JP2001290843A - Device and method for document retrieval, document retrieving program, and recording medium having the same program recorded - Google Patents

Abstract

PROBLEM TO BE SOLVED: To automatically retrieve important document data. SOLUTION: This document retrieval device which retrieves document data from a document data group having link relation is equipped with a link importance imparting unit 21 and an access part 25. The link importance imparting unit 21 weights the link relation of document data 30 and imparts link importance 31 being a degree of importance to the document data 30 and the access part 25 accesses the document data 30 according to the link importance 31.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document retrieval apparatus for retrieving a large group of document files stored in an information processing apparatus based on the contents of the document, the link relation of the document, the registered location of the document, and the like. The present invention relates to a method and a recording medium.

[0002]

2. Description of the Related Art Today, with the development of computer networks, a large amount of online document information is overflowing. In order to search and organize such a large amount of online document information, a service for indexing online document information is provided.

For example, the Internet WWW (Wor
In ld Wide Web page search, one famous service provides a directory service. This is a collection of links to WWW pages for each hierarchical category, and has the following advantages. 1. By simply selecting (clicking) a category, a link to a WWW page that the user wants to browse can be obtained. 2. Since the WWW pages are organized for each category, a great deal of information is not searched for at the time of document search. 3. Since the WWW page is registered in the category through manual operation, garbage, that is, information irrelevant to the information that the user wants to browse, is less likely to enter.

Because of these advantages, this service is very widely used on the Internet. But,
Since this service manually creates categories and manages WWW pages, there is a problem that operation costs are high.

[0005] In automation of the entire directory service, there are the following problems to be solved. 1. Selection of important documents 2. Management of category hierarchy, for example, addition and deletion of topics Assignment of categories from documents: automatic classification Here, selection of important documents will be described. The number of WWW pages is increasing on both the Internet and the intranet. Therefore, similar information may be separately created in various places, and a search based on the presence or absence of a character string (keyword) included in a document may result in too many results. As a result, there is a problem that it is not possible to determine which information is important among the many search results. The following measures have been taken as solutions to the selection of important documents. 1. The search results are sorted in an order satisfying the search request, that is, the search results are sorted and ranked based on the number of search keywords included in the document. 2. Access support is performed by visualizing the search results, that is, a group of documents of the search results are grouped together with similar contents (clustering). 3. The search results are sorted based on document attributes, for example, size, creation date and time. 4. In addition, the search results are sorted in the order of importance of the documents given by some means, for example, based on metadata such as link relationships, user access log analysis, and ratings created by a third party organization, and the like, and the search results are sorted.

As a final example, in recent years, document importance assignment using a link relationship of a hypertext such as a WWW page has become an important technology at a research and service level. Here, the simplest realization of the link importance assignment based on the link relationship is based on the intuition that “a document having a large number of linked pages has a high importance”.

[0007]

However, in order to facilitate the navigation of information, W in the same server is generally used.
WW pages are often linked to each other. For example, in a personal WWW page or the like, there are many links to a home page top such as "return to the top of XX". Therefore, there is a problem in that the server (site) or individual having a large number of pages has a high importance simply because the number of