JPH0749875A - Document information classification method, document information collection method using the same, and document information collection system - Google Patents

Abstract

(57) [Summary] (Modified) [Configuration] The document collection server system 100 automatically connects to a plurality of information sources to acquire a new document, and calculates the degree of conformance 1
According to 06, the matching degree with the search condition previously described by the user is checked. The document storage processing 107 forms a classification system from the relationship between the search conditions, classifies the matched documents, and stores them in a folder. The folder management process 108 monitors the collection of information in each folder and automatically subdivides the folders, integrates them, and changes the structure to organize the information. [Effect] It is possible to improve the classification system and the search conditions according to the degree of information gathered in each classification, and to suppress the amount of information classified into each classification to a number that can easily grasp the whole.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information collecting system for automatically collecting, classifying and organizing information via a computer network.

[0002]

2. Description of the Related Art The development of computer networks is advancing rapidly, and there is an environment where the so-called broad catch of information can be performed, such as online information search services, information collection from net news, and question and answer using electronic mail and electronic bulletin boards. It's getting ready.

Although the usefulness of these latest information has been recognized, it has not been effectively used due to the following problems.

(1) The usage varies depending on the information source, and the operation of collecting information from a plurality of information sources is complicated.

(2) The search expression must be input as a logical expression. It is difficult to write an appropriate search formula for obtaining desired information.

(3) It takes time and effort to classify and organize the collected information.

"The 21st Century Information Society" (Nikkei Byte 1
Wide-area information server WAIS described in November 991, pages 320 to 331) makes the problem (1) common to the protocol (expands NISO Z39.50), and connects to information sources and searches. Is solved by automating
The problem of (2) was solved by relevance feedback. Relevance feedback is a method of refining the following search conditions. When the user describes the contents to be searched, WAIS searches and presents the information that matches the contents, using the contents as the search condition. When the user selects the information that he / she wants, WAIS feeds back the information selected by the user to the search condition and improves the search condition. The information search using this relevance feedback allows the user to search for desired information without describing a search formula.

In order to solve the problem (3), various document automatic classification systems have been devised.

For example, the document classification system of Japanese Patent Laid-Open No. 1-188834 obtains information on the frequency of appearance of keywords in each field by examining a sample document group, extracts the keywords from the input document, and extracts the keywords for each field. Calculate scores and classify into the highest scoring areas.

The notification document processing system of Japanese Patent Laid-Open No. 63-214832 analyzes the format of the notification document, adds the weights of the words appearing in the message to each category, calculates the sum of them, and selects the maximum category. Classify by

[0011]

[Problems to be Solved by the Invention] WAIS is based on the above (1),
Although the problem of (2) was solved, no consideration was given to the classification and organization of the collected information.

It is desired to classify and organize information hierarchically, but the conventional method is not suitable for this.

Further, in the conventional automatic classification system that solves (3), it is necessary to establish a classification system for classification in advance. Furthermore, it is necessary to give or obtain in advance the keyword group that characterizes each field and the appearance frequency of the keyword group.

However, it is difficult to provide an appropriate general-purpose classification system in advance. Inappropriate classification systems can concentrate a lot of information in a classification. When the amount of information of a certain category becomes too large, it becomes difficult for the user to grasp the whole content of the collected information.

Further, in the most advanced fields, the classification system and technical terms accepted by many people are often unfixed,
And it changes frequently. It is difficult to properly classify documents related to the state of the art with a conventional automatic classification system.

A first object of the present invention is to provide a document information classification method suitable for hierarchically classifying and organizing information, and a document information collecting method and system using the same.

A second object of the present invention is to provide a document information collecting method and system for automatically improving a classification system and a search condition used for classification based on a collection condition of collected document information.

[0018]

According to a first document information classification method of the present invention, a search condition consisting of one or a plurality of search conditions corresponding to each of a plurality of folders which are associated with each other in a hierarchical relationship. A group is stored, the matching degree between the information to be classified and each folder is detected based on the search condition group stored corresponding to each folder, and the detected between each folder and the information is detected. Based on the compatibility and the hierarchical relationship, one or more of the plurality of folders is determined as a folder to which the information corresponds, and the determined one folder or each of the plurality of folders is determined. Storing the information.

A second document information classification method according to the present invention is
A search condition group consisting of one or a plurality of search conditions is stored in correspondence with each of a plurality of folders which are associated with each other in a hierarchical relationship, and the search condition group stored in advance corresponding to each folder is predetermined. Based on the determination criteria, one or more of the plurality of folders is determined as a folder to which the information to be classified is associated, the information is stored corresponding to the determined folder, and the plurality of information to be classified For each of the above, the above determination and storage are performed, and it is determined whether or not the plurality of pieces of information stored corresponding to each folder satisfy a predetermined condition defined for the reconstruction of the folder. And a step of reconstructing a plurality of information stored corresponding to the one folder and a group of search conditions stored corresponding to the folder when one folder satisfies the predetermined condition.

A third document information classification method according to the present invention is
A search condition group consisting of one or a plurality of search conditions is stored in correspondence with each of a plurality of folders which are associated with each other in a hierarchical relationship, and the search condition group stored in advance corresponding to each folder is predetermined. Based on the judgment criteria, one or more of the plurality of folders is determined as a folder for associating the information to be classified, and the information is stored in correspondence with the determined folder and a plurality of classifications are performed. The above-mentioned determination and storage are performed for each piece of information that should be stored, and a plurality of pieces of information stored corresponding to some of the plurality of folders are determined for reconstruction of the plurality of folders. It is determined whether or not a predetermined condition is satisfied, and when any one of the plurality of folders satisfies the predetermined condition, a plurality of pieces of information stored in correspondence with the some of the plurality of folders are stored. , Some of the Comprising the step of reconstructing a set of search conditions stored in correspondence with the da.

