US 5,440,481 AGrant
System and Method for Database Tomography
Issue Date:1995-08-08
•14 Claims
•4 Drawing Sheets
Abstract
A Process for analyzing full-text is provided for identifying often-repea, high user interest, word phrases in a database. Often-repeated, high user interest, word phrases are defined as pervasive theme areas (PTAs). The process also allows the relationship defined as connectivity among the various PTAs to be identified. In addition, phrases that are in proximity to the PTAs and which are strongly supportive of the PTAs are identified. Numerical indices, figure of merit, and user defined thresholds are used to quantify relations between PTAs and among PTAs and phrases.
Metadata
Assignee
- The United States of America as represented by the Secretary of the Navy
Inventors
- Ronald N. Kostoff
- David L. Miles
- Henry J. Eberhart
Application Information
Application Number:US 9673415
Filing Date:1992-10-28
Priority Date:1992-10-28
Art Unit:231
Classifications
IPC:
G06F 1720G06F 1730
Field of Search:
364395419.08;419.07;419.13;419.19600
Patent Drawings (4 sheets)
Description
Field of the Invention
The present invention relates to database searching and particularly to full text database search methods and their mechanization and implementation.
Background of the Invention
Due to the rapid expansion of electronic media storage capabilities, research analysts now have available massive databases of information which can be analyzed for various purposes. A major problem in practice is how to extract the essential information from these databases in a form readily amenable to analysis and interpretation. In other words, how does the analyst extract the collective wisdom contained in these large databases in a concise, readily understandable form.
The user may directly interact with the process of the present invention to select the often repeated phrases of high user interest referred to herein as pervasive theme areas (PTAs). Or, at the start of the process provide criteria for selection of the PTAs to enable the process of the present invention to define and select the PTAs from the sorted list content. Until recently, the co-word methodologies used key or index or title words as input, rather than input from the text directly. This led to universal criticism that the method input was subject to sources of bias and error from the indexing process, and the validity of the final results were questioned. In addition, the text database tended to consist of published papers, which had the index and key words. This also limited the types of text available for analysis. Recently, some progress has been reported in computer-assisted indexing using only the words found in the published literature.
Modern quantitative techniques utilize computer technology extensively, usually supplemented by network analytic approaches, and attempt to integrate disparate fields of information. One class of techniques exploits the use of co-occurrence phenomena. In co-occurrence analysis, phenomena that occur together frequently in some domain are assumed to be related, and the strength of that relationship is assumed to be related to the co-occurrence frequency. Networks of these co-occurring phenomena are constructed, and then maps of evolving topic fields are generated using the link-node values of the networks. Using these maps of structure and evolution, the information analyst can develop a deeper understanding of the interrelationships among the different information fields and the impacts of external intervention, and can recommend new directions for more desirable information portfolios.
One approach to co-occurrence analysis is co-word analysis. The origins of Co-word phenomena can be traced back to the pioneering work in: 1) lexicography to account for co-occurrence knowledge, and 2) linguistics to describe how affinity of two language units correlates with their appearance in the language.
In early co-word studies, words were classified on the basis of their co-occurrence with other words as well as their meanings. It was, however, observed that the reasons for two words co-occurring in the same context are not always relevant to a general linguistic description of a given language. The well-formedness of sentences to their lexical levels; i.e., how sensitive the meaning of a sentence is to substitution for one member of co-occurrence pair has been studied. A recent study included collocations as part of a linguistic model, whose goal was to relate any given meaning and all the texts that express it. Information retrieval research has focused on designing more efficient indexing tools using pairwise lexical affinities instead of keywords. Methods have been developed for locating interesting collocational expressions in a large body of text. These methods were based principally on the distribution of types and tokens in the body of text and on the analysis of the statistical patterns of neighboring words.
In the mid-1970s, a study was performed to examine relationships among themes in a novel using co-occurrence phenomena. An important term in the book was chosen, and a dictionary was constructed of all words in the book occurring in the same sentences as that word. A co-occurrence matrix which contained the co-occurrences among these related terms was constructed, and analyzed to eventually show the relations among all the associated terms in the mini-dictionary as they occurred in the original text. While the dictionary was restricted to single words, and the co-occurrence domain was restricted to sentences, the methodology did represent a major step forward in extracting word relations from text by their co-occurrences.
