JPH0727532B2 - Information retrieval method and system - Google Patents

Description

Detailed Description of the Invention

[0001]

The present invention relates to the storage, retrieval, and storage of information such as documents, drawings, and photographs so that general users can easily understand the information.
The present invention relates to a search method and system that can be displayed.

[0002]

2. Description of the Related Art Conventionally, the management of a database that enables the storage and retrieval of a large amount of information and the processing of retrieval have been entrusted to experts, and end users obtain information through the intervention of such experts. Was there. However, with the recent realization of small-sized large-capacity storage devices such as optical disk storage devices,
A document filing apparatus for an office, which is directly operated by an end user, has been realized. In addition, word processors have become widespread, and a large amount of documents have been stored in electronic devices.

However, in the conventional computer technology, objects such as these documents are given unique names,
Bibliographic items such as titles and author names are displayed in a table format for management, or keywords or classification codes are further added to facilitate retrieval, but there are the following problems.

First, in most of the file systems of computers, files are managed with unique names (about 20 characters), but it is difficult to name them so that they are easy to remember, and the contents are imagined from the character strings. However, finding the file you want is extremely difficult even for the person who created it.

[0005] Although there is no difficulty in the case bibliographic data is to be registered because it is an objective fact, the situation is less likely to use it to search means. It is limited to cases where the documents and references are clearly known.

In many cases where a document or the like is retrieved, a title that is only vaguely remembered or the content of the document is a clue. For this reason, keywords and classification codes are used, but it is difficult to add them when first registering a document. In other words, it is a difficult problem to use what is the keyword so that it can be appropriately selected later. For example, if a large number of keywords are added in order to enable search from various angles, many unnecessary items will appear when searching. If there are few, you will not know if you will be selected correctly. Conventionally, experts have extracted and assigned keywords and classification codes in literature databases and the like.

Further, it is difficult to recall a keyword when searching. For example, in the case of a document search, it is not uncommon to find a document that is similar to what you want from the general catalog, pick up the keyword that is attached to it, and use it as a reference when creating a search formula that consists of keywords. From this it can be seen how difficult it is to assume.

Further, the document filing by classification also causes the ambiguity of the classification tree (hierarchical tree) and the intersection of classification trees (when one document is multiply classified) when it is subdivided. There is. Another problem is that the classification criteria change over time and often become completely useless after several years.

As described above, in the filing method for documents and the like according to the conventional technique, it is a very important subject to make the management and retrieval easy, especially for the end user.

For these problems, the literature [J. F.
Sowa: “Conceptual Graphs for Adebase Interface” Ivy MG.
Research and Development, Vol. 20, 19
76 Phi 336-357 (JF Sowa: "Conceptual gr
aphs for a Data Base Interface, ”IBM J. Research a
nd Development, Vol.20, 1967, pp.336-357)], a method of schematizing a search condition and a method of deriving a formal search condition expression when performing a search in a natural language are proposed. In addition, the literature [F. N. Toe. Etol, “Rabbit: Anne
Enrichent Database Assistant ”, Proceedings of National Conference of Air Ey, 1982, Phi: 314-318 (F.
N. Tou, et al. “RABBIT: An Intelligent Database
Assistant, ”Proceedings of National Conference of
AAAI, 1982, pp.314-318)], a method is proposed in which knowledge about the contents of a database is presented from a computer to support the construction of a search conditional expression. These methods focus only on the support for searching the database, and do not include the support for storing information such as data input and update.

[0011]

When filing a document by an end user, it is important to make it easy to search for registration of a new document, maintenance of a file system (review of appropriateness of classification, etc.). The above proposed method is not sufficient.

Furthermore, the following problems remain in the search. In other words, a new concept (view) that has not been clarified at the time of storing the information does not provide a means for reviewing old information or searching with that view. For example, after several years, it is required to reclassify from a new angle or to classify from a viewpoint different from that of a person, which is an important factor assuming ease of use of the information storage device.

An object of the present invention is to solve the above problems and allow an end user to easily retrieve a desired document or the like from ambiguous and fragmentary information, and at the same time register the document or the like. To make it easier.

[0014]

In order to achieve the above object, the retrieval system of the present invention makes it clear to the end user by making the storage system of information in the machine conform to the human's memory characteristic and the way of understanding the meaning. An easy-to-use and easy-to-use search system such as a document filing device is provided.

Particularly, in the present invention, for the purpose of supporting the input of search conditions as described later, a plurality of terms and concepts indicating the concept and the concept
The knowledge base consists of multiple predicates that represent relationships between
It is stored as a search condition for information retrieval.
Knowledge base when a phrase that represents the first concept is selected
To show the concepts that are related to the first concept
With a predicate that represents the relationship between one phrase and the first concept
The set is displayed on the display screen, and the set of the displayed phrase and predicate is displayed.
Select a set of words and predicates that shows the desired second concept
And the words that represent the second concept are related by the selected predicate.
The words that indicate the first concept attached are used as new search conditions.
Generated, and referring to the knowledge base,
At least one phrase and a second synopsis indicating related concepts
It is characterized by displaying on the display screen a set of predicates that represent the relationship with the mind .

[0016]

The present invention specifically realizes easy registration of new information, easy input of search conditions, realization of ambiguous and semantic search, and realization of search adapted to various viewpoints of objects. Is possible.

To realize these, the following new functions are provided.

(1) Registration support function To register a new document, bibliographic items (author name,
In addition to inputting the title, source, etc.), it is necessary to enter the subject and the character of the document. Moreover, in order to enable a high-level semantic search, it is necessary to add more detailed information about the information.

For example, if the subject is a computer, what kind of computer it is, what characteristics it has, which company developed it, where it is located, and which country it belongs to. Information such as, is required. If this information is stored,
"A computer developed by a company in the country A, B
It is possible to search for "a computer document having the following characteristics" by using the inference function.

According to the present invention, knowledge about concepts such as "computer" and "company" is stored in the storage device, and interactively presenting to the user what kind of attribute data should be added when new information is added. However, not only will you be able to enter data in a short time, but you will not be able to enter incorrect information.

Further, when similar information is already registered, a function of inputting only the points different from the attributes of the similar object without newly inputting all the attribute data is used to input the information. Measure ease. For example, consider a case where a person named "Taro Nakayama" has already been registered and his brother "Jiro Nakayama" is registered. In this case, if "Taro Nakayama" is selected as a similar concept, the system displays the attribute list. For example, ((Father is Motoo Nakayama) (Mother is Hanako Nakayama) (Date of birth is May 4, 1960) (Gender is male) (Hobby is music) (1) is displayed. Of which, only the different ones,
For example (birth date is June 7, 1963), (hobby is sports).

(2) Search condition input support function When end users search, most of them have only a vague image, and it is difficult to express all of them properly in natural language or the like.

In the present invention, interactively, starting from the most important concept, information for searching is sequentially added. At this time, similar to the registration support function, knowledge of the filing target world is stored in the storage device, and based on this, the user is presented with the attribute names that can be entered and the concepts (classes of things) that can be entered. To do.

For example, assuming that the user wants a "paper", the user first inputs "paper". The storage device knows from the "thesis" that it has attributes such as author, title, and subject, and sets of attribute names and concepts on the terminal, such as (author, person's name), (title, text),
Present (theme, concept). The user sees them, selects what he has as information, and inputs data. For example, select "subject" and enter "computer". This process can be repeated recursively. For example, if the subject is "computer," the system presents (developer, organization, company), (running language, computer language), (OS, OS), etc., and the user, for example, (running language, LISP ), You can add conditions.