[0021]

In the first document information classification method according to the present invention, the matching degree between the search condition stored in association with each folder and the search target document information and the hierarchical structure of the plurality of search conditions are considered, Since the folders to be associated with the search target document information are determined, the collected document information can be classified by regarding the search condition group described by the user as a classification system having a hierarchical structure.

In the second document information classification method according to the present invention, folders can be reconfigured such as dividing each folder depending on the document information stored corresponding to each folder. Therefore, the classification system can be automatically changed according to the degree of collection of the document information obtained by the search.

In the third document information classification method according to the present invention, it is possible to reconfigure folders that span a plurality of folders.

[0024]

EXAMPLE One example of the present invention will be described below.

The target of the document information collecting system of the present embodiment is a group of contents which are meaningful to the user and which can be obtained electronically through an online document information search service, electronic mail, electronic bulletin board and the like. It is the information represented by the character. Hereinafter, such information will be referred to as document information.

Each of these services is operated by various companies and organizations. Hereinafter, these services will be referred to as information sources. Since the document information provided by each information source generally has a wide variety, it is presented to the user in a plurality of fields. These fields are called domains. Individual information provided in the domain is called a document. A search result storage area that is stored when a document matches the search condition is called a folder.

FIG. 1 shows an example of a system configuration including a document collection system of this embodiment and an external information source for collecting documents by the document collection system of this embodiment. The document collection system of this embodiment includes a document collection client 500 and a document collection server 510.

A plurality of document collection clients 500 exist on the network and can simultaneously access the document collection server 510.

Memory 52 of document collection client 500
In the document collection client system 501 on 2, the user creates a folder for storing the collected documents, registers a search condition indicating what kind of document is to be collected in each folder, or is collected in the folder. It provides a graphical user interface for viewing and viewing documents.

The document collection server system 100 on the memory 523 of the document collection server 510 provides document information in response to a request from the document collection client system 501, while automatically automatically providing the news server 520 and the document server 521. Documents that meet the search condition group registered by the user are collected from external information sources, and further classified and organized.

First, the document collection client system 50
1 will be described.

The user collects the document collecting client system 5
When 01 is activated, an interface screen 400 as shown in FIG. 3 is displayed on the CRT 502. The user can use the keyboard 503 and mouse 5 on the interface screen 400.
Perform various operations using input devices such as 04 to create / delete folders that store collected documents, describe search conditions for collecting documents, view collection results, and evaluate To do.

FIG. 7 shows a flow chart of the processing performed by the document collection client system 501. When the document collection client system 501 is activated, first, the document collection server system is connected (step 120). Next, the interface screen 400 shown in FIG. 3 is displayed (step 12).
1).

After that, the event loop 122 is entered, and steps 123 to 126 are repeated. That is, the operation of the user is accepted and analyzed (step 123), the command corresponding to the operation is transmitted to the document collection server system 100 (step 124), and the execution result is sent to the document collection server system 10.
It is received from 0 (step 125) and the execution result is reflected on the interface screen 400 (step 12).
6) is repeated.

When the user selects the end from the menu, the end command is issued to the document collection client system 5
01, the process exits the event loop 122, disconnection processing with the document collection server system 100 is performed (step 127), and the processing ends.

The interface screen 400 shown in FIG.
A specific example of will be described. On this screen, the user has already created a folder and registered search conditions for documents to be collected in the folder, and the document collection server system 100
Is an example at the time when documents matching and collecting search conditions registered by the user are collected and classified.

Reference numeral 402 is the name of the folder whose contents are being displayed. In this example, the contents of the folder named voice, which is a subordinate directory of the folder named user1, are being displayed.

In 403, search conditions registered by the user in the folder user1 / voice are displayed. The search conditions can be updated by directly editing the displayed text. In this embodiment, the search condition stored for each folder is a word (hereinafter referred to as a word) or a phrase,
Alternatively, it is a list of things that the user can describe in natural language, such as sentences.

In 404, a list of folders under the folder user1 / voice is displayed. For each folder, the folder name, the number of documents collected in the folder, and the writing of search conditions corresponding to the folder are displayed. In this example, recognition and synt are listed under user1 / voice, respectively.
There are two subfolders of hesis.

By clicking an item in this subordinate folder list, it is possible to move to the clicked subordinate folder.

At 405, a list of documents already collected in the folder user1 / voice is displayed.

For each document, title, folder user
The degree of conformity to the search condition of 1 / voice, the word in the matched search condition, the information source name, the domain name, etc. are displayed.

By clicking an item in this document list, the contents of the clicked document can be viewed. The content of the document is displayed at 406.

Creating / deleting a folder can be done from the menu 401.
This is done using the File menu. You can also use the Goto menu to move to another folder.

Further, the user can give an evaluation to the collected document or the folder in which the document is stored by using the Edit menu of the menu 401. That is,
When the user uses the menu to give an evaluation that it is a useful / useless document or a useful / useless folder, the corresponding command is the document collection server system 100.
Sent to. The document collection server system 100 reflects the evaluation of the documents and folders in the search condition, and collects the documents more suitable for the user from the next document collection.

The document collection server system 100 on the memory of the server 510 processes the request from the client 500, while collecting and classifying the documents based on the folder group created by the user and the search condition registered in each folder.・
Organize.

That is, the document collection server system 100
Regularly accesses an external information source such as the news server 520 or the document server 521, acquires the documents accumulated in each information source after the previous access, and matches the search condition group with which the user is registered. Search for what you want to do. On this occasion,
The number of appearances of each word in the search condition in the target document is defined as the matching degree between the target document and the search condition. From the folders in which the matching search conditions are registered, a folder into which the target document is classified is selected in consideration of the hierarchical structure of the folders and stored in that folder. Furthermore, the documents are organized according to the document collection status such as automatically dividing the folder in which many documents are accumulated.

The external information source that is the target of the document collection may be on another network accessible from the server.

Before describing in more detail the collection / classification / arrangement of documents, first, folders created by the user and search conditions will be described with reference to the example shown in FIG.

