Text

Method for extracting company names from text

5287278

Abstract

A method for extracting company names from textual information uses a combination of heuristics, exception lists, and extensive corpus analysis. The method first locates company name suffixes (i.e., Company, Corporation) and attempts to locate the beginning of the company name. The method works on both mixed-case text and capitalized text. Upon identification of a company name, the method proceeds to generate variations of the name for later extraction.


Claims

What is claimed is:

1. A method, for practice on a computer, for extracting company names from text comprising the steps of:

scanning said text for a company name indicator;

reading words backwards through said text until a stop condition comprising reading a word which is on a stop list occurs; and

extracting the words read by said computer, before the stop condition occurred.

2. A method, for practice on a computer, for extracting company names from text comprising the steps of:

scanning said next for a company name indicator;

determining if there is parallel sentence structure in the text including said company name indicator;

reading words backwards through said text until a stop condition occurs; and

extracting the words read by said computer, before the stop condition occurred.

3. A method, for practice on a computer, for extracting company names from text comprising the steps of:

scanning said text for a company name indicator;

determining if a verb following said company name indicator is plural;

reading words backwards through said text until a stop condition occurs; and

extracting the words read by said computer, before the stop condition occurred.


Description

A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure, as it appears in the Patent and Trademark Office patent files or records, but otherwise reserves all copyright rights whatsoever.

BACKGROUND OF THE INVENTION

The present invention relates generally to automated information retrieval, and more particularly to a system and method for extracting company names from text.

One of the major problems in the accurate analysis of natural language is the presence of unknown words, especially names. While names account for a large percentage of the unknowns in a text, they can also be the most important piece of information in a text; for what the text is about (topic analysis), for extracting information from text (data base generation), and for indexing a text for full-text retrieval.

Company names are particularly important for knowledge-based financial applications. With mixed case input, a program can easily extract company names by looking backward from a company name indicator (i.e., Incorporated, Corporation, etc.) to the first non-capitalized word. This simple heuristic fails to correctly identify approximately 10% of real company names and fails entirely with upper case input.

What is needed is a more accurate method which extracts company names from mixed case text and which also works for upper case text.

SUMMARY OF THE INVENTION

In accordance with the present invention, there is provided a method for extracting company names from text using a combination of heuristics, exception lists, and extensive corpus analysis. The method first locates a company suffix and then reads backwards from the suffix. With mixed case input, the method utilizes a combination of searching for the first non-capitalized word and special heuristics for dealing with embedded conjunctions and other non-capitalized words. With upper case input, the method utilizes an empirically derived stoplist, company word length restrictions, and lexical lookup to determine the start of a company name. Once a company name is identified, the invention generates the most likely variations that those names may go by for use in subsequent retrieval.

BRIEF DESCRIPTION OF THE DRAWINGS

While the novel features of the invention are set forth with particularity in the appended claims, the invention, both as to organization and content, will be better understood and appreciated, along with other objects and features thereof, from the following detailed description taken in conjunction with the drawings, in which:

The single Figure is a flowchart of the method of the present invention.

DESCRIPTION OF THE INVENTION

The present invention is specifically directed to a company name extraction method. It is contemplated that the present invention could be utilized in a larger automated natural language text processing system which includes other modules, for example, for performing syntactic segmentation of the text as an aid to understanding certain sentences and helping to identify points within sentences which could not be included in the boundaries of a company name (phrase boundaries). The present invention has been implemented on a SUN Workstation using LISP as a source language, although it is contemplated that other languages and/or computers could be used to practice the present invention.

The LISP source code which can be used to implement the present invention is provided in Appendix A. An overall description of the process, as performed by a computer in response to the source code in Appendix A, makes up the rest of the detailed description of the present invention.

The program first scans through the input text, looking for company name indicators. If the text is mixed-case, a word must begin with a capital letter in order to be an indicator. The following words are indicators (words prefaced by ABBREV must have periods after them). 

    ______________________________________
    ABBREV.sub.-- INC
               ABBREV.sub.-- LTD
                             ABBREV.sub.-- CORP
    ABBREV.sub.-- CO
               ABBREV.sub.-- PLC
                             ABBREV.sub.-- AG
    ABBREV.sub.-- COS
               ABBREV.sub.-- LP
                             ABBREV.sub.-- L.P
    CORP       INC           LTD
    CO         PLC           AG
    NV         CSF SA        ABBREV.sub.-- ENTRP
    ABBREV.sub.-- S.A
               ABBREV.sub.-- SA
                             ABBREV.sub.-- PTY.LTD
    ASSOCIATES COMPANY       COMPANIES
    CORPORATION
               INCORPORATED  LIMITED
    PARTNERS
    ______________________________________


After a company name indicator is detected, the program looks backwards to determine where the company name begins. If the indicator is NV or CO, it makes sure the previous word is not a city in Nevada or Colorado, respectively. If it is, no company name is extracted. Otherwise, the program looks at up to six words, not including punctuation, that appears before the indicator. If no other stop condition occurs, all six words are taken to be the constituents of the company name and it is extracted.