With such a support function, for example, "LISP
You can set a search condition such as "A paper written by an employee of Company A" (2) for a paper on a computer that runs. As will be described later in detail, the same search condition is formally represented by the following expression in the method of the present invention.

(TECHNICAL-PAPER (SUBJECT-IS (COMPUTER (RUNS LISP))) (AUTHOR-IS (EMPLOYEE (WORKS-AT A-COMPANY)))) (3) This can be expressed in Japanese as follows. Write in.

(Paper (the subject is (computer (running language is Lisp))) (author is (employee (affiliation is company A)))) (3) The above notation is an S-expression (symbolic e in the Lisp language.
xpression) [Reference: P. Etch. Winston "Lisp" Addition-Vesley Publishing Company 1981, P18 (PH Winston, "LIS
P, "Addison-Wesley Publishing Co. , 1981, to comply with the written side of the p.18)].

(3) Semantic ambiguous search function When a user wants to search for a certain property, it usually has only fragmentary information and is ambiguous. On the other hand, it is stored in a storage device (database or the like) with a very specific name. Therefore, it is required to close these gaps.

In this case, there are the following five types of ambiguity, roughly classified.

(I) Incompleteness of Name Only part of the name of an object or concept is remembered.

(Ii) Synonyms Often, the same thing is remembered by different words. For example, “artificial intelligence, artificial brain, AI” and the like refer to the same concept.

(Iii) Incompleteness of numbers It is rare to remember numbers exactly.

Examples thereof include "around 1980s, about 85, between 83 and 87, and before 1960".

(Iv) Conceptual Abstraction (Part 1) Objects and concepts are often remembered by their higher-level concepts, and specific ones are often forgotten. Here, the higher-level concept is often based on classification (clustering). "I forgot the company name, but the organization was not a university or research institute, but a company anyway." Or "that was an electronics manufacturer."
Is an example.

In this case, the manufacturer of the electric machine is, for example, "A
"BC Co., Ltd.", the relational expression "(ABC Co., Ltd. IS-A Electric Maker) (Electric Maker, IS-A Company)" holds. Schematically, the concept "ABC Co., Ltd." and the concept "Electric Maker" are " It will be connected by a link called "IS-A". Here, the link "IS-
"A" is a relation defined between the above two concepts.
Ship) is called inclusive relation.

It is generally considered that all concepts are classified hierarchically by IS-A links. The hierarchical tree will be called a concept tree.

(V) Conceptual Abstraction (Part 2) The abstraction in the previous section is a set-theoretic abstraction of the concept, but humans often memorize it in the upper part of the conceptual partial relation. For example, “I forgot which factory A belongs to, but it was ABC Corporation anyway” or “I forgot how many cities I had in Saitama prefecture”.

[0038] usually, in the database that stores the fact Toka "Mr. A is an affiliation of XYZ factory", "ABC Co., Ltd. is located in Tokorozawa" are Toka memory, the above-mentioned conventional information retrieval technology You cannot search from the ambiguous information in the example.

In this case, the relationship of (HAS-PART-OF XYZ factory of ABC Co., Ltd.) (HAS-PART-OF Tokorozawa city, Saitama prefecture) plays an important role. What should be noted here is not (Tokorozawa City IS-A Saitama Prefecture) but (Tokorozawa City IS-PART-OF Saitama Prefecture), which should be clearly distinguished from conceptual inclusive relations (upper and lower relations). Is. Relationship "IS-PART-OF"
Is the inside-out relationship of the relationship “HAS-PART-OF”.

Strictly speaking, a directional relationship (IS-P
ART-OF and HAS-PART-OF) are called relations, and a relationship in which they have no directionality (existence as a link) is called a relationship.
In the above example, IS-PART-OF and HAS-PART-
OF is "Part-whole relationship"
It is called.

In addition to the above five types of ambiguity, the characteristics of human memory are that relationships between concepts are more likely to remain in memory than the concepts themselves. For example,
The fact that it was "developed" was important in a search from fragment information that "the subject of an article was an operating system, but it was developed in a US laboratory." It is a "relationship" defined between the two concepts of place. Specifically, it is represented by a relational expression of (UX operating system IS-DEVELOPED-BY B Laboratory), and IS-DEVELOPED-
BY is relation. This "relationship" is important in the semantic fuzzy search.

Among the characteristics of human memory described above, incompleteness of names and incompleteness of numbers are taken into consideration in the conventional information retrieval. For example, a matching function of a partial character string, a numerical range specification, etc. are known.

The semantic ambiguous search function, which is a feature of the present invention, is novel in the above classification, particularly in the handling of abstraction concepts (No. 1 and No. 2). That is, the following semantic fuzzy search is possible by using the search condition input support function described above.

Search condition: "Article about a computer developed by a company in Saitama prefecture and running an operating system developed by a certain research institute" (4) Here, the concrete concept appearing in the conditional sentence is Only "Saitama Prefecture" is available, and other words that are likely to be keywords are abstract "computer", "laboratory", "operating system", etc. Conventional information retrieval method,
For example, keyword search cannot obtain sufficient search information.

Conditional statement (4) is a "semantic search" in the sense that the relationship between Saitama prefecture and company, company and computer, operating system and computer, etc. is also treated as search information. A company, a computer, an operating system, etc. have realized an "ambiguous search" in the sense that they are generic names (abstract concepts). Since the conventional search method does not describe the relationship between keywords, for example, "articles on computers installed by a research institute in Saitama prefecture running an operating system" are also included at the same time. In that sense, it is a “non-semantic search method”.

[0046]

EXAMPLES The present invention will be described below based on examples.

FIG. 1 is a block diagram of an image information filing apparatus adopting an information storage system which is an embodiment of the system of the present invention. An outline of the configuration and operation of the entire device will be described.

First, the device is a control device (CPU) 100,
Main memory 300, magnetic disk device group 400, operating terminal 200 (CRT 210, keyboard 220, mouse 23
0), and an image information processing unit including an image scanner 700, an image printer 750, an optical disk device 450, an image buffer memory 350, a high-speed image processing circuit (IP) 600, and a high-definition image display device 500. Become. Also, these are connected via a bus adapter 805.

The main operations are registration of image information from a document or the like, retrieval / display / output of desired information, and input / edit of knowledge about a field to be filing. When registering a document image, the document image is read from the scanner 700 into the memory 350, compression-encoded by the IP 600, and then stored in the optical disc 450. At that time, the image in the memory is displayed on the CRT 500 to check whether or not the document was correctly read, and the bibliographical items (title, author, source, etc.) of the document and the position in the field knowledge are input from the terminal 250. To do. Information on bibliographic items, physical addresses (back address, track address, sector address) of the document image on the optical disc and image attributes (size, scanning density, encoding method, etc.) are provided in the file 4.
It stores in 20. Information such as the position of the document in the field knowledge is stored in the file 430.

In the search / display, a desired document is fixed on the terminal 250 by a dialog for searching, which will be described later, and is displayed on the image display CRT 500. If a hard copy is desired, it is output from the printer 750. The location information (physical address of the optical disk, etc.) of the fixed document is read from the file 420, and is sent to the optical disk control circuit 455 as a read control command of the optical disk via the bus adapter 805. The read image data is stored in the shared memory 350, and the image of the displayed page is sequentially decoded by the IP 600.