When a user is registered in the document collection server system 100, one folder is assigned to each user. The user can freely create subordinate folders hierarchically under the folder assigned to him and register the search condition for each folder as to what documents should be collected in that folder. .

In the example of FIG. 4, there are two users, user1 and user2.
Are registered, and a folder 540 and a folder 550 are respectively assigned. user1 is in folder 54
Folders 541 to 544 are hierarchically created under 0, and search conditions 545 to 548 are registered in each folder.

On the other hand, user2 does not create a subordinate folder,
In the folder 550, the search conditions 551 are simply registered, listing the matters of interest.

In addition to the user creating and updating the folders and the search conditions corresponding to the folders, the document collecting server system 100 may also automatically create or update the folders according to the document collection status. is there. Details will be described later.

Therefore, the collected documents are automatically classified / arranged simply by listing the matters of interest without creating a hierarchical folder like user2.

The document collection server system 100 will be described with reference to the flowchart of FIG.

The document collection server system 100 always responds to requests from a plurality of users and, at the same time, periodically collects documents, so that there is always a user connection request or whether a certain time has elapsed. Is being monitored (step 101). When the user makes a connection request, the client request processing 110 is started. When the fixed time has elapsed, the document collection process 102 is started. In either case, the process of the document collection server system 100 main body immediately returns to step 101 to continue the connection request from the user and the monitoring of the elapse of a fixed time.

FIG. 6 shows the correspondence between the flow of the client request processing 110 and the data structure used during command execution.

When the client request processing 110 is started, first, as a preparation for processing the request from the client, connection with the client (step 111) and loading of the folder table (step 112) are performed.

Thereafter, the command transmitted from the client 500 is repeatedly analyzed (step 113) and executed (step 114) until the end command is received from the client.

When the end command is received and the repetition is ended, the client is disconnected and the client request processing 110 is ended.

When executing each command, various tables are loaded, referenced, updated, and saved as necessary.

For example, when the user performs an operation of requesting the display of the document list stored in a certain folder, the document collection client system 501 sends the corresponding command and the target folder name. Client request processing 1
When 10 receives this command and the folder name, it refers to the folder table and refers to the information of the documents stored in the folder (title of each document, degree of conformity, word in conforming search condition, source name, etc.) ) Is sent to the client.

Document collection process 102 shown in FIG. 5 (FIG. 2)
The document collection process will be described in accordance with the correspondence between the flow of the above and the data structure used in the document collection process.

First, a table for collecting documents (document number table 300, folder table 310, word / folder table 330, word / document table 350) is loaded from the internal DB 511 onto the memory (step 1).
03).

The document number table 300 shows what types of information sources are available, what domains each information source has, what number of documents each of these domains has, and It represents information up to which number has been acquired.

The folder table 310 shows what kind of folder has what kind of hierarchical structure, and what kind of document is stored in each folder.

The word / folder table 330 represents what words appear in the search condition associated with each folder.

The word / document table 350 shows what kind of word appears in which document. Details of each table will be described later.

Next, for all new documents of each information source,
Repeat steps 105 to 107.

In the new document acquisition processing of step 105, the information source is connected to check whether there is a newer document than the document number registered in the document number table 300, and if there is, obtain the document.

Next, the fitness calculation in step 106 calculates the fitness in each folder of the acquired document. First, a folder search table 370 that represents what words appear in the acquired document is created, and a compatible folder table 390 is created and initialized to store the degree of suitability in each folder. Then, the folder search table 370 and the word folder table 330 are collated to calculate the matching degree for each folder, and the matching degree is registered in the matching folder table 390. Details of the fitness calculation will be described later.

Next, in step 107, the document storing process determines the folder for storing the document from the degree of conformity in each folder registered in the conforming folder table 390 and the hierarchical structure between folders represented by the folder table 310, and The document is registered in the folder table 310 and the word / document table 350. Details of the document storage processing will be described later.

Next, the folder management processing of step 108 analyzes the documents in the folder using the word frequency distribution in each document represented by the word / document table 350, and performs automatic folder division and integration. The table 310 and the word folder table 330 are updated. Details will be described later.

The document collection table updated during the repetition of the above steps 105 to 107 is saved in the internal DB 511 (step 109).

Up to this point, the general document collection processing 102
To finish.

The data structure and processing used in the document collection processing 102 described above will be described in more detail.

[0077] The data structure of the document number table 300 ^-
13 shows. The document number table 300 is a hash table, and each entry has a document number list 30 shown in FIG.
Pointing to 2. The entry is determined by the value of the hash function that takes the source name and domain name as input.

The document number list 302 includes a pointer 303 to the information source name, a pointer 304 to the domain name, the number 305 of the oldest document in the domain, and the number 30 of the latest document.
6, a set of the document number 307 of the document which the document collection system has already collected, and a pointer 308 to another document number list of the same hash value.

The document number table is loaded at the beginning of document collection and is updated each time a document is acquired from the information source.

The internal DB stores what kind of information source exists, what kind of domain exists, and how many documents in which domain have been collected and processed. First, the stored information source name, domain name, and acquired document number are read from the internal DB 511, and the document number list 302 is read.
Is created, and the value of the hash function with the information source name and the domain name as input is registered in the document number table 300 as an entry. Next, the oldest document number and latest document number of each domain are acquired from each information source and written in the document number list. At this time, if there is a domain not registered in the document number table 300, this is a domain newly created in the information source, so that the document number list is generated with the already acquired document number set to 0, and the document number table 300
Register with.

For example, the document number list 302 shown in FIG.
-A indicates that there are 123 to 145 documents in the domain fj.ai of the internet news source, and 130 of them have been collected and processed.

The data structure of the folder table 310 is shown in FIG. The folder table 310 is a hash table, and each entry has a folder list 314 shown in FIG.
Pointing to. The entry is determined by the value of the hash function that takes the folder name as input.

