subroutine 224 of the CSPDM 66. Similarly, the four pool display subroutines 508 receive data from the corresponding pool search subroutine 212 of the CSPDM 66. Once the display subroutines have processed the data received by the search subroutines, the data is sent to the integrator 540. The integrator 540 prepares the data to be displayed in the proper format on the display 38.

FIGS. 5B through 5H depict screens generated by the textual object display subroutines, CIDS 512, CADS 516 and SCDS 520. The three types of screens are the Cases In screen 1000, the Cases After screen 1004 and the Similarity Screen 1008, respectively. The Similarity Screen 1008 provides the most "intelligent" information, but all three screens generated by the textual object display subroutines 504 work in tandem as a system. The other screens created by the pool display subroutines are variances of these three, and also work in tandem with each other and with the three textual object display screens.

FIG. 5B depicts the "Cases After" 1004 Screen created by the CADS 516 for the textual object, Terry v. Ohio, 392 U.S. 1 (1968). The Cases-After subroutine 236 search produces all of the textual objects. in the designated field (here D.C. Circuit criminal cases since 1990) that cite Terry. The number "12" 1080 in the upper left hand corner indicates that there are a total of 12 such textual objects. The vertical axis 1012 indicates the degree to which a given textual object relied upon Terry. The number "10" immediately below the 12 indicates that the textual object in the field which most relied upon Terry namely U.S. v. Tavolacci, 895 F.2d 1423 (D.C. Cir. 1990), discusses or refers to Terry in ten of its paragraphs.

The Tear-Off Window 1016 feature is illustrated in FIG. 5B by the Tear-Off Window 1016 for U.S. V. McCrory, 930 F.2d 63 (D.C. Cir. 1991). The four Tear-Off Window active boxes 1020 (displayed on the Tear-Off Window 1016): 1) open up the full text 1104 of McCrory to the first paragraph that cites Terry; 2) run any of the three searches, namely Cases-In Subroutine 232 Cases-After Subroutine 236 or similar cases Subroutine 240 for McCrory itself (the default is to run the same type of search, namely Cases-After Subroutine 236 again); 3) hide the Terry execute search window 1024; and 4) bring the Terry Execute Search window to the foreground, respectively. The weight numeral 1028 indicates the number of paragraphs in McCrory that discusses or refers to Terry, in this textual object (in this example there is only one).

The Cases After screen 1004 for a given Textual object B displays a Textual Object Active Box 1032 representing every subsequent textual object in the database 54 that refers explicitly to Textual object B. The analysis starts with the same pool of material as a Shepards.TM. list for Textual object B. As well as some additional material not gathered by Shepards. However, the Cases After screen 1004 conveys a wealth of information not conveyed by a Shepards.TM. list.

The horizontal axis 1036 may represent time, importance or any other means of measurement to rank the textual objects. The Shepards list itself contains no information as to when a case was decided. The vertical axis 1012 similarly may represent any means of measurement to rank the textual objects. In the preferred embodiment, the vertical axis 1012 represents the degree to which the subsequent Textual object C relied upon the original Textual object B. The display 38 makes it obvious when a textual object has received extensive discussion in another textual object, or provides key precedent for a subsequent textual object, or merely mentions the earlier textual object in passing. It also provides guidance as to possible gradations in between extensive, or merely citing.

The "shape" of the overall pattern of active boxes on the Cases After screen 1004 provides a rich lode of information to be investigated. For example, a "dip" in citation frequency immediately after a particular textual object suggests that the particular textual object, while not formally overruling Textual object B, has largely superseded it. A sudden surge in citation frequency after a particular Supreme Court case may indicate that the Supreme Court has "picked up" and adopted the doctrine first enunciated in Textual object B. The researcher can instantly determine if the holding of Textual object B has been adopted in some circuits but not in others, if Textual object B is losing strength as a source of controlling precedent, etc. None of this information is now available to lawyers in graphical or any other form.

As with the Cases In screen 1000, every Textual Object Active Box 1032 on the Cases After screen 1004 is active, and includes a Tear-Off Window 1016 that may be moved by dragging on the tear-off window 1016 with a mouse 42, and that tear-off window 1016 becomes a text Tear-Off Window 1040, visible even when one moves on to other searches and other screens. Thus one may "tear off" for later examination every relevant citation to Textual object B, or even for a group of textual objects. The text tear-off windows 1040 "tile"; that is, they can be stacked on top of one another to take up less room. There is also a "Select All" feature (not shown), that creates a file containing the citations of every textual object retrieved in a given search.

In Cases After screen 1004 mode, clicking on the expanded-view button 1044 of the text tear-off window 1040 opens the text of the subsequent Textual object C to the first place where Textual object B is cited. A paragraph window 1048 displays a paragraph selection box 1052 indicating what paragraph in Textual object C the researcher is reading, and a total paragraph box 1056 indication how many paragraphs Textual object C contains in total. The user can view paragraphs sequentially simply by scrolling through them, or see any paragraph immediately by typing its number in the paragraph selection box 1052. Clicking on a Next paragraph active box 1060 immediately takes the researcher to the next paragraph in Textual object C where Textual object B is mentioned. Traditional Shepardizing allows the researcher to explore the subsequent application of a doctrine in a range of different factual situations, situations that help to define the outer contours of the applicability of a rule. Combining the expanded-view button 1044 functions and "Next Paragraph" active box 1060 functions allows the researcher to study how Textual object B has been used in all subsequent textual objects, in a fraction of the time the same task currently requires with available searching methods.