A recent update of this method employed frequency of co-occurrence to extract relatedness information from text. The study looked at co-occurrence using the sense-definition as the textual unit (entire definition of a sense of a word). The database used was the Longman Dictionary of Contemporary English (LDOCE) rather than free text. The method used single word frequencies only, and resulted in construction of networks of related words. It was concluded that co-occurrences of words in the LDOCE-controlled vocabulary in the definitions in LDOCE appeared to provide some useful information about the meanings of those words. Co-occurrences frequency correlated significantly with human judgements of relatedness, and the relatedness functions on co-occurrences yielded even higher correlations.
While the methods described above were useful for showing how relations among words and terms could be quantified and extracted from text, none were applied to extract relationships among topics of interest from large text databases.
It has been concluded in a recent study that co-word analysis may be satisfactorily performed on a set of documents by using either title words or keywords and that the main difference between results obtained is that keywords provided a much more detailed account of the subject studied for the case in point.
While there has been some progress in overcoming the dependency of co-word analysis on key or index words, limitations remain. From the above it is clear that the prior art has taken many varied approaches to search out related topic information from individual or multiple target databases. In light of the various problems and limitations of the foregoing approaches a simpler methodology less dependent on user bias and influences was imperative.
Summary of the Invention
It is thus an object of the present invention to provide a database search methodology that is less dependent upon or eliminates user input bias.
It is yet another object of the present invention to eliminate the need for user specified search criteria such as keyword requirements.
It is further another object of the present invention to provide a search methodology capable of performing upon full-text database elements.
It is still another object of the present invention to provide a data-base search methodology employing co-word relationships.
The present invention is a system and method of performing full-text data base analyses without the requirement for any key or index words. The database may include any form of combination of forms of text, be it published article, report, or memo. The invention is the most direct means of extracting messages from large textual databases without relying on interpreting intermediate abstractions of text such as citations, key or index words, or titles.
The invention provides a revolutionary approach for identifying often-repeated phrases (a phrase in the present context is from one to "N" words, e.g., [METAL, METAL MATRIX, OR METAL MATRIX COMPOSITES], etc.) in any size document or combination of documents (defined as the database). Those often-repeated phrases which also have high interest to the user(s) of the invention can be defined as pervasive theme areas (PTA). The invention allows the relationships (defined as connectivity) among these PTA to be identified. The invention also allows phrases which are in close physical proximity to the PTA, and which are strongly supportive of the PTA, to be identified. The PTA can be research areas of emphasis, organizations, journals, people, patents, etc. A major unique feature of this invention is that the database consists of full text (any combination of journal papers, reports, memos, written speeches, etc.), not abstractions of text such as key words or index words, as previous methods have used. The full text method displays the richness of the fine structure relations in the text, and provides orders of magnitude more detail and useful information than the methods which rely on key or index words.
Detailed Description of the Drawings
FIG. 1 is a functional block diagram of the method of the present invention.
FIG. 2 is a functional block diagram expanding on Block 18 of FIG. 1.
FIG. 3 is a functional block diagram expanding on Block 22 of FIG. 1.
FIG. 4 is a functional block diagram expanding on Block 26 of FIG. 1.
FIG. 5 is a functional block diagram expanding on Block 30 of FIG. 1.
Detailed Description
The present invention is a system and method as shown in FIG. 1. The system employs a scanner for digitizing document information content or other means for creating a data base in digital form. A computer program, for example as shown in Appendix A, can be used to make possible digital execution of the methodology of the present invention by means of a digital computer. A number of functions of the methodology of the present invention may be performed by human interaction. In the alternative a fully automated form of the system can perform all functions without human interaction where it is desirable to have the computer controlling the method of the invention by developing thresholds, numerical indices, figures of merit and other means of quantitative measurement directly from the database content itself. To those skilled in the art, manual data entry, optical scanning, and verbal input are means for information or data entry to a database of the type envisioned for application by the invention. Likewise any number of means for performing the functions of the invention including various types of digital computers and means for presenting or displaying the output of the inventions methodology may be used.