The mouse 230 is a CRT 210 and a CRT.
It is possible to specify the position for both 500, and the display position of the image on the CRT 500 can be specified by the mouse. This function can be used to convert documents on multiple pages to CRT5.
It is also possible to superimpose and display it on any position on 00. In order to display a large number of pages on the same screen, a document image for one page can be reduced and displayed by IP. CR
The CPU 100 manages the screen on T.

The field knowledge is input / edited on the terminal 200 while displaying the document on the CRT 500 as necessary. Field knowledge is a set of facts described as concepts about the world that are registered or that are described in documents registered and relationships between the concepts. In addition, it also includes knowledge about natural language that expresses concepts and relationships. Naturally, the document itself is included in the above world as one of the concepts. These pieces of knowledge are stored in the file 430.

The above-mentioned three main functions are modeless and can be arbitrarily called at any time. For example, while adding / editing field knowledge, necessary information is called up on the CRT 500 by using the search function, or knowledge about the same document is added while registering a document image. You can do it.

Next, the field knowledge data expression format will be described. Knowledge is expressed by two types of elements: concepts that include things and relationships between them. FIG. 2 is a diagram conceptually expressed in the form of a Semantic Network. Nodes represented by ellipses represent concepts, and the words written therein are representative words representing the concepts. It is called a concept name. Links between ellipses (arrows represented by solid lines or broken lines) represent "relationships" between concepts. For example, if the super computer 1012 is a kind of the computer 1011, the IS-
It is represented by the link A. UNI here
VERSAL 1010 is a special concept and is defined as one that encompasses everything. Therefore, all concepts are UNI
A concept tree with VERSAL as the root
e), which represents a taxonomic hierarchy. IS-A is a kind of "relationship", but it is a way to pass down the attribute of a concept to a subordinate concept, and in that sense, consider it separately from other "relationships". Therefore, FIG.
Also in, the arrows are shown separately as a solid line and a broken line.

For example, there is a general attribute that "computer runs software". At the same time, it is the attribute that "software runs on a computer" when turned over. Here, these are referred to as "general relationships" (gene
ric relationship). Formally expressed as (Computer RUNS software) (Software RUNS-ON computer) (5). These general relationships are inherited by subordinate concepts as follows. That is, it can be said that "a super computer runs software", "X-800 runs software", or "an operating system runs on a computer", and "UX runs on a computer". These relationships are derived from the general relationship (5) and are not directly described in the knowledge base.

In FIG. 2, the concepts "X-800" and "U
Unlike the above general relation, the relation 1005 connecting “X” is a unique relation defined between these two concepts, and is referred to as “instance relation” (simply relation). However, the concrete relationship 1005 is an embodiment of the general relationship 1004.

In this way, an article "ART # 01
8 ”1018 is the subject of supercomputer X-800
The fact that the operating system UX runs on the same super computer is represented by the diagram shown in FIG. All concepts are connected by IS-A link warp threads and by general and concrete relationship weft threads to form a concept network as a whole.

It should be noted here that not only the attributes associated with a concept are viewed from the side of the concept, but also the perspective of the concept appearing in the attribute is given equally. This is done by changing the view of "attribute" to take the view of "relationship" defined between two concepts, and by making the relationship visible from both sides. In this way, for example, article AR
When T # 018 and its subject X-800 are registered, it is also registered that "X-800 is on ART # 018" at the same time.

The above knowledge expression can be more formally expressed as shown in FIG. The system stores and manages these in the form of four types of data: concept C, subsumption relation S, general relation R, and concrete relation r. C (i) is the i-th concept, S (k, i) is that C (k) is a kind of C (i), R (m) is the m-th general relation, and r (m,
k, l) is a concrete relation of R (m), that is, a concrete relation, which is defined between C (k) and C (l).

More specifically, these four types of data are represented in a tabular format, and the entity relation model (Entity Relation Model) is used.
It can be expressed as shown in Fig. 4 when it is represented by the diagram of del: ER model). Here, the concept C and the general relationship R are regarded as entities,
The inclusive relation S and the concrete relation r are regarded as a relation connecting these entities. The general relation R defines the class of the concrete relation r, and is considered as an entity here. These specific tabular formats are shown in FIGS. The contents of the table represent FIG.

FIG. 5 shows a table (ta
ble), and is basically a concept unique number C #, a word (word; or notation) CNAME that represents it, and the same notation CNAME is a primary notation (Primary) or a secondary notation (Secondary). And a column that represents For example, concept # 58 is "computer,"
"Calculator" is also given as a notation.

FIG. 6 shows the subsumption relation S, which is composed of a concept unique number C # and a column of the unique number S # of the superordinate concept (Superclass) of the concept. As a special case, "UN
The upper concept of "IVERSAL" is "UNIVERSAL"
Is defined as

FIG. 7 is a table defining the general relation R. It is composed of a general relationship specific signal RS #, a name RSNAME of the relationship, a reading LR when the relationship is read from left to right, and a reading RL when the relationship is read from right to left. However, "from left to right" here refers to a basic form in which two concepts connected by the same relationship are arranged side by side, and the basic form is defined in the table of the specific relationship r described in FIG. In Table R, for example, 16
There is a relationship of "SUBJECT" as a general relationship of the number, which means that it is used as a specific relationship of IS-SUBJECT-OF or SUB JECT-IS.

It should be noted here that the same reading relations are defined as distinct and different. For example, in FIG. 7, there are two HAS-PART-OFs, but these are related PART-WHOLE1 and PART.
-For a different relationship of WHOLE2.
This meaning means "part-wholerelationsh".
ip) also treats different concepts separately. For example, PART-WHOLE1 is applied to the case of "organization" (ABC Co., Ltd. HAS-PART-OF XYZ factory (6), PART-WHOLE2 is applied to "region", and (Saitama HAS-PART- OF Tokorozawa) (7) Used to express the relationship.

FIG. 8 is a table showing the relation r defined between concepts. Both general relations and concrete relations are registered in the table. Column G / I is to distinguish them,
G indicates a general relationship, and I indicates a concrete relationship.

In the table r, each record has a relationship unique number R #, a general relationship unique number RS # indicating the type of relationship, a concept unique number CL located on the left side in the basic form, and a concept unique number CR also located on the right side. Has an element.

For example, the relation # 4 is a general relation (G / I = G) defined in the concept # 58 (“computer”) and the concept # 64 (“software”), and the general relation #
It has the meaning defined in 7. More specifically, (computer RUNS software) (software RUNS-ON computer) (8) has two superficial relationships, and one record in table r. Similarly, relationship # 724 is an instantiation of general relationship # 7, with concepts # 1512 and #
It is tied to 1051. That is, as can be seen from Table C, the relationship (X-800 RUNS UX) (UX RUNS-ON X-800) (9) is expressed.

The four data described above (Tables C, S,
R, r) is stored in the file 430 as a knowledge base.

Next, the table D that defines the document image will be described with reference to FIG. Table D shows a document unique number D #, a concept unique number C #, a document size SIZE, a compression coding method CODE,
Image scanning density DENS, physical address P on optical disc
It has columns of HYSA and recording sector number LENG. The record shown in FIG. 9 relates to Concept # 313 (Article ART # 018; see FIG. 5), Document # 98 is A4 size, is scanned with a scanning density of 16 lines / mm, and is compressed with the MH code. 40 of optical disc
It indicates that the data is stored in 13 sectors from address 0207.

As can be seen from the above, it is possible to add a document image to all the concepts, so that the description can be attached to a "computer", for example, as an image, even if it is not necessarily a "document". it can. At the same time, it is possible to attach a plurality of different documents to one concept.