Perhaps the most fundamental form of legal research is "Shepardizing." A researcher starts with a textual object known to be relevant, "Textual object B," and locates the "Shepards" for that textual object. The "Shepards" is a list of every subsequent textual object that explicitly refers to Textual object B. The researcher then looks at every single textual object on the list. Shepardizing is often painstaking work. Many subsequent references are made in passing and have almost no legal significance. Although Shepards includes some codes next to its long lists of citations, such as "f" for "followed" and "o" for "overruled," the experience of most lawyers is that such letters cannot be relied upon. For example, the researcher may be citing Textual object B for a different holding than that recognized by the anonymous Shepards reader, interpreting Textual object B differently, or interpreting the subsequent textual object differently. However, for really thorough research, checking a Shepards type of list is essential. The researcher must make absolutely sure that any textual object cited as legal authority in a brief, for instance, has not been superseded by later changes in the law.

Very often, textual objects located on the Shepards list for Textual object B refer back to other important textual objects, some of which may predate Textual object B, all of which may be Shepardized in turn. This "zig-zag" method of research is widely recognized as the only way to be sure that one has considered the full line of textual objects developing and interpreting a doctrine. The real power of the Cases After screen 1004 emerges when it is used in conjunction with the Cases In screens 1000 and Similarity screens 1008. Using the preferred embodiment, the researcher may engage in the same kind of careful "zig-zag" study of a legal doctrine in a much more efficient manner.

For example, consider the following hypothetical search. The researcher reads Textual object B, and makes a list of every Supreme Court textual object it substantially relies upon, perhaps six textual objects. The researcher then Shepardizes Textual object B and reads each of those textual objects, in order to find other Supreme Court textual objects that they relied upon, perhaps eight. One then Shepardizes those fourteen Supreme Court decisions, in order to find any Court of Appeals cases in a selected circuit within the last three years on the same basic topic. This process would take at least an hour, even using Shepards through an on-line service. The same search can be performed with the present invention using the Cases In screens 1000 and Cases After screens 1004 in under five minutes.

In order to perform the same search, a researcher can pull up both the Cases In screens 1000 and Cases After screens 1004 for Textual object B simultaneously. The researcher can then "tear-off" all of the Supreme Court Cases on both lists, run Cases-After Subroutine 236 searches on every Supreme Court Case mentioned on either list, then examine the Cases In screens 1000 for all of the Supreme Court cases produced by these searches. The researcher can locate every recent Court of Appeals case from a selected circuit mentioned in any of those Supreme Court cases. Use of the Similarity screen 1008 as well, allows the researcher to find the pool of relevant Court of Appeals full textual objects even faster.

FIG. 5C depicts the Cases After Screen 1004 for U.S. v. Lam Kwong-Wah, 924 F.2d 298 (D.C. Cir. 1991). FIG. 5C shows a text Tear-Off Window 1040 on a Cases After Screen 1004, (in this textual object the Tear-Off Window 1016 for U.S. v. Barry, 938 F.2d 1327 (D.C. Cir. 1991), is opened using the full text active box 1064. A text Tear-Off Window 1040 containing the text of Barry opens, to the first cite of U.S. v. Lam Kwong-Wah at paragraph 15. Clicking on the Next Paragraph active box 1060 will open the text of Barry to the next paragraph that cites Lam Kwong-Wah.

The number "34" in the lower-left corner of the total paragraph box 1056 indicates that Barry has a total of 84 paragraphs in the cite U.S. v. Lam Kwong-Wah. Dragging the small squares 1068 to the left and below the text allow the researcher to move within a paragraph, and from paragraph to paragraph, in the text of Barry, respectively. The empty space below the text 1072 would contain the text of any footnote in paragraph 15. The compress window active box 1074 now closes the window and replaces it with the corresponding Textual Object Active Box 1032.

FIG. 5D depicts the Cases In Screen 1000 for U.S. v. North, 910 F.2d 843 (D.C. Cir. 1990). FIG. 5D contains a Textual Object Active Box 1032 representing every textual object or node with persuasive authority, cited in the text of North. The vertical axis 1012 represents the degree to which North relied upon a given textual object. In this example it is immediately apparent that Kastigar v. United States 406 U.S. 441 (1972) is the most important precedent, and its Tear-Off Window 1016 have been activated. The weight numeral 1028 indicates that Kastigar is referred to in 77 paragraphs of North.

A highlighted Textual Object Active Box 1076 can be created by clicking on it, as has been done with U.S. v. Lily, 651 F.2d 611. The number "212" in the case number box 1080 indicates that citations to two-hundred-twelve distinct texts appear in North. Fewer are visible because the textual object active boxes 1032 "tile" on top of one another; the "Zoom" feature is used to focus on a smaller area of the screen, and ultimately resolves down to a day-by-day level, making all the textual object active boxes 1032 visible.