Referring to block 14 in FIG. 1 the invention 10 as depicted requires as a first step the assembly of information of interest to the user in a full-text database. Journals, papers, memos, reports, patents, and police records are just a few examples of the unlimited variety of documents that may be introduced into a database. The database may, of course, contain one or a combination of two or more documents. In the preferred embodiment of the invention the documents of interest are stored, regardless of the means of entry, on electronic media such as hard or floppy disk. CD-ROM, optical, and other high-density, rapid access storage media is also preferred for the system. Each individual or combination of documents may occupy a separate computer file in the system. In a preferred embodiment storage of digitally introduced documents has been in ASCII files. Other types of files will be dictated by the types of computer processing and the programs employed.
In the discussion that follows a phrase is defined as one or more words, where a word is used in the usual sense such as "the printed word". More specifically a phrase is one word, adjacent double words, adjacent triple words . . . to adjacent m*words. Thus a single word phrase might be "metal"; an adjacent double word phrase "metal matrix", and an adjacent triple word phrase "metal matrix composite".
The user is required to prepare a list of trivial phrases, e.g. it, to, if, and, but, etc., that the user wishes to have excluded from the database as it is subjected to the invention's methodology. In the preferred embodiment the number of trivial phrases that can be selected is unlimited except for limitations imposed by computer memory size. The system can exclude trivial phrases at any time, however, in a preferred embodiment trivial phrases are excluded immediately prior to or during processing to identify pervasive theme areas discussed hereinafter.
The next step in the method of the present invention is the identification of pervasive theme areas (PTAs) as depicted in block 18 of FIG. 1 and in FIG. 2. Thus the system and methodology are required to use the entire full-text database to create lists of phrases as at 40 in FIG. 2. More specifically the system and methodology require that separate lists of all one (1) word phrases, adjacent two (2) word phrases, adjacent three (3) word phrases . . . and adjacent n*word phrases be prepared. The phrases in each list are then counted as in 42 and then sorted in order in each list by frequency of occurrence as shown in 44 and in Table 1. The user, of course, can as an option sort the phrases alphabetically by consecutive word order in the phrase, i.e. e.g., first word, or second word, or through to the last word. The sorted phrases are then displayed as at 46 by any of various means known to those skilled in the art such as print-out list or graphs or video means to permit user interaction as at 48.
At 48 FIG. 2 the user selects pervasive theme areas (PTAs) which are often repeating phrases in the data base of high user interest. See Table 2.
The user may directly interact with the process of the present invention to select the often repeated phrases of high user interest (PTAs). Or, at the start of the process provide criteria for selection of the PTAs to enable the process of the present invention to define and select the PTAs from sorted list content. Many applications of the invention to date have focused on identifying PTA in databases which describe sponsored research programs. In these cases, phrases of high user interest are words which have high technical content, e.g., REMOTE SENSING, SIGNAL PROCESSING, HIGH TEMPERATURE SUPERCONDUCTOR. Thus, the selection of PTAs in some database analyses performed by the invention have used, for example, the 60 most often repeated phrases of high user interest as PTA. More or less than 60 phrases per PTA can be used, and, as stated above this can be determined by user interaction or pre-processing user supplied criteria for system definition of PTA composition.
The invention next identifies all phrases related to each PTA as depicted in 22 of FIG. 1 and FIG. 3. As shown at 50 the user identifies a range R of interest of plus or minus a number of single words about each PTA for the purpose of extracting phrases from the full-text database which occur physically close to the PTA. The invention has been found to produce good results using a set range of 150 single words of the PTA. It should be understood that selection of the size and range about the PTA is at the discretion of the user and may depend on the characteristics of the database. Next, for each PTA count all the phrases within the range of the PTA as depicted at 52 in FIG. 3. Sort by frequency of occurrence all phrases found for each PTA as depicted at 54. See Table 3.
Next, list each PTA and its related sorted list of phrases as at 56. See Table 3.