The table D is stored and managed in the file 420 shown in FIG.

[0072] above, the device configuration of the present invention embodiment, since has been described in the data representation format, presenting a software configuration and processing scheme below. FIG. 10 shows a software configuration diagram. Here, the software is processed by the control device 100 (FIG. 1), and the program is the file 41.
It is stored in 0.

As shown in FIG. 10, the software includes a dialogue control module 2001, a search / concept matching module 2002, a search expression generation module 2003, a concept network editor 2004, a network negotiation module 2005, a table operation module 2006, and an image display module 2010. Made of.

The function of the table operation module, which is the basic part, will be described first. The purpose of the table manipulating module is to provide a function for manipulating and searching the tabular data described above at a high level. The main functions are explained below. The program is written in the LISP language, and the S-expression is used below. In order to make it easier to understand,
Variables and functions are written in lower case, and constants are written in upper case.

First, as a function for creating a new table, crea
There is a te-table. For example, in the table R of FIG. 7, the framework is newly defined by calling the same function as follows.

(Create-table'R '(RS # RSNAME LARL)) (10) Here, the first argument is a table name, and the second argument is a list (set) of column names. The symbol '(quotation mark) means that the next symbol is not a variable but a constant.

Next, to add a record to the table, insert
There is a function.

(Insert 'R' (RS # RSNAME LR RL '((7 RUN RUNS RUNS-ON) (15 DEVELOPMENT HAS-DEVELOPED IS-DEVELOPED-BY) (16 SUBJECT IS-SUBJECT-OF SUBJECT-IS))) By setting (11), the three records in table R of FIG. 7 will be added.

The update function is used to update the record. For example, by setting (update 'R' RSNAME "THEME '(eq (v RS #) 16)) (12), the value of column RS # in table R is 1
The value of column RSNAME of the record that is 6 is'THE
It can be ME. The record is deleted by the following formula.

(Delete'R '(eq (v RS #) 16)) (13) The second argument is a conditional term as in the expression (12), and it is possible to write an arbitrary proposition function for both up date and delete. it can.

A record is selected (that is, retrieved) by selecting
Use a function.

(Select '(LR RL' R '(Smatch' * SUBJ (v RSNAME))) (14) Example In the expression (14), the substring * SUBJ and the column RSN of table R are used.
A record that partially matches AME is selected, and as a result, a list of columns LR and values of LR of the same record is returned.
Here smatch is a propositional function that performs substring matching.

The main functions of the table operation module have been described above.

Next, the network negotiation function and a method of implementing the main functions will be described.

The concept network is as described in FIG.
It consists of a concept node and a link called "relationship".
The system stores the concept that is the focus of the user's interest as the current node. The user can roam around the network while using the functions described below. Walking around here is equivalent to moving the current node.

The main ones of the network negotiation function will be described.

[0087] KarenToThe first way to move a node
There is a way to put the concept name or its substring in the eye.
It Here, if the concept of matching substrings
If there is a number, display the menu table and let the user
You can select one of the. Karen to the result
Node is moved. Figure 11 shows the substring * DATABASE
The matching concept is displayed and number 7 in the menu table
The example which selected the eye is shown. Where the * symbol is
Indicates a substring. The above function selects the concept by (select '(C # CNAME)' C '(smatch str (v CNAME))) (15)
It is realized by calling a function (select-one a-list)
It Here, str is an example of characters read from the terminal.

The second means of negotiation is the concept tree (hierarchical tree).
Is a method of sequentially tracing by selecting a menu. Figure 1
Reference numeral 2 indicates a concept directly below the top concept as a menu displayed by the l command. After that, as shown in FIG. 13, it is possible to go down the hierarchical tree by inputting a number. In the example of FIG. 13, “6, 1, 1, 2”
By entering, you have reached the concept of "Organization". At this point, if the previous partial character string is input, the concept that matches the partial character string is selected only from the concept set included under the current node. For example, of the 11 concepts shown in Fig. 11, the 11th "RELATIONAL-DATABASE-SYSTEM-IN"
"C" is the company name, and only this will be selected. That is, only this is included in the concept of "organization".

Here, the concept encompassed by a certain concept C is obtained as follows. First, (Select '(C #)' S '(eq (v S #) c)) (16) selects a concept that is directly included in the concept (located directly below in the concept tree). Therefore, by recursively calling the same function, all the concepts included by the concept c can be selected.

Next, the third means for negotiating is a method of graphically displaying the concept tree. FIG. 14 is a conceptual tree starting from the concept of "space" and ending with the lower two layers. By pointing an arbitrary node of the same concept tree displayed on the CRT 210 in FIG. 10 with the mouse 230, the current node is moved to the designated concept, or a lower-level concept tree starting from the designated concept is displayed. You can You can specify it with the mouse here. The dialogue control unit 2001 shown in FIG. 10 receives the position information on the CRT designated by the mouse. Since the control unit controls the graphic display, the CRT
What is displayed at each position is stored, and by inputting position information, it is possible to know which concept is displayed at the closest point at the same point. Therefore, the control unit 2001 inputs the cursor position information and returns the concept name to the network negotiation unit 2005.

The graphical negotiation function is not limited to the subsumption relation of the concept shown in FIG.
relation). Figure 15 shows the concept "Earth (e
"arth)" is displayed hierarchically as a concept tree in another sense. Exactly the same as the above example, it is possible to walk around along the partial whole relationship. The partial relation can be extracted as follows. First, (select '(RS #)' R '(eq (v LR)' HAS-PART-OF)) (17) is the general relation number R that shows the partial whole relation from Table R in Fig. 7.
Returns the set of S #. In the example of FIG. 7, (2223) is returned.
This is temporarily stored in the variable x.

Now, letting c be the concept of finding a part, the concept of “part” of c can be obtained by the following equation.

(Select '(CR)' r '(and (eq (v CL) c) (member (v RS #) x))) (18) where r is a lowercase letter, but the name of the table in FIG. Is.

Similarly, also in the case where the HAS-PART-OF is in the column PL in Table R of FIG. 7, by exchanging LR and RL in the expressions (17) and (18) and executing them, the concept c A partial concept is obtained.

The fourth means of negotiation is chained negotiation through the frame representation of the concept. An example will be described with reference to FIG. Enter the substring * so w a of the person's name and only enter "JF S
"OWA" is found and the current node moves to the same concept. A frame of the same concept is displayed by the fr command for displaying a frame (Operant * means displaying the frame of the current node). From the same frame, "JF SOWA" is "Man (MAN)" and PAPER #
I know that I am the author of 0012 and BOOK # 0007. Here, for example, when it is desired to know what the PAPER # 0012 is, by inputting the command "fr2", the conceptual frame on the second line of the displayed frame is displayed. In this example, PAPER # 001
The second is "Thesis (TECH-PAPER)" and the author is "JF SOW".
"A", pages 336 to 357, "IBM-R
You can see that it is listed in "ES &DEV-76-20". In the example of FIG. 16, JF SO
WA → PAPER # 0012 → IBM-RES & DEV-76-
The current node is moved from 20 to IBM-CORP.

The frame display is performed as described above. Now, when displaying the frame of concept c, (select '(RS # CR)' r '(eq (v CL) c)) (19) is the "relationship" where c is on the left in the basic form of the relationship. And a set of pairs of concepts on the right side at that time is obtained. Similarly, (select '(RS # CL)' r '(eq (c CR) c)) (20) gives a set of pairs of "relationships" where c is on the right and the left concept at that time. . Let these values (sets) be x and y, respectively (select '(LR)' R '(eq (v RS #) x _1i )) (21) (select' (RL) 'R' (eq (v RS #) y _1i )) (22) returns the name of each concrete relationship as a result.