The folder list 314 is a folder ID.
A reference numeral 315, a pointer 316 to a folder name, a pointer 317 to a folder list showing a higher folder, a pointer 318 to a lower folder list 321, a pointer 319 to a stored document list 324, and a folder list showing another folder having the same hash value. Is a set of pointers 320 to.

The lower folder list 321 is a set of a pointer 322 to a folder list representing a lower folder and a pointer 323 to a lower folder list representing another lower folder of the folder represented by the folder list 314.

The stored document list 324 is a pointer 325 to an information source name of a stored document, a pointer 326 to a domain name, a document number 327, and a matching degree of a document represented by the stored document list 324 in a folder represented by a folder list 314. 328 is a set of pointers 329 to the stored document list representing other documents stored in this folder.

For example, in the folder list 314-a of FIG. 15, the upper folder of the folder voice whose folder ID is 1003 is the folder user1 represented by the folder list 314-b, and is represented by the folder list 314-c. Have a folder synthesis, which is a subordinate folder, and that this folder has an internet new
This indicates that the 120th document in the domain fj.ai of the information source s is stored.

Word folder table 3 shown in FIG.
Reference numeral 30 is a hash table, and each entry points to the word folder list 333 shown in FIG. The entry is defined by the value of the hash function that takes a word as input.

The word folder list 333 is a set of a pointer 334 to a word, a pointer 335 to a folder frequency list 340, and a pointer 336 to another word folder list of the same hash value. The folder frequency list 340 is a set of a folder ID 341 of a folder corresponding to a search condition in which this word appears, an appearance frequency 342 of a word in the search condition, and a pointer 343 to another folder frequency list.

For example, the folder list 333-a and the folder frequency list 340-a shown in FIG. 17 indicate that the word "language" appears once in the search condition corresponding to the folder with the folder ID 1003.
3-b and folder frequency list 340-b, 340-c
Indicates that the word “voice recognition” appears once in the search condition corresponding to each of the folder with ID 1003 and the folder with ID 1004.

The data structure of the word / document table 350 is shown in FIG. The word / document table 350 is a hash table, and each entry points to the word / document list 354 shown in FIG. The entry is determined by the value of the hash function that takes a word as input.

The word / document list 354 of FIG. 18 is a set of a pointer 355 to a word, a pointer 356 to a document frequency list 360, and a pointer 357 to another word / document list having the same hash value. Document frequency list 360
Is a pointer 361 to the information source name of the document in which this word appears, a pointer 362 to the domain name, and a document number 36.
3, the appearance frequency 364, and a pointer 365 to another document frequency list in which this word appears.

For example, the word / document list 33 in FIG.
4-a and document frequency list 360-a indicate that the word language appears 5 times in the 56th document of domain fj.sci.lang of the information source internet news, and word and document list 334-b. Document frequency list 360-b, 360
The word c is the word "voice recognition" in the source of information internet news
It appears twice in the 120th document of domain fj.ai and twice in the 56th document of domain fj.sci.lang.

The contents of the folder table 310, word / folder table 330, and word / document table 350 are stored in the internal DB 511. These tables are stored when the document collection process 102 is started, when the client request process 110 is started, or when a command is executed.
It is loaded into the memory as needed, is referenced / updated while each process is being executed, and internal DB 511 is used when it ends.
Will be saved to. However, each table is updated exclusively. The update of the folder table by creating / deleting a folder and the update of the word / folder table by updating the search condition are immediately saved.

As an example, the word folder table 350
FIG. 8 shows a flow chart for the loading of No. The word folder table 350 is loaded by the folder table 310.
After loading.

First, the word folder table 350 is initialized (step 160).

Next, the word registration of the folders (steps 164 to 166) is repeated for all the folders registered in the folder table 310 (step 16).
1).

In the word registration of a folder, first, search conditions corresponding to the folder are stored in the internal DB 511 to the memory 5
23 is read (step 164) and the word is extracted (step 165). The word folder list 333 of FIG. 16 is created for each extracted word, and the hash value of the word is calculated and registered in the word folder table 330 of FIG. 17 (step 166).

When word registration is performed for all folders, this repetition is ended, and the word / folder table load processing 151 is ended.

The search process performed by the fitness calculation 106 will be described with reference to FIG.

This processing is performed by checking the similarity between the word appearing in the search condition group and the word appearing in the document,
Check the conformity between the acquired document and each folder.

Here, the matching degree between the document to be searched and any of the folders used in this embodiment will be described.

There are several possible methods of matching the search target document with any of the folders, but in the present embodiment, among the words in the document, words that match the folder (that is, the folder). Of each word in the document that matches the word contained in the search condition stored in association with the folder and the sum of the degrees of conformity between those words and the folder is calculated. The sum is the fitness of the document and the folder.

Here, the fitness of the word adapted to the folder and the fitness of the folder can also be obtained by various methods, but in this embodiment, as a more preferable one, the weight of the word in the document is set. And the weight of the word in the folder to obtain the fitness of the word and the folder.

Here, the weight of the word in the document can be detected by various methods, but in this embodiment, it is more preferable that the weight is the frequency of appearance of the word in the document. It is the weight of the word in the document.

Further, the fitness between the word and the folder can be detected by various methods, but in the present embodiment, it is more preferable that the fitness within the search condition stored corresponding to the folder is satisfied. Use the number of occurrences of that word.

Therefore, in this embodiment, the matching degree between the word and the folder is represented by the product of the appearance frequency of the word in the document and the appearance frequency of the word in the search condition corresponding to the folder. The fitness between the search target document and the folder is given by the sum of the fitness of each word thus obtained.

More specifically, a word / folder table 330 in which the words appearing in the acquired document are registered in the folder search table 370 of FIG. 21, and the words appearing in the search conditions corresponding to all folders are registered. Collation is performed and the goodness of fit is totaled for each folder, and the folders are sorted in order of goodness of fit.

First, the folder search table 370 is initialized (step 170) and the compatible folder table 390 of FIG. 23 is initialized (step 171).