The invention next quantifies the strength of relationship between extracted phrases and each pervasive theme area (PTA) as depicted at 26 in FIG. 1 and 62 in FIG. 4. The user must first define, develop, and provide numerical indices and figures of merit for each of the extracted phrases. Examples of indices used in a preferred embodiment include the absolute frequency of occurrence of the PTA designated C.sub.j, the absolute frequency of occurrence of the extracted phrases, C.sub.i, and the extracted phrases within a set range of single words of the PTA designated C.sub.ij.
Examples of figures of merit used include the following ratios of the frequencies of occurrence described above: C.sub.ij /C.sub.i, C.sub.ij /C.sub.j, and (C.sub.ij 2)/(C.sub.i *C.sub.j) where 2 denotes squared and * denotes multiplication by. Other numerical indices and figures of merit may be used instead of or in addition to those described above. Table 3 shows the results of the quantification function of the invention. The user is thus in a position at 64 in FIG. 4 to select phrases that have a relationship to each PTA having determined relationships strengths based on quantifications obtained by applying the above numerical indices and figures of merit.
Pervasive theme areas (PTAs) which are closely related are next identified by the invention as depicted at 30 in FIG. 1 and in FIG. 5. The user is required to have defined and provided threshold values above some predetermined number of the figures of merit for selecting phrases of high user interest. In the presently embodied invention the selection of threshold values for the identification of closely related PTAS is at the discretion of the user. The system can, however, be implemented to select threshold values based on an automated analysis of the full-text database being operated upon. The user or system can then select phrases of high interest whose figures of merit are above the threshold. See Table 4.
The commonality, i.e. the degree of similarity or close relatedness, of extracted phrases among the different PTA is next computed to identify phrases in common between each PTA's related phrases as at 74 in FIG. 5. In the preferred embodiment of the present invention commonality was defined as the numbers of phrases in common among PTA, and in particular the number of extracted phrases in common between all the PTAs. Other definitions of commonality may, of course, be formulated.
PTA which are related based upon the number of phrases in common are next identified as at 76 in FIG. 5. Specifically groups of PTA are generated such that each PTA in a given group has extracted phrases in common with at least one other PTA in the group. See Table 5.
In a preferred embodiment each PTA in a group contains extracted phrases in common beyond some predetermined threshold value with at least one other PTA in the group. Other inclusion or membership criteria may be used.
Finally, as shown in FIG. 1, the relationships among related PTA and among PTA and related phrases are graphically displayed. In a preferred embodiment groups of related PTAs with each PTA in the group are listed. A generic title is used for each PTA group. See Table 6.
Table 6
Groups of Closely Related Phrases
GROUP 1. IONOSPHERIC HEATING/MODIFICATION: *RADIO WAVE; *WAVE PROPAGATION; *QUANTUM ELECTRON; *IONOSPHERIC MODIFICATION; *PHASE CONJUGATION.
GROUP 2. IMAGE/OPTICAL PROCESSING: *PARALLEL PROCESSING; *PATTERN RECOGNITION; *IMAGE PROCESSING; *COMPUTER VISION; *DIGITAL COMPUTER; *ARTIFICIAL INTELLIGENCE; *DATA PROCESSING; *COMPUTER SCIENCE; *OPTICAL PROCESSING; *SPATIAL LIGHT MODULATOR; *SIGNAL PROCESSING; *LIQUID CRYSTAL; *LIGHT MODULATOR; *PROGRAMMING LANGUAGES; *INTEGRAL EQUATIONS.
GROUP 3. AIR-SEA INTERFACE: *SURFACE WAVE; *OCEANIC PHYSICS; *INTERNAL WAVE; *SEA SURFACE; *BOUNDARY LAYER; *ATMOS OCEANIC PHYS; *REMOTE SENSING.
GROUP 4. LOW OBSERVABLE: *LOW OBSERVABLE; *THIN FILM.
GROUP 5. EXPLOSIVE COMBUSTION: *KINETICS AND CATALYSIS; *SOLID FUEL; *EXPLOSION AND SHOCK; *SHOCK WAVE; *CHEMICAL PHYSICS; *EXPLOS SHOCK WAVE; *STRENGTH MATER; *FRACTURE MECHANICS; *COMPOSITE MATERIALS.