However, x _i and y _i are i-th elements of the sets x and y, x _1i is the first element of x _i , and y _1i is the first element of y _i . The name of this concrete relationship corresponds to the first column of the frame (FIG. 16). That is, it is a slot name when each line of the frame is called a slot. The second column of the frame is the slot value, each pair x _i ,
The second element x _2i of y _i, a y _2i. Therefore, if you create a pair of slot name and slot value and display it in tabular form,
It becomes like 6. Here, since the concept is internally handled by the concept unique number, when displaying, the table C (FIG. 5) is converted into the concept name and then displayed.

In the above example, the line number (slot number) is specified to jump to the next frame, but the concept that one wants to jump to can be specified directly on the CRT by the mouse, and the frames can be overlapped one after another. It can also be displayed.

Next, the conceptual network editor 2004 in FIG. 10 will be described.

The concept network editor is for maintaining the knowledge base, and is for defining, adding, changing and deleting new concepts and relationships.

FIG. 17 is an example of additional dialogue for a new concept. First, by typing man, the current node is moved to the concept MAN "male" and a new person "superman" is registered. "MR. SUPERMAN" can be registered in the lower order of MAN by the command crc, that is, as (MR.SUPERMAN IS-A MAN) (23). Next, the command crr defines and registers the attribute of the newly registered concept MR. SUPERMAN in the form of a concrete relationship.

The system has "knowledge" in the form of general relations, such as "human beings are the ones who come up with new things,""has a degree," and "have a job. "
It has items that require you to enter those knowledge as clues and prompts. As shown in FIG. 17, M
R.SUPERMAN HAS-TITLE-OF (ACADEMIC-TITLE): Encourages the input of MR. SUPERMAN degree. The concept ACADEMIC-TITLE indicated by {} indicates a class of attributes (here, one concept) to be input. The concept network negotiation function described above is valid at this point, at which point the current node has been moved to the concept ACADEMIC-TITLE. Therefore, any snooping function can be used to find the attributes to enter. In the example of FIG. 17,
Although the concept name "phd" is directly input, in order to search for the same concept, the command l can display the subordinate concepts of ACADEMIC-TITLE as a menu (see Fig. 13) or the concept tree. (See FIG. 14) You may find the concept you want. However, exploration and crosstalk are limited in this case to the scope of subordinate concepts of the concept ACADEMIC-TITLE.
After moving the current node to the concept representing the attribute desired to be input by the negotiation function, by inputting "ok", the next attribute is input.

As described above, it is possible to newly define and register while selecting only the attributes to be input according to the prompt displayed by the system.

One of the features of the method of the present invention is that both concepts are treated equally by capturing the attributes in the form of concrete relationships, as described above.
Speaking of the example above, (RED IS-GIVEN-TO MR.SUPERMAN) (AAAI HAS-MEMBER-OF MR.SUPERMAN) (JAPAN HAS-PEOPLE-OF MR.SUPERMAN ) ( HITACHI-LTD HAS-EMPLOYEE- OF MR.SUPERMAN) will be defined at the same time.

The new registration of the concept can be realized internally as follows. Now, set the maximum C # (see FIG. 5) to C _max
Then, first, the concept MR.SUPERMAN is registered in Table C by (insert'C '(C # CNAME P / S) (list (add1 C _max )'MR.SUPERMAN'P)) (24). Here, the required number add1 is a function that increments by one.

[0106] in Table S by (insert'S '(C # S #) (list (add1 C max) (get-C #' MAN))) (25), can be registered the fact of (23). Here, the function get-C # is a function for obtaining the unique number of the concept from the concept name.

The specific relationship can be registered for the name relationship (attribute) as follows.

(Insert'r '(R # RS # CL CR G / I) (list (add1 r # _max ) rs (get-C #' MR.SUPERMAN) (get-C # 'PHD)' I)) Here, it is assumed that rs is the unique number of the general relation "ACADEMIC-TITLE", and r # _max is the maximum R # at that point in the table r.

Another function of the Conceptual Network Editor is
It is a change or modification of a fact or concept that has already been registered. Specifically, the concept name can be modified, the position in the concept tree can be modified (classification modified), the concrete relationship can be modified, and the general relationship can be modified.

FIG. 18 shows a modification example of the concept tree. Concept PSYC
This is an example of moving the HOLOGY position from SOCIAL-SCIENCE to the source of NATURAL-SCIENCE.

By executing (update'S'S # '(get-C #' NATURAL-SCIENCE) '(eq (v C #) (get-C #' PSYCHOLOGY))) (27), the above example can be moved. Done.

Although it has been described that the concept network editor can use the network negotiation function at any time, similarly, the network editor function itself can be used recursively. For example, while registering a new concept definition,
This is because it is often necessary to newly register another concept. In the example of FIG. 17, “Academic Society” (AC
If "AAAI" to be entered is not registered in ADEMIC-ASSOCIATION), a command AAAC can be entered at the same point to newly register the concept AAAI. If necessary, the attribute of AAAI may be registered at that time by the command crr.

Alternatively, an erroneous registration of “fact” or the like may be found during new registration of a concept, and in this case also, it has a function capable of immediately executing correction / change.
Due to these recursion properties, new knowledge can be efficiently added and registered.

Next, the search expression generation unit 200 in FIG.
3 and the search / concept matching unit 2002 will be described.
These parts are the main parts that enable semantic content retrieval from fragmentary fuzzy information. First, in order to explain the function, the flow of processing will be described by using a search example.

"Article on computers running UNIX (registered trademark) on computers developed by a company in California" (2
We consider that the search for 8).

As shown in FIG. 19, first, the current node is moved to the concept ARTICLE (article), and the command q is input to start the search expression generation. The system obtains the general relationships defined in the concept ARTICLE, including those inherited from the superordinate concept, and displays it in frame format as shown in FIG. "Article is part of a publication", "How many pages does an article have", "An article is referenced by an article"
A frame made from general knowledge such as is displayed. In this case, the user can specify the slot by the number and add more specific information than the currently displayed slot. In the example of FIG.
Since the subject of the article is a computer as information, we select the fifth slot. The system moves the current node to the superordinate concept of the item to be input, in this case UNIVERSAL. The user can now use any network negotiation function (see Figure 10). In the example of FIG. 19, "Computer" is directly input to move the current node to the concept COMPUTER. However, if the concept name to be input cannot be thought of, it is possible to select it by inputting a partial character string or displaying a menu. In this example, since there is more information regarding the computer, the command q is further input after "Computer".

FIG. 20 shows the next dialogue screen. Concept COMPUT
A frame representation of the general relationship for ER is displayed.
As before, the condition slot to be added, in this case 4
Select No. The class of items to be entered is "ORGANIZA
"TION. WORK PLACE" is displayed. Menu selection (see Fig. 13) can be used here as well, but in this example, "Co-company" means "Co
You have entered "mpany". Since there is information to be added to "Company", "q" is continuously input to continue the search expression generation process.

FIG. 21 is a frame representation of the general relationship to the concept COMPANY. I selected the 7th slot and entered the condition that the company is in California. In this case, enter the partial character string "Cal" and enter "Ca
lifornia ”. Concept SPACE. PLACE. LOCA
The subordinate concept of TION is shown in FIG.