Next, words are extracted from the acquired document (step 172) and each word is registered in the folder search table 370 (step 173).

The folder search table 370 is a hash table, and each entry points to the folder search list 372 shown in FIG. The entry is determined by the value of the hash function that takes the word as an argument. The folder search list 372 is a pointer 373 to a word in the document, a pointer 374 to a matching folder list 380, and an appearance frequency 37 in the document.
5, a set of pointers 376 to other folder search lists having the same hash value. The matching folder list 380 is a folder of folders corresponding to the search condition in which the word appears.
It is a set of an ID 381, a matching degree 382 in the folder, and a pointer 383 to another matching folder list.

For example, the folder search list 372-a of the folder search table 330 shown in FIG. 21 indicates that the word "language" appears twice in the document to be searched. Since no search has been performed yet, the pointer 374-a to the matching folder list is NULL. Similarly, the folder search lists 372-b and 372-c respectively show that the word "silent time" appears three times and the word "voice recognition" appears five times in the target document.

Next, the folder search table 370 and the word / folder table 330 are collated, and if there is a matching folder, a matching folder list is created and registered in the folder searching table (step 174).

That is, the word registered in the folder search table 370 is the word / folder table 3
If it is also registered in 30, the folder search list 372
Value 375 and the frequency 342 of each folder frequency list 340 registered in the word / folder list 333 are taken as the matching degree in each folder of the word, and the matching folder list 380 corresponding to each
Is created and registered in the folder search list 372.

For example, the word "language" registered in the folder search table 370 appears once in the search condition corresponding to the folder ID 1003 as shown in the folder frequency list 340-a of the word folder table 330. . Therefore, the word ID 1
The matching degree in the folder 003 is 2 points, and the pointer to the matching folder list, which was NULL before the search is executed, has a matching folder list 380-a as shown in the folder search table 370 after the search shown in FIG. Refers to.

Similarly, the word voice recognition is
As shown in the folder frequency list 340-b of the folder table 330, each appears once in the search condition corresponding to the folder of ID1003 and the folder of ID1004. Therefore, the word ID 1003
There are 5 points of conformity in each of the folder and the folder of ID 1004. Therefore, after executing the search,
The compatible folder lists 380-b and 380-c are registered in the second folder search table 370. However, the word silent time is not registered in the word folder table. That is, since there is no matching folder, the pointer to the matching folder list is NULL even after the search is executed.

Finally, the goodness of fit in the folder of each word is tabulated for each folder, and the folders whose non-fitnesses are not 0 are arranged in descending order of goodness of fit.
(Step 175). Each entry in the compatible folder table points to the compatible folder list 381. The conforming folder list has the same data structure as the conforming folder list of FIG. 20, but the total of the conforming degrees in the folders of each word is the conforming degree, and the pointer 383 to another conforming folder list is not used.

The document storing process 109 selects and stores the folder in which the document is to be stored by using the adaptive folder table 390 created by the above-described fitness calculation and the folder table 310 in which the hierarchical structure of folders is registered. . The flow of the document storage processing 109 is shown in FIG.

This document storage processing is roughly divided into two stages (steps 180 and 181).

First, in step 180, it is determined which folder the target document should be stored in based on the conformance of each folder registered in the compatible folder table 390 and the hierarchical structure of the folders registered in the folder table 310. The document is registered in the folder table 310 as a document storing the document.

FIG. 24 shows an explanatory diagram of the method of determining the storage folder. In this embodiment, the folder is stored in the lowest folder among the matched folders in each branch of the folder hierarchical structure. A
If there are folders from H to H, and each has the matching degree shown in the figure, the lowest D among the matching folders of A-B-D-G, as well as branch A-B-E E inside,
Store in H of branch A-C-F-H. In this method, it is considered that the lower folders inherit the search conditions of the upper folders, and the search conditions are stored in a folder in which the search conditions are described in more detail.

Next, in step 181, the words appearing in the target document are registered in the word / document table 350.

That is, the word / document list of FIG. 18 is created for all the words appearing in the document, and registered in the word / document table 350.

By this processing, the word / document table 3
In 50, what words appear in each document are recorded for all the stored documents. The word / document table 350 is used to analyze the documents in the folder in the folder management process described below.

The folder management process 108 will be described.

When the retrieval conditions corresponding to folders having a hierarchical structure are regarded as a classification system and documents are collected and classified,
Documents may be concentrated in a specific folder, and the number of documents in the folder may increase beyond the user's knowledge. In addition, documents are often stored in a plurality of folders in duplicate, which may be wasteful.

These phenomena occur when the user does not properly configure the classification system, or when the situation and research trends in the world change and the classification system does not match.

The folder management processing 108 detects these phenomena by analyzing the collection state of documents in each folder, and improves the hierarchical structure of folders and the search condition corresponding to the folders. As a result, the number of documents in a folder can be suppressed to a number that the user can grasp, and documents can be prevented from being stored in multiple folders unnecessarily redundantly. To maintain.

FIG. 11 shows the flow of folder management processing.

Steps 201 to 205 are repeated for each folder (step 200).

First, the number of documents stored in the folder is monitored (step 201). If the folder stores more than the given number of documents in advance, the documents in the folder are analyzed using a statistical method (step 202).
A method of collecting similar individuals and classifying them into groups among objects having different properties is known as cluster analysis. For example, “Multivariate Analysis Handbook” (Hyundai Mathematics Co., Ltd. 1986). Year). Step 202 uses the method of cluster analysis,
Reclassify documents based on the frequency of words that appear in the documents in the folder. A set of reclassified documents is called a cluster.

Next, the relationship between clusters is analyzed to determine the hierarchical structure of clusters (step 203). The relationship analysis between clusters performed here will be described later. Folders and search conditions are generated corresponding to the clusters, and folders are created hierarchically corresponding to the hierarchical structure of the clusters (step 204). Next, from the word / document table 350, words that appear frequently in the generated documents in each folder are extracted and added to the search condition corresponding to the folder (step 205). As a result, the search condition can be refined.