The unique Cases In screen 1000 provides a schematic representation of the precedent from which Textual object A is built. The Cases In screen 1000 contains a textual object active box 1032 representing every textual object which is relied upon, or even mentioned, in Textual object A. Any citation in textual object A to a textual object that possesses potential persuasive authority, whether a statute, constitutional provision, treatise, scholarly article, Rule of Procedure, etc., is treated as a "textual object." The textual object active boxes 1032 are color-coded to indicate the court or other source of each textual object. Supreme Court cases are red, Court of Appeals cases are green, District Court cases are blue, and statutes are purple, for example. Each Textual Object Active Box 1032 contains the full official citation 1084 of its textual object. Clicking on any Textual Object Active Box 1032 immediately pulls up a larger window, known as a tear-off window 1016, also containing the full citation 1084 to the textual object (Tear-Off Window Citation 1088), its date 1092, its circuit 1096, and its weight numeral 1028 to the textual object being analyzed. The user may then drag the Tear-Off Window 1016 free of the Textual Object Active Box 1032 and release it.

This creates a text Tear-Off Window 1040 that remains visible until the researcher chooses to close it, no matter how many subsequent screens the researcher examines. The text Tear-Off Window 1040 can be moved anywhere by dragging it with the mouse 42. The text Tear-Off Window 1040 contains small text active boxes 1100 allowing the researcher to access or "pull up" the full text 1104 of the textual object it represents with a single click of the mouse 42. This feature also allows the researcher to run Cases-In Subroutine 232, Cases-After Subroutine 236, and Similar Cases Subroutine 240 searches on the textual object. (See below for a description of the Similarity screen 1008).

The organization of the boxes on the screen, including their position on the horizontal axis 1036 and vertical axis 1012, represents the real "intelligence" behind the Cases-In screen 1000. The horizontal axis 1036 in the preferred embodiment represents time, with the left margin 1108 corresponding to the present, i.e., the date 1992 when the search is run. The right margin 1112 represents the date of decision of the earliest textual object cited in Textual object A. (Certain special materials, such as treatises updated annually, and the U.S. Constitution, are located in a column 1116 to the left of the margin.)

The vertical axis 1012 in the preferred embodiment represents the degree to which Textual object A relied upon each particular textual object it contains. For example, if the Cases In screen 1000 is run on a district court case (Textual object A) which happens to be a "stop and search" textual object that mainly relies upon Terry v. Ohio, 392 U.S. 1 (1968), Terry will be at the top of the screen, with all other textual object active boxes 1032 appearing far below. The researcher can thus access the text of Terry directly without ever reading the text of Textual object A. Of course, the full text 1104 of Textual object A is also instantly available if desired. If the researcher wants to see where Terry "came from," the researchers can instantly, by clicking on a text active box 1100 within the Terry text Tear-Off Window 1040, run the Cases-In Subroutine 232 for Terry--and so on. There is no limit to the number of "levels" or "generations" the researchers may explore using this technique. It is therefore possible (assuming a sufficient database 54) to find, in a matter of seconds, without having to read through layers of texts, the possibly long-forgotten eighteenth-century precursors to a modern doctrine.

The Cases In screen 1000 creates an instant visual summary or "blueprint" of a textual object. The blueprint can help a researcher make a preliminary judgment about whether a particular textual object is worth closer examination. Viewing the Cases In screens 1000 for a group of textual objects allows a researcher to recognize whether there are precedents common to that group. The blueprint tells the researcher whether Textual object A is primarily a statutory construction case, a textual object that relies on local Court of Appeals cases without Supreme Court support, a textual object relying on precedent outside the circuit 1096 as persuasive authority, etc.

The initial Cases In screen 1000 presents every citation within a given textual object. In a textual object with an unusually large number of citations, the screen will be crowded with textual object active boxes 1032. The GUI therefore contains a "zoom" feature that allows the researcher to expand any small portion of the screen. To get back to the "big picture," the researcher simply selects the "Fit in Window" menu item, or else selects the "zoom out" feature. The same "zoom," "zoom out," and "Fit in Window" functions are present in the Cases After screen 1004 and Similarity screen 1008 as well.

The routine that calculates "degree to which Textual object A relies upon the cited textual object" clearly ranks major textual objects at the top, textual objects mentioned only in passing at the bottom, and textual objects of potentially greater relevance in between via display the appropriate textual object active boxes 1032 in the appropriate place. In addition, the routine can recognize when a highly relevant textual object is mentioned only in passing and give a higher weight to that textual object than it would otherwise receive in the ranking procedure.

The "intelligence" behind the entire GUI is driven by the knowledge that the lawyers do not want the computer to do legal analysis or make judgments for them, but simply guide them through the great mass of irrelevant material to those texts where lawyerly analysis of a problem begins.

The Cases In screen 1000 is designed with practical legal research in mind. It is common in legal research to locate a lower court textual object on the correct topic, call it "local Textual object A." However, the researcher desired to find the most persuasive authority available. The aim of this type of research is to find the "lead" textual object or textual objects o