The end of the condition input is notified to the system by inputting "ok". FIG. 22 shows a situation in which the condition input for COMPANY is completed and the condition input for COMPUTER is performed again. This time, I chose the 13th slot and said, "The computer is an operating system.
Run under UNIX ".

When all the above steps are completed, (28)
The expression search condition will be replaced by the formal expression of the following expression.

(ARTICLE (SUBJECT-IS (COMPUTER (RUNS-UNDER UNIX) (IS-DEVELOPED-AT (COMPANY (IS-LOCATED-IN CALIFORNIA)))))) (29) It is conducted by being guided by the general knowledge described as a general relationship as described above. Much of this general knowledge is inherited from superordinate concepts. For example, in FIG. 21, (COMPANY IS-A ORGANIZATION.WORKPLACE), but most slots (general relations) of the same frame are defined for the superordinate concept ORGANIZATION.WORKPLACE. Only the (PRODUCES MACHINERY. DEVICE) is unique to COMPANY in the example in the figure. In this way, the general relationship as general knowledge is literally, the more general (in a higher-level concept) the more powerful it is. That is, it can be applied to many subordinate concepts.

The general relation of the concept c can be obtained as follows. Now, let x be a set of superordinate concepts of the concept c.

(It is sufficient to follow the IS-A link to UNIVERSAL in Table S.) (Select '(RS #CR)' r '(and (member (v CL) x) (eq (v G / (I) 'G))) (30) extracts the general relationship inherited by the concept c, where the concept c is defined on the left side. In equation (30),
By substituting CR → CL and CL → CR, the general relationship in which the concept c is defined on the right side is obtained. As can be seen from the equation (30), what is obtained is a RS # and CL or CR pair. By referring to Table R from RS #,
The slot name of the frame can be obtained. Now
The answer to the equation (30) is ((y _{1 1} , y _{2 1} ) ... (y _{1 i} ,
y _2i ) ...), the slot name corresponding to the i-th y can be obtained from (select '(LR)' R '(eq (v RS #) y _1i )) (31). it can. With respect to the expression (30) in which CR and CL are replaced, in the expression (31), LR may be replaced with RL (see FIG. 7).

The answer to equation (31) is ((z ₁ ) ... (z _i ).
...), the pair (z _i y _2i ) consists of a slot name and a concept unique number.

(Select '(CNAME)' C '(and (eq (v C #) y _2i ) (eq (v P / S)' P))) (32) returns the conceptual name C ₁ . After all, the pair (Z _i C _i ) constitutes the slot name and the slot value. By displaying the list ((Z ₁ C ₁ ) (Z ₂ C ₂ ) ... (Z _i C _i ) ...) in a tabular form, a frame of general relation can be obtained as shown in FIG.

When a search command is input to the automatically generated search condition expression (eg, expression (29)), the terminal node of the current node (the concept corresponding to the "leaf" of the partial concept tree). Concept matching is performed on the set of. That is, each terminal concept (most specific concept) is compared with the "abstract concept" expressed by the search condition expression, and the conceptually included terminal concept is the search result.

FIG. 23 shows a case where a search is performed for the search condition 3000 of the expression (29). The conceptual frame display 3001 of the article that matches the condition and the image 3 of the text.
051 can be displayed. The conceptual frame of the symbol ARTICLE # 0014 obtained as shown in the figure does not seem to match the search condition formula (29) at all on the surface. This is exactly what happens with conventional information retrieval methods such as databases. That is, since the database stores the most specific information, if the search conditions are abstract, they will not match at all. However, in the method of the present invention, the gap between the concrete contents of the conceptual frame and the abstract expression of the search condition is filled by inference using the knowledge about the field.

Next, the matching method will be described by using the inference between the abstract concept for filling the gap and the concrete concept. Here, this process is referred to as concept matching (conc
eptmatching). FIG. 24 shows a process of concept matching between the abstract concept expressed by the equation (29) and the concept ARTICLE # 0014.

In the search process, the terminal concept ARTICLE # 0001-ART
Each of ICLE # 0040 is compared with the equation (29), but here, only ARTICLE # 0014 for which the matching is successful will be described.

First, (ARTICLE) and (ARTICLE # 0014) are compared. From frame 3001, it turns out that (ARTICLE # 0014 IS-A NEWS. ARTICLE) and (NEWS. ARTICLE IS-A ARTICLE), and the conclusion (ARTICLE # 0014 IS-A ARTICLE) is obtained. Therefore, the conditional expression is made deeper and it is next checked whether (ARTICLE (SUBJECT-IS COMPUTER)) (33) holds. From the frame 3001, first, it can be seen that it is (ARTICLE # 0014 (SUBJECT-IS HP-9000)). Therefore, you can check whether it is (HP-9000 IS-A COMPUTER). Following the link 3003 (HP-9000 IS-A SUPERMINICOMPUTER) and following the link 3032 (SUPERMINICOMPUTER IS-A COMPUTER), it can be seen from the frames 3002 and 3004 that the expression (33) is successful. Therefore, next, the establishment of (ARTICLE (SUBJECT-IS (COMPUTER (RUNS-UNDER UNIX) (IS-DEVELOPED-AT COMPANY)))) (34) is examined. As for (COMPUTER RUNS-UNDER UNIX), it is directly understood that it is established from the frame 3002.
Regarding (COMPUTER IS-DEVELOPED-AT COMPANY), the frame 3002 says (HP-9000 IS-DEVELOPED-AT HP.HEWLETT-PACKARD-CO), so further investigation is necessary.
Following the link 3033, since it is written in the frame 3004 as (HP.HEWLETT-PACKARD-CO IS-A COMPANY), it is established. Therefore, the equation (34) is also established.

The search condition expression is further deepened by one step and compared with expression (29). The frame 3004 is written as (HP.HEWLETT-PACKARD-CO IS-LOCATED-IN PALO-ALTO) (35), so it is necessary to check whether it is (COMPANY IS-LOCATED-IN CALIFORNIA) (36). There is. Following link 3034, it is recorded in frame 3005 as (PALO-ALTO IS-PART-OF CALIFORNIA) (37), so in addition to formula (35), (HP.HEWLETT-PACKARD-CO IS-LOCATED- IN CALIFORNIA) is inferred. After all, it turns out that the abstract concept (29) includes the concrete concept ARTICLE # 0014.

The concept matching has been described above by using a concrete example. When this process is expressed recursively, a process of checking the success or failure of the subsumption relation of the concept and a process of checking whether or not the respective slots match each other. Are called recursively alternately. Also, as you can see from the explanation, it is a backward-chaining inference. Therefore, the search time is basically proportional to the number of concrete properties to be searched.

In concept matching, a concrete concept is returned as a value when the matching is successful. By embedding this value in the abstract concept, it will be understood later why a specific concept matched. In the method of the present invention, when the command why is input, the matching reason of the retrieved concept is displayed and output. In the example of the expression (29), (ARTICLE # 0014 (SUBJECT-IS (HP-9000 (RUNS-UNDER UNIX) (IS-DEVELOPED-AT (HP.HEWLETT-PACKARD-CO (IS-LOCATED-IN PALO-ALTO) ))))) (38) is obtained. Therefore, the computer described abstractly at the time of searching is HP-9000, and a company is HP.HEWLETT-P.
You will know that it is ACKARD-CO.