When the above processing is applied to each folder,
The document group stored in each folder is analyzed, and the folders are reconstructed, that is, the folders are integrated and the hierarchical structure is changed (step 206). The analysis performed here will be described later.

A method of analyzing the inter-cluster relationship performed in step 206 will be described with reference to FIGS.

As shown in FIG. 25, it is assumed that there is a search condition consisting of a word group w1 and a word group w2, and a document group d is stored in a folder 450 corresponding to this search condition. At this time, regarding the documents in this folder, the pattern of the relationship between the word and the document, which is obtained by statistically analyzing the data obtained from the word / document table, is shown by 451 in FIG. 26, 455 in FIG. 26, and 458 in FIG. Shown in.

FIG. 26 shows a case where the document group d is classified into two independent clusters, a document group d1 in which the word group w1 appears and a document group d2 in which the word group w2 appears. In this case, a search condition composed of the word group w1 and a search condition composed of the word group w2 are generated, and folders 453 and 454 corresponding to them and a folder 452 above them are provided,
The hierarchical structure shown in FIG. 26 is used.

FIG. 27 shows a document group d1 in which only word group 1 appears, a word group w1 and a document group d in which word group w2 appears.
This is the case of being classified into two clusters of 2. The word group w1 also appears in the document group in which the word group w2 appears. Therefore, a search condition including the word group w1 and a search condition including the word group w2 are generated, and folders 456 and 457 corresponding to the search conditions are provided to form the hierarchical structure shown in FIG.

FIG. 28 shows a document group d1 in which only the word group w1 appears, a document group d3 in which only the word group w2 appears, and a document group d2 in which both the word group w1 and the word group w2 appear.
This is the case of being classified into three clusters. in this case,
A search condition including only the word group w1 and a search condition including only the word group w2 and a search condition including the word group w1 and the word group w2 are generated, and folders 45 corresponding to the search conditions are generated.
6, 457, 458 and folders 455 above these are provided to form a hierarchical structure as shown in FIG.

Using the same FIGS. 25-30, step 205
The method of analyzing the relationship between folders will be described below.

If there are more documents stored in the folders 453 and 454 in duplicate when there are folders having a hierarchical structure as shown in FIG. 27, the pattern of the relation between words and documents is 457 from the pattern of 451 or 457. It is thought that the pattern changed to 458. It can be detected from the folder table 310 that the documents are redundantly stored. Therefore, with respect to the documents stored in the folders 453 and 454, statistical analysis similar to the inter-cluster relation analysis is performed based on the word appearance frequency in each document obtained from the word / document table 350, and the distribution of the words and the documents is analyzed. Examine patterns and reconfigure folders and search conditions according to the patterns.

Further, when there is a hierarchical structure as shown in FIG. 27, among the documents stored in the folder 457, the folder 4
When the ratio of documents that do not conform to 56 exceeds a predetermined ratio, it means that the hierarchical relationship of the search conditions has changed. This can be detected by checking the compatibility of the document stored in the folder 457 registered in the folder table 310 with the folder 456. Cluster analysis is performed on the documents stored in the folders 456 and 457 to reconfigure the folders.

[0141]

According to the present invention, the following effects can be obtained.

(1) Information matching the search condition group described by the user can be collected from a plurality of information sources, and the collected information can be classified by regarding the hierarchical structure of the search condition group as a classification system.

(2) The classification system can be changed according to how information is collected in the search result storage area corresponding to each search condition.

As a result, the amount of information stored in a certain search result storage area can be suppressed to a number that allows the user to grasp the entire information. Alternatively, an appropriate classification system can be maintained according to changes in the situation.

[Brief description of drawings]

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

FIG. 2 is a flowchart of the document collection server system of this embodiment.

FIG. 3 is an example of an interface screen of this embodiment.

FIG. 4 is an example of folders and search conditions corresponding to the folders of the present embodiment.

FIG. 5 is a flowchart of a document collection process of this embodiment.

FIG. 6 is a flowchart of client request processing according to the present embodiment.

FIG. 7 is a flow chart of the document collection client system of the present embodiment.

FIG. 8 is a flow chart of a word folder table loading process of the present embodiment.

FIG. 9 is a flowchart of the fitness calculation in this embodiment.

FIG. 10 is a flowchart of a document storage process of this embodiment.

FIG. 11 is a flowchart of a folder management process of this embodiment.

FIG. 12 is a data structure of a document number list of this embodiment.

FIG. 13 is a data structure of a document number table of this embodiment.

FIG. 14 is a data structure of a folder list, a subordinate folder list, and a stored document list according to this embodiment.

FIG. 15 is a data structure of a folder table according to the present embodiment.

FIG. 16 is a data structure of a word / folder list and a folder frequency list according to this embodiment.

FIG. 17 is a data structure of a word folder table according to the present embodiment.

FIG. 18 is a data structure of a word / document list and a document frequency list according to this embodiment.

FIG. 19 is a data structure of a word / document table according to the present embodiment.

FIG. 20 is a data structure of a folder search list and a compatible folder list according to this embodiment.

FIG. 21 is a data structure of a folder search table of this embodiment.

FIG. 22 is an example of a folder search table after the search processing of the present embodiment.

FIG. 23 is a data structure of a compatible folder table of this embodiment.

FIG. 24 is an explanatory diagram of a method of determining a folder for storing a document according to this embodiment.

FIG. 25 is an explanatory diagram of folders for performing in-folder document analysis according to the present embodiment.