GROUP 6. PARTICLE BEAMS: *NEUTRAL BEAM; *PARTICLE ACCELERATOR; *ATOMIC ENERGY; *PLASMA PHYSICS; *ELECTRON BEAM; *CHARGED PARTICLE ACCELERATOR; *CHARGED PARTICLE.
GROUP 7. AUTOMATIC/REMOTE CONTROL: *AUTOMATIC CONTROL; *REMOTE CONTROL
GROUP 8. FREQUENCY STANDARDS: *FREQUENCY STANDARD; *HYDROGEN MASER
GROUP 9. RADAR CROSS SECTION: *CROSS SECTION; *ELECTROMAGNETIC WAVE; *RADIO ENGINEERING.
Other relational display mechanisms such as Venn-type diagrams may be used in the alternative. Also in a preferred embodiment the graphical display of the relationship between each PTA and its extracted phrases present the extracted phrases assigned to quadrants based on the magnitudes of the two figures of merit (C.sub.ij /C.sub.i) and (C.sub.ij /C.sub.j). Each quadrant depicts a specific type of relationship between the PTA and the extracted phrases in that quadrant. See Table 7.
Alternative means for displaying relationships would include, but not be limited to, two or three-dimensional plots of the relationship between each PTA and its extracted phrases using figures of merit of the extracted phrases as plotting variables.
The invention as described is easily mechanized using a variety of existing hardware elements such as optical scanners computers and displays. A program used by the system of the present invention to implement the full-text database process of the present invention appears in the Appendix to this specification. The code was written in Think Pascal.TM. on the Macintosh. The Macintosh was chosen for the user interface, speed, and memory management capabilities. An application on a Macintosh can access all of the free memory management capabilities of the machine without any limitations like the 640K "barrier" on IBM PC's. Think Pascal.TM. was chosen as the programming language because of its efficient compiler and powerful debugger.
The variable and procedure names indicate what they are used for. There are a few subtleties that should be understood. First, whenever a list is mentioned in the block diagrams of the process it is implemented using a balanced binary tree. The main advantages of a binary tree are the fast search time and dynamic memory allocation. Each time the binary tree is changed it is checked to see that all branches are close to the same length. By doing this it should be understood that to search for a given item in a tree containing n items will take at most log.sub.n (n) comparisons. This is much better than n/2 for a sort list. The fact that the tree is made up of individual data structures linked together by pointers means that the user doesn't have to know how big the tree will get. The user simply has to know if there is enough memory to add the nest item to the tree. Second, there is no double counting of words.
It should be understood that there are numerous variations and modifications of the invention that will be readily apparent to those skilled in the art in light of the above teachings. It is, therefore, to be understood that, within the scope of the appended claims, the invention may be practiced other than as specifically described. ##SPC1##
Claims
What we now claim for our invention is:
1. A system for full-text database searching, for identification of often repeated phrases which by virtue of their repeated occurrence, frequency of occurrence above a user-set threshold, or user input constitute phrases having a high user-interest designated as pervasive them areas (PTAs), said phrases consisting of one to n words (n*words), where n is an integer, in one or more documents defined as the database, relationships defined as connectivity among said PTAs, and phrases in close physical proximity to and which are supportive of said PTAs, comprising: means for introducing document information content into a full-text database in digital form; means for digitally storing said database; means for processing said digitally stored database; means operatively associated with said processing means and said storing means for identifying pervasive theme areas (PTAs) defined as often-repeating word phrases consisting of one or more adjacent words such that said phrases are one word phrases, adjacent 2 word phrases, adjacent 3 word phrases . . . and adjacent n* word phrases, and for entering said phrases in said storing means; means for identifying phrases in said database related to said PTAs, said phrases being defined as m words, where m=1,2,3, . . . n and where each word phrase for m=2,3, . . . n is composed of adjacent words, said word phrase for m=1 being a single word phrase, for m=2 a double word phrase, for m=3 a triple word phrase . . . and for n=m an nth word phrase, by applying a user specified range of interest R expressed as a number of single words appearing both before and after said PTAs, and for storing said identified phrases in said storing means; means for counting for each PTA the extracted phrases within said range of said PTA stored in said storage means, sorting all phases found for each PTA by frequency of