The search method of the present invention also provides a means for converting the knowledge expressed in the concept network into a table format that is usually familiar to the user, displaying the knowledge, and searching the table. FIG. 25 shows a specific example. When the command tab is input after moving the current node to the abstract concept NEWS.ARTICLE, the system internally generates the concept frame of the lower-level end concept of the same concept. Next, the slot names appearing in the same frame are presented as a table. In the case of FIG. 25, there are six types of slots. The user can select which slot is to be the table with columns.
In the example, the sixth and fifth slots, ie, TITLE-
Select IS and SUBJECT-IS to display a table with three columns in total. The user can, for example, find what he or she wants from the title text. When a command for displaying a text image is input using the selected result as the current node, the text image is displayed on the image CRT 500.

The search method according to the present invention can be applied not only to document filing but also to general data and factual information. Concept matching also provides a method for automatically testing the inclusive relation of two concepts,
It can be applied not only as information retrieval but also as information analysis technology such as examination of inclusive relations of patent claims.
Alternatively, conceptual cluster analysis becomes possible by collecting a large number of concrete concepts and applying the same concept matching.

As an extension of the present invention, the concept of time can be introduced into the knowledge base. In general, facts and the like often need to specify times when they are "true." For example, the fact that (MR.SUPERMAN IS-MEMBER-OF AAAI) is "true" from the time of admission to the time of withdrawal. Such a concept of time can be realized by expanding the table r holding the concrete relationship.

Yet another extension is the ability to record the origin of each fact. For example, the source of information as to who said it and where it was written can be recorded by expanding the table r to manage the information. Furthermore, it is also possible to add a degree of reliability.

The present invention does not depend on how the system is implemented. For example, special hardware may be introduced to speed up the process. Further, when the knowledge base becomes large, a variation in implementation can be considered, for example, a method of dividing a table storing concepts and relationships into clusters and storing them, which are included in the method of the present invention. Further, when a concept or the like is written in Japanese, a kana-kanji conversion function peculiar to Japanese is required, but the present invention includes those expanded.

[0139]

As described above, according to the present invention,
It becomes possible to easily extract information such as a desired document from the fragmentary information without knowing the specific construction method of the file or the database. Furthermore, it is also extremely easy to store it so that it can be easily pulled out.

Specifically, by expressing the field knowledge about the object to be filed by a concept network and making it into a knowledge base, it is possible to easily add / edit new knowledge and interactively input search conditions. Knowledge is expressed in the form of general relations and concrete relations. When inputting information, the system uses the general relations to determine what the user should input next, and which class of concept it should be. Has a function of estimating and displaying on the terminal. Since this function is built on the basic principle, it is possible to adapt to a new world or a more detailed world by simply adding or changing the knowledge base represented by the concept network.

Further, there is a conceptual network editor for adding / changing a knowledge base, and the state of the knowledge base having four or more types of network negotiation functions can be displayed in a conceptual tree, a menu table, a frame, or a relational database. It is possible to give the function of displaying in the form of a tabular form, and to interact with different concepts one after another on these expressions. What makes these rich data views possible is the adoption of a newly devised conceptual relationship model. In particular, the function of jumping to different concepts while displaying the concept frame matches the human thought pattern, and a fairly advanced concept search can be executed only with this function.

Further, according to the present invention, it is possible to perform semantic content retrieval from ambiguous fragment information by performing inference using field knowledge. Even abstract information that is not useful in the conventional technology can be searched with high accuracy by "semanticly" capturing the content. Further, the search condition can be estimated at any item and to any depth, and a search expression can be created in accordance with the level that a person can remember.

[Brief description of drawings]

FIG. 1 is a system configuration diagram of an embodiment of a system of the present invention.

FIG. 2 is a diagram illustrating a concept network.

FIG. 3 is a conceptual network based on a formal expression.

FIG. 4 is a data representation diagram of a conceptual relation model based on an ER diagram.

FIG. 5 is a diagram showing a specific example of knowledge representation based on a conceptual relation model.

FIG. 6 is a diagram showing a specific example of knowledge representation based on a conceptual relation model.

FIG. 7 is a diagram showing a specific example of knowledge representation based on a conceptual relation model.

FIG. 8 is a diagram showing a specific example of a knowledge expression based on a conceptual relationship model.

FIG. 9 is a diagram showing a specific example for managing image data.

FIG. 10 is a software-like functional block diagram.

FIG. 11 is a diagram illustrating partial character string matching.

FIG. 12 is a view showing a menu.

FIG. 13 is an explanatory diagram of network negotiation using menu selection.

FIG. 14 is a view showing a concept tree display.

FIG. 15 is a view showing a hierarchical tree based on a partial whole concept relationship.

FIG. 16 is an explanatory diagram of network negotiation using a conceptual frame.

FIG. 17 is an explanatory diagram of a new concept definition / registration method.

FIG. 18 is an explanatory diagram of a conceptual network editor.

FIG. 19 is an explanatory diagram of an interactive search expression generation method.

FIG. 20 is an explanatory diagram of an interactive search expression generation method.

FIG. 21 is an explanatory diagram of an interactive search expression generation method.

FIG. 22 is an explanatory diagram of an interactive search expression generation method.

FIG. 23 is an explanatory diagram of semantic content search.

FIG. 24 is an explanatory diagram of concept matching.

FIG. 25 is an explanatory diagram of a function of displaying the state of the conceptual network in a table format.

[Explanation of symbols]

100 ... Control device, 200 ... Operation terminal, 300 ... Main memory, 400 ... Magnetic disk device group, 450 ... Optical disk device, 500 ... High-definition image display device, 600 ... High-speed image processing circuit, 700 ... Image scanner, 750 ... Image Printer, 2001 ... Dialogue control unit, 2002 ... Search / concept matching unit, 2003 ... Search expression generating unit, 2004 ... Concept network editor, 2005 ... Network negotiation unit,
2006 ... Front operation unit, 2010 ... Image display unit.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-60-254326 (JP, A) JP-A 61-5344 (JP, A) JP-A 60-14323 (JP, A) JP-A 59- 108144 (JP, A) IEICE Technical Report AL82-71 P. 77-86

Claims (16)