FIG. 26 is an explanatory diagram of a word generated as a result of document analysis in a folder and a first distribution pattern of a document according to the present embodiment, and a folder generated according to the word;

FIG. 27 is an explanatory diagram of a word generated as a result of document analysis in a folder and a second distribution pattern of a document according to the present embodiment, and a folder generated according to the word;

FIG. 28 is an explanatory diagram of a word generated as a result of analyzing a document in a folder and a third distribution pattern of a document according to the present embodiment, and a folder generated according to the word;

[Explanation of symbols]

Reference numeral 100 ... Document collection server system, 102 ... Document collection processing, 106 ... Fitness calculation, 107 ... Document storage processing, 10
8 ... Folder management processing, 110 ... Client request processing.

Claims (28)

[Claims] 1. A method for automatically classifying information by a computer, which corresponds to one of a plurality of folders associated with each other in a hierarchical relationship, and one or more words to be searched respectively. Storing a plurality of search conditions that specify the search conditions, detecting the matching degree between the search conditions stored corresponding to each folder and the document information to be classified, and detecting the correspondence between each folder and the information. Based on the matching degree and the hierarchical relationship, one or more folders in which the document information should be registered are determined from among the plurality of folders, and one or more folders corresponding to the determined one or more folders are determined. An information classification method for storing the document information. 2. The detection detects the degree of conformity between the document information and each of the words designated by the search condition stored corresponding to each folder, and is stored corresponding to each folder. 2. The document information according to claim 1, further comprising the step of determining a sum of matching degrees detected between each of the words designated by the search condition and the document information as a matching degree between the folder and the document information. Classification method. 3. The step of determining the matching degree between the document information and each of one or a plurality of words included in a search condition stored in correspondence with each folder includes the step of determining the matching degree in the document information. 3. The document information classification method according to claim 2, further comprising a step of determining a product of a weight of the word and a weight of the word in the search condition as a matching degree between the word and the document information. 4. The weight in the document information of each of one or more words included in the search condition stored corresponding to each folder is proportional to the number of appearances of the word in the document information. The document information classification method according to claim 3, which is a value to be set. 5. The weight in each of the one or a plurality of words included in the search condition stored corresponding to each folder in the search condition is the weight of the word in the plurality of words in the search condition. 5. The document information classification method according to claim 3, which is a value proportional to the number of appearances. 6. The determination is performed by detecting one or a plurality of folders matching the document information according to the matching degree detected between the document information and each folder, and detecting the plurality of folders by the detection. Is detected, one or a plurality of folders corresponding to the document information in the detected plurality of folders is set to the relative position in the hierarchical relationship between the detected plurality of folders. The document information classification method according to claim 1, wherein the selection is performed depending on the relationship. 7. The selection is representative of a group of folders when the plurality of folders detected as being compatible with the document information include a group of folders having a relatively upper and lower relationship. 7. The document information classification method according to claim 6, further comprising the step of selecting one folder as a folder to which the document information is associated. 8. The selection is performed when the detected plurality of folders include another group of folders that is not in a relative hierarchical relationship with the group of folders. 8. The document information classification method according to claim 7, further comprising the step of detecting one folder representing a folder as another folder corresponding to the document information. 9. The document information classification method according to claim 7, wherein one folder representing the group of folders is a folder located at the lowest level of the group of folders. 10. A method for automatically classifying document information by a computer, which corresponds to one of a plurality of folders mutually associated in a hierarchical relationship, each of which should be one or a plurality of searches. A plurality of search conditions for designating a word are stored, and based on the search condition stored corresponding to each folder and a predetermined judgment criterion, the document information to be classified is associated as a folder of the plurality of folders. Determine one or more, store the document information corresponding to the determined folder, perform the above determination and storage for each of the plurality of document information to be classified, and store the document information corresponding to each folder. It is determined whether a plurality of document information satisfy a predetermined condition defined for the reconstruction of the folder, and when any one of the folders satisfies the predetermined condition, one of them Document information classification method comprising the step of reconfiguring correspondingly stored search condition to a plurality of document information and the folder stored to correspond to the folder. 11. The document information classification method according to claim 10, wherein the predetermined condition is that the total number of registered document information in the one folder exceeds a predetermined value. 12. The reconstructing step divides the search condition stored corresponding to the folder into a plurality of new search conditions each designating a part of a plurality of words designated by the search condition. , Dividing a plurality of document information registered in the one folder into a plurality of document information groups, replacing the one folder with a plurality of new folders, and corresponding to each of the plurality of new folders, 11. The document according to claim 10, further comprising a step of storing one or a plurality of words designated by one of the new plurality of search conditions and a part of the document information obtained by dividing the plurality of document information. Information classification method. 13. The step of dividing the plurality of pieces of document information includes a plurality of pieces of document information that conforms to a plurality of new search conditions obtained by dividing the search information stored corresponding to the one folder. 13. The document information classification method according to claim 12, comprising the step of dividing the document information into the document information subgroups. 14. The step of reconstructing comprises reconstructing a part of a plurality of document information registered in the one folder and a part of a plurality of words specified by a search condition stored corresponding to the folder. Select, arrange at least one new folder in the lower hierarchy of the one folder, and correspond to the new folder, the selected part of the words and the selected part of the document information 11. The document information classification method according to claim 10, further comprising the step of storing. 15. The step of selecting a part of the plurality of pieces of document information and the part of the words includes storing the plurality of pieces of document information in association with the one folder. Part of the word group that is searchable, but some other words in the word group cannot be searched, and some of the document information and the other part of the word The method further comprises a step of separating the document information from other document information that can be searched by any of the above, and selecting a part of the document information obtained by the separation and the part of the words used for the separation.
Document information classification method described in 4. 16. The step of reconfiguring creates a new folder, and registers a plurality of pieces of document information registered in the one folder and another folder in a duplicated manner in correspondence with the new folder. Then, the registration of the plurality of overlapping documents in the one folder and the other one folder is deleted, and the word groups stored in correspondence with the one folder and the other one folder are deleted. Of the other word groups stored as
11. The document information classification method according to claim 10, further comprising a step of storing a word group for searching the plurality of overlapping documents in association with the new folder. 17. The new folder is arranged in the same hierarchy as the one folder and the other folder.