occurrence, listing each PTA and its related sorted list of extracted phrases, and storing said counts and said lists of PTA's and their related sorted list of extracted phrases in said storing means; means for quantifying the strength of relationship between extracted phrases and each pervasive theme area (PTA) applying user-predefined numerical indices and figures of merit, and providing the results of said quantifying means to said storing means; means for obtaining the results of said quantification from said quantifying means and said storing means and presenting said results to said user for user-selection of phrases having a relationship to each PTA predicated on the relationship strengths obtained by said quantifying means; means for identifying PTAs which are closely related, said means employing user-input figure of merit threshold values above a user-predetermined number for selecting phrases of high-user interest, said means storing identified closely related PTAs in said storing means; means for identifying phrases in common among PTA and storing those identified in said storing means; means for identifying and grouping related PTA based upon the number of phrases in common among the PTA, said number being above a user-input predetermined threshold, each group having at least one PTA having extracted phrases in common with one or more other PTA in said group, said groupings of PTA's stored in said storing means; and means for displaying relationships among related PTA and between PTA and related phrases said display means connected to said processing means.
2. The system of claim 1 wherein said means for identifying pervasive theme areas in said database, comprises: a means for counting frequency of occurrence of said n* word phrases; a means for creating a list of all n* word phrases and the frequency of occurrence for each of said n* word phrases; a means for sorting said list of n* word phrases by frequency of occurrence: a means for defining pervasive theme areas from said list of sorted phrases; and a means for selecting the number of said n* word phrases to be used as pervasive theme areas.
3. The system of claim 2 wherein said means for identifying phrases in said database related to said pervasive theme areas (PTAs), comprises: a means for extracting phrases for each PTA from the full-text database which occur within a set range of single words of the PTA; and, a means for listing the extracted phrases for each PTA and their frequency of occurrence in the database.
4. The system of claim 3 wherein said means for quantifying the strength of relationship between said phrases and each pervasive theme area, comprises: a means for producing numerical indices and figures of merit from said database; and a means for applying said numerical indices and figures of merit to quantify the strength of relationship of each said phrase and each pervasive theme area.
5. The system of claim 4 wherein said means for producing numerical indices and figures of merit employs definitions for numerical indices selected from the group consisting of absolute frequency of occurrence, denominated C.sub.j, of the pervasive theme area, absolute frequency of occurrence of the extracted phrases C.sub.i for single, adjacent double, adjacent triple . . . adjacent nth, and C.sub.ij the frequency of occurrence of the extracted phrases (single, adjacent double, adjacent triple, . . . adjacent mth) within a set range of single words of the pervasive theme area.
6. The system of claim 5 wherein said means for producing numerical indices and figures of merit employs definitions for figure of merit selected from the group consisting of the ratios of the frequencies of occurrence C.sub.ij /C.sub.i, C.sub.ij /C.sub.j, and (C.sub.ij 2)/(C.sub.i *C.sub.j).
7. The system of claim 6 wherein said means for identifying pervasive theme areas (PTAs) which are closely related comprises: a means for defining threshold values of the figures of merit for selecting single, adjacent double, adjacent triple, . . . adjacent mth word phrases of high user interest; a means for selecting phrases of high interest having figures of merit above said threshold, said selection being made from the list of extracted phrases; a means for computing commonality defined as the degree of similarity or close relatedness of extracted phrases between different pervasive theme areas; and a means for generating groups of PTAs, each PTA in a group having extracted phrases in common with at least one other PTA in the group.
8. The system of claim 1 wherein said means for identifying pervasive theme areas in said database comprises a means for creating a list of all phrases and for each phrase indicating its order in said list in accordance with a user prespecified sort criteria; a means for sorting said phrases in accordance with said user pre-specified sort criteria; a means for defining pervasive theme areas from said phrases; a means for selecting a number of said phrases to be used as pervasive them areas.