[Claims] 1. A plurality of terms representing a concept and a plurality of predicates representing the relationship between the concepts are stored in advance as a knowledge base, and a word indicating the first concept is used as a search condition for information retrieval. When selected, referring to the knowledge base, a set of at least one word indicating a concept having a relationship with the first concept and a set of predicates representing the relationship between the first concept are displayed on the display screen. Then, from the displayed set of words and predicates, a desired set of words and predicates showing the second concept is selected, and the words showing the second concept are related by the selected predicates. Further, at least one word and a second concept indicating a concept having a relation with the second concept by generating a word or phrase indicating the first concept as a new search condition and referring to the knowledge base. Display a set of predicates that represent the relationship with Information retrieval method characterized by displaying on the face. 2. The information retrieval method according to claim 1, wherein the knowledge base is a relation between the concepts, and an inclusion relation indicating that two specific concepts among the concepts have a superordinate concept and a subordinate concept. , A general relationship indicating that two specific concepts have a specific relationship among the above concepts, and has a general relationship with respect to the first concept and the second concept. At least 1 to show the concept
A set of predicates that represent the relationship between two terms and the above concept,
And an at least one word indicating a concept having a general relationship with the superordinate concept of the concept and a set of predicates representing the relationship between the concept and the concept are displayed on a display screen. . 3. A plurality of terms indicating a concept and a plurality of predicates representing the relationship between the concepts are stored in advance as a knowledge base, and a term indicating the first concept is used as a search condition for information retrieval. When selected, referring to the knowledge base, a set of at least one word indicating a concept having a relationship with the first concept and a set of predicates representing the relationship between the first concept are displayed on the display screen. Then, from the displayed set of words and predicates, a set of words and predicates showing a desired second concept is selected, and a new word included in the second concept is inputted, and the input is made. The generated new phrase is related to the second concept by generating a phrase indicating the first concept associated with the selected predicate as a new search condition, and referring to the knowledge base. At least one phrase indicating a concept with Information retrieval method characterized by displaying a set of predicates that representation the relationship between the second concept on the display screen. 4. The information retrieval method according to claim 3, wherein a plurality of terms indicating the concept included in the second concept and stored in the knowledge base are displayed on a display screen. Information retrieval method characterized by the following. 5. The information retrieval method according to claim 3, wherein the knowledge base is a relation between the concepts, and an inclusion relation indicating that two specific concepts among the concepts are in a superordinate concept and a subordinate concept. , A general relationship indicating that two specific concepts have a specific relationship among the above concepts, and has a general relationship with respect to the first concept and the second concept. At least 1 to show the concept
A set of predicates that represent the relationship between two terms and the above concept,
And an at least one word indicating a concept having a general relationship with the superordinate concept of the concept and a set of predicates representing the relationship between the concept and the concept are displayed on a display screen. . 6. A plurality of words and phrases indicating a concept and a plurality of predicates representing the relationship between the concepts are stored in advance as a knowledge base, and a) the first concept which is a search condition for information retrieval is defined. With respect to the words and phrases shown, by referring to the knowledge base, a set of predicates representing the relationship between at least one word and a concept having a relationship with the first concept and the first concept are displayed on the display screen. And b) receiving a selection of a set of words and predicates showing a desired second concept from the set of words and predicates displayed, and showing the second concept, or the second concept A new phrase input by the input means that is included in the above is generated as a new search condition that represents the first concept associated with the selected predicate, and c) the second concept described above. As a first concept, the operations of a), b), and c) are repeated, and finally, Information retrieval wherein the but to obtain a semantic relationship chain search condition. 7. The information search method according to claim 6, wherein the knowledge base as the relation between the concepts, and subsumption relation indicating that the particular double-concept is the relationship of the preamble and the subordinate concept of the above concepts , A general relationship indicating that two specific concepts have a specific relationship among the above concepts, and has a general relationship with respect to the first concept and the second concept. At least 1 to show the concept
A set of predicates that represent the relationship between two terms and the above concept,
And an at least one word indicating a concept having a general relationship with the superordinate concept of the concept and a set of predicates representing the relationship between the concept and the concept are displayed on a display screen. . 8. The information search method according to claim 7, wherein the search condition in which the concepts have a semantic relationship and are chained is a central subject to be searched for a search concept indicating n concepts to be searched. The first search concept and the second search concept are added to the first search concept with a relationship between the first concepts, and are sequentially added to the (n-1) th search concept. An information search, characterized in that n search concepts are described as a series of search concepts added with a relationship between (n-1) th concepts and connected with a relationship between concepts. Method. 9. The information search method according to claim 8, wherein the terminal concept, which is the lowest concept stored in the knowledge base, is collated with the first search concept to include the first search concept. Is extracted, and the first relational concept, which is a concept in the relationship between the first concepts, is collated with the second concept, and the second concept is extracted from the extracted final concept. An end concept in which one relational concept is included in the second search concept is extracted, matching is sequentially repeated for the extracted end concept, and finally the subordinate concept of the first search concept is compared with the first concept. A subordinate concept of the second search concept is added with a relationship between the first concepts, and a subordinate concept of the nth search concept is sequentially added to a subordinate concept of the (n-1) th search concept. Added with a relationship between the (n-1) th concept,
An information retrieval method, characterized in that an end concept described as a series of connected concepts having a relationship between concepts is extracted as a retrieval result. 10. The information search method according to claim 9,
The information corresponding to the terminal concept which is the lowest concept stored in the knowledge base is stored in the database in advance, and the information stored in the database corresponding to the terminal concept extracted as the search result is output. Information retrieval method characterized by the following. 11. A plurality of terms indicating a concept and a plurality of predicates representing the relationship between the concepts are stored in advance as a knowledge base, and search conditions for information retrieval are set as the plurality of terms and the plurality of terms. Is defined by a predicate relating to the above, and with respect to the word or phrase indicating the first concept appearing in the search condition, at least one word or phrase indicating a concept related to the first concept is referred to by referring to the knowledge base. A set of predicates that represent the relationship with the first concept is displayed on the display screen together with the search condition, and a phrase and a predicate indicating a desired second concept from the displayed set of phrases and predicates. Of the second concept, or a new phrase input by the input means included in the second concept, which is related by the selected predicate. New words that show the concept Generated as a condition, and the information retrieval method of the displayed search criteria and displaying on a display screen rewriting in accordance with the new search condition. 12. A storage unit for storing, as a knowledge base, a plurality of words and phrases indicating a concept and a plurality of predicates representing a relationship between the concepts, a display unit having a display screen, and inputting a word and phrase indicating a retrieval concept and The knowledge base is referred to for the input means capable of designating the search concept displayed on the display screen and the word or phrase indicating the first concept selected as a search condition for information search by the input means. The at least one word indicating a concept having a relationship with the first concept and a set of predicates representing the relationship between the first concept on the display screen; A selection of a combination of a phrase and a predicate indicating a desired second concept from the displayed combination of a phrase and a predicate, and a phrase indicating the second concept or an input included in the second concept. Entered by means Information Retrieval System new phrase and having a retrieval condition generating means for generating a phrase indicating the first of the concepts associated with the predicate which is the selected as a new search condition. 13. The information retrieval system according to claim 12, wherein the knowledge base is a relation between the concepts, and an inclusion relation indicating that two specific concepts among the concepts are in a superordinate concept and a subordinate concept. , A general relationship indicating that two specific concepts have a specific relationship among the above concepts, and has a general relationship with respect to the first concept and the second concept. A set of predicates that represent the relationship between at least one word indicating a concept and the above concept, and a relationship between the above concept and at least one word that indicates a general relationship with a superordinate concept of the above concept; An information retrieval system characterized by displaying a set of represented predicates on a display screen. 14. The information search system according to claim 13, wherein the search condition generation means includes a first search concept, which is a central subject to be searched, for the n search concepts indicating the concepts to be searched, and the first search concept. The second search concept is added to the one search concept with a relationship between the first concepts, and the nth search concept is sequentially added to the (n-1) th search concept.
-1) Information search characterized by finally generating search conditions added as a relationship between concepts and described as a series of search concepts connected with a relationship between concepts system. 15. The information retrieval system according to claim 14, wherein the terminal concept, which is the lowest concept stored in the knowledge base, is collated with the first retrieval concept to include the first retrieval concept. Is extracted, and the first relational concept, which is a concept in the relationship between the first concepts, is collated with the second concept, and the second concept is extracted from the extracted final concept. An end concept in which one relational concept is included in the second search concept is extracted, matching is sequentially repeated for the extracted end concept, and finally the subordinate concept of the first search concept is compared with the first concept. Subordinate concepts of the second search concept are added with the relationship between the first concept, and sequentially added to the (n-1) th concept.
The subordinate concept of the n-th search concept is added to the subordinate concept of the search concept with the relationship between the (n-1) th concepts, and is connected with the relationship between the concepts. An information retrieval system having a concept retrieval means for extracting, as a retrieval result, a terminal concept described as a series of concepts. 16. The information search system according to claim 15, further comprising a database storing information corresponding to a terminal concept, which is the lowest concept stored in advance in the knowledge base.