Document information classification method described in 6. 18. Determining one or a plurality of folders for associating the document information with the document information to be classified based on a word group designated by a search condition stored corresponding to each folder. The degree of matching between each folder is detected, and based on the detected degree of matching between each folder and the document information and the hierarchical relationship, the plurality of folders are registered as the folders for registering the document information. 11. The document information classification method according to claim 10, further comprising the step of determining one or a plurality of folders. 19. A method for automatically classifying document information by a computer, each of which corresponds to one of a plurality of folders which are associated with each other in a hierarchical relationship, and one or more of each of which should be searched. A plurality of search conditions for designating a word are stored, and based on the search condition stored corresponding to each folder and a predetermined determination criterion, the document information to be classified is associated as a folder, One or more are determined, the document information is stored corresponding to the determined folder, and the above determination and storage are performed for each of the plurality of document information to be classified. It is determined whether a plurality of pieces of document information stored corresponding to a plurality of folders of a copy satisfy a predetermined condition defined for reconstruction of the plurality of folders, F When the folder satisfies the predetermined condition, a plurality of document information stored corresponding to the some folders and a plurality of search conditions stored corresponding to the some folders are displayed. A document information classification method having a step of reconstructing. 20. The predetermined condition includes a condition in which the some of the plurality of folders relate to an upper folder and a lower folder of the folder, and the reconfiguration corresponds to the lower folder. The document information group registered by the above and the document information group registered corresponding to the folder on the upper side are redistributed to the folder on the upper side and the folder on the lower side, and after the redistribution, , Based on the new document information group registered corresponding to the lower folder and the new document information group registered in the upper folder, registered corresponding to the lower folder 20. The document information classification method according to claim 19, further comprising the step of redistributing the document information group and the document information group registered corresponding to the upper folder to the upper folder and the lower folder. . 21. The predetermined condition relates to a relative size of the number of pieces of document information registered corresponding to the lower folder and the number of document information registered corresponding to the upper folder. The document information classification method according to claim 19, which is a condition. 22. The condition is that the number of pieces of document information registered corresponding to the lower folder is smaller than the number of document information registered corresponding to the upper folder. 21. Document information classification method described in 21. 23. The determination of one or a plurality of folders for associating the document information is performed between the document information to be classified and each folder based on a search condition stored corresponding to each folder. Of the plurality of folders is detected as a folder for associating the document information based on the detected degree of matching between each folder and the document information and the hierarchical relationship. 20. The document information classification method according to claim 19, further comprising a determining step. 24. In a computer system having a storage device for holding a database and a computer for selectively retrieving document information specified by the user from the computer, a user-specified computer system associated with each other in a hierarchical relationship. Corresponds to one of multiple folders, stores multiple search conditions that specify one or more words to be searched, and monitors whether there is newly registered document information in the database. When there is newly registered document information, the matching degree between the document information and each search condition is determined, and based on the matching degree between each search condition and the document information and the hierarchical relationship, the document information is searched. Document information collection method for determining one or a plurality of folders to be associated with each other. 25. It is determined whether or not each of the plurality of folders satisfies a condition regarding reconstruction of a folder related to a plurality of document information stored corresponding to the folder, and any one of the folders is determined. When the reconstruction condition is satisfied, a new folder is generated, a new search condition for searching at least a part of the plurality of document information stored corresponding to the one folder is generated, and the new search condition is generated. 3. The method further comprising the step of storing different search conditions and the part of the document information in association with the new folder.
4. Document information collection method described in 4. 26. It is determined whether or not any of the folders satisfies a division condition for dividing a plurality of pieces of document information stored corresponding to the folder into a plurality of new folders. When the folder satisfies the division condition, a plurality of new folders are generated, and the plurality of document information is divided into a plurality of groups based on the plurality of document information and the search condition stored corresponding to the one folder. Determining a plurality of search conditions for the document information, and selecting one of the plurality of groups of document information obtained by the division and one of the plurality of groups of search document information obtained by the division. 25. The document information collecting method according to claim 24, further comprising a step of storing the document information of 1. and the document information corresponding to one of the new plurality of folders. 27. It is determined whether or not a set of a plurality of folders satisfies a condition for separating and storing a plurality of pieces of document information that are correspondingly stored in duplicate, and a set of the plurality of folders is determined. When any of the folders in the set satisfies the separation condition, a new folder is generated, and a search condition for searching a plurality of pieces of document information redundantly stored in those folders is set to any one of the set of folders. Generated based on the search conditions stored corresponding to each other, store the generated search conditions corresponding to the new folder, and correspond the plurality of duplicate document information to the new folder 2 further comprising the step of re-memorizing
4. Document information collection method described in 4. 28. A first computer having means for storing a database containing document information to be provided to a user, and a second computer for communicating with the first computer to retrieve the document information in the database. And a user-operable terminal connected to the second computer, the terminal having names of a plurality of folders designated by the user, which are related to each other in a hierarchical relationship, And a means for sending to the second computer a plurality of search conditions including a plurality of words designated by the user corresponding to each folder, and the second computer has a means for sending the plurality of sent folders to the second computer. Means for storing the name and search condition, means for communicating with the first computer to detect whether or not there is document information newly registered in the database, and when there is newly registered document information , Its document information And the degree of matching of the plurality of search conditions with each other, and based on the degree of matching between each search condition and the document information and the hierarchical relationship, determine one or a plurality of folders to which the document information is associated, Means for storing the document information and its name corresponding to each of the determined folders, and a plurality of names of the plurality of folders and a plurality of folders stored corresponding to the respective folders in response to a request from the terminal. And a means for sending the name of the document information of the folder to the terminal, the terminal displaying a plurality of folders having the names of the plurality of sent folders in a mode in which the hierarchical relationship can be identified, A document information collecting system further comprising means for displaying the names of the plurality of sent document information in association with folders corresponding to the respective document information.