9. The system of claim 8 wherein said sort criteria is alphabetical by consecutive word order in said phrase.
10. The computer implemented method of full-text database searching, comprising the steps of: a. assembling information into a full-text database by scanning documents and storing digitized results in said computer; b. eliminating trivial phrases from said databases by comparing a user-input list of such phrases with the entire contents of said database and deleting matches with said list; c. using the definition of phrase as m*word=phrase where m=1,2,3, . . . n and where each word phrase for m=2,3 . . . n is composed of adjacent words, said word phrase for m=1 being a single word phrase, for m=2 an adjacent double word phrase, and for m=3 an adjacent triple word phrase, . . . and for m=n an adjacent nth word phrase, creating a list of all single word phrases, a list of all adjacent double word phrases, a list of all adjacent triple word phrases, . . . , and a list of all adjacent nth word phrases and their frequencies of occurrence in the database; d. sorting each list of said phrases by their frequency of occurrence in said list; e. identifying pervasive theme areas in the information in said database; f. defining pervasive theme areas from said sorted list of phrases as the most frequently occurring phrases of high user-interest. g. identifying phrases in said database that are related to said pervasive theme areas; h. quantifying strength of relationships between said identified phrases and said pervasive theme areas; i. identifying pervasive theme areas which are closely related; j. displaying relationships among related pervasive theme areas and pervasive theme areas and related phrases; wherein the step of identifying phrases related to pervasive theme area further comprises the steps of k. extracting phrases for each pervasive theme area (PTA) from the full-text database which occur within a user-identified range of interest plus or minus a range of words of the PTA; and l. listing the extracted phrases and their frequency of occurrence in the database for each PTA.
11. The method of claim 10 wherein the step of quantifying the strength of relationship between phrases and each pervasive theme area (PTA) further comprises the steps of: a. preparing numerical indices and figures of merit for quantifying strength of relationship between extracted phrases and their PTA; and b. applying said numerical indices and figures of merit to said phrases relative to their respective PTA.
12. The method of claim 11 wherein the step of identifying pervasive theme areas (PTA) which are closely related further comprises the steps of: a. defining threshold values above some predetermined number for the figures of merit for selecting phrases; b. selecting, from the list of extracted phrases for each PTA, phrases of high user interest having figures of merit above the threshold; c. computing commonality of extracted phrases among the different phrases in terms of the numbers of phrases in common among PTA, or equivalent; and, d. generating groups of PTA such that each PTA in a given group has extracted phrases in common with at least one other PTA in the group.
13. The method of claim 12 wherein said step of identifying pervasive theme area further comprises the steps: creating a list of all phrases sorted and ordered in accordance with frequency of occurence of said phrases; sorting said ordered phrases in accordance with a user pre-specified user-interest criteria; defining pervasive theme areas from said user interest phrases; and selecting the number of said phrases to be used as pervasive theme areas.
14. The system of claim 13 wherein said sort criteria is alphabetical by consecutive word order in said phrase.
Patent Citations (16)
| Patent | Date | Inventor | Cited By |
|---|---|---|---|
| US4839853 | 1989-06-01 | Deerwester et al. | |
| US4849898 | 1989-07-01 | Adi | |
| US4942526 | 1990-07-01 | Okajima et al. | |
| US4992972 | 1991-02-01 | Brooks et al. | |
| US4994967 | 1991-02-01 | Asakawa | |
| US5056021 | 1991-10-01 | Ausborn | |
| US5070456 | 1991-12-01 | Garneau et al. | |
| US5146405 | 1992-09-01 | Church | |
| US5276616 | 1994-01-01 | Koga et al. | |
| US5280573 | 1994-01-01 | Kuga et al. | |
| US5295261 | 1994-03-01 | Simonetti | |
| US5301109 | 1994-04-01 | Landauer et al. | |
| US5311429 | 1994-05-01 | Tominaga | |
| US5321833 | 1994-06-01 | Chang et al. | |
| US5325298 | 1994-06-01 | Gallant | |
| US5333313 | 1994-07-01 | Heising |