There are several stop conditions which will each be described.

Additional company name indicators that appear before the final indicator are included as a part of the company name being extracted. One stop condition occurs when the program encounters one of the following words in all-caps input: 

    __________________________________________________________________________
    ABOUT ABOVE ACQUIRE ACQUIRES ACQUIRING AFFILIATE AFFIRMS
    AFTER
    AGAINST ALL ALLOW AN APPROVES ARE AS AT
    BELIEVES BE BEFORE BEGIN BETWEEN BOTH BOUGHT BUY BUYS BY
    CERTAIN COMPANY COMPLETES CONCERN CONNECT CONTACT COVER
    DIRECTORS DISTRIBUTE DOWNGRADES
    EST EVEN EXPECT
    FILES FOR FORCE FORMER FORMERLY FRIDAY FROM
    GROUP
    HAD HAS HAVE HE HELD
    IN INACTIVE INCLUDE INCLUDES INCLUDING INITIAL
    INVOLVE INTO IT ITS IS
    JOINS
    LEAVING LEFT LONGTIME
    MAKER MEAN MONDAY
    NAME NEWSWIRE
    ON ONE OR OTHER OUT OUTSTANDING OVER OWN OWNS
    PARENT PARTNER PR PRESIDENT PUBLISHER PURCHASE
    REQUIRE RESUMED RETAILER
    SAID SAYS SAY SATURDAY SHOWS SOLD SPLIT
    STOP SELL SUBSIDIARY SUBSIDIARIES SUNDAY
    TEXT TO TODAY THAN THAT THE THEIR
    THREATENING THROUGH THURSDAY TUESDAY
    UNDER UNIDENTIFIED UNIT UNTIL UPI USE USING USUAL
    VIA VS
    WAS WEAKENS WEDNESDAY WERE WHEN WHEREBY WHICH WIRE WITH
    YESTERDAY
    __________________________________________________________________________


If any of the above words are encountered, the company name extracted beings after the word. for the following words, the Company name is assumed to start with the word: 

    __________________________________________________________________________
           UNITED APPLIED ALLIED CONSOLIDATED DIVERSIFIED
    INTEGRATED ADVANCED
    __________________________________________________________________________


Another stop condition occurs when the program encounters a non-capitalized word in mixed-case input that is not a coordinator. Coordinators are: 

    ______________________________________
    AND        DE     VAN         DU   OF
    ______________________________________


Company names containing only an indicator are not allowed.

If an AND appears within the six-word window and either there are more than 2 commas within this window, the company name extracted begins with the word after the AND.

If an OF appears within the six-word window and the word directly before the OF is one of the following: 

    ______________________________________
    BOARD          DIVISION    OFFICER
    PROGRAM        PROGRAMS    DIRECTOR
    SHAREHOLDERS               EXECUTIVE
    ______________________________________


or the words directly before the OF constitute a person name, then the company name extracted begins with the word after the OF.

If the word AND appears and conjoins another company name, and there is parallel sentence structure or plural verbs, the company name extracted begins with the word after the AND. An example of parallel sentence structure is IBM, GE and HP each . . . or IBM, GE, and HP all.

If the symbol & appears, the company name extracted terminates according to the normal stop condition or at the first comma detected.

If there is a comma within the words under consideration, the sentence is bracketed with respect to syntactic segmentation. If the word or words before the comma belong to a separate syntactic constituent, they are not included in the company name.

Referring now to the single Figure, there is shown a flowchart of the method of the present invention. The method begins at 101 by detecting a company suffix. As described above, the program checks the suffixes CO and NV to make sure they are not state abbreviations. At block 103 words are read one at a time up to six words before the suffix. If the text is mixed case, the method checks for an uncapitalized word at 105. If the word is uncapitalized and not de, van, or du, then the company name is extracted and the program exits at 107. If the word is capitalized, it is taken as part of the company name unless it is on a sentence or phrase boundary as determined at 113. If a sentence or phrase boundary is detected, the program exits at 115. The test for a conjoined name at 109 refers to the situation when the word AND appears in a company name as discussed above. If the AND is not part of the company name, the program exits at 119. If the text is all caps, the various stop conditions described above are checked for at 111. If a stop condition is found, the program exits at 117. Company names located can be stored in a database and used for future detection of those company names (and variations, such as without the suffix, etc.) previously identified.

While specific embodiments of the invention have been illustrated and described herein, it is realized that modifications and changes will occur to those skilled in the art. It is therefore to be understood that the appended claims are intended to cover all such modifications and changes as fall within the true spirit and scope of the invention. ##SPC1##

«Previous Next»
Word processor having a word frequency count unit Data entry and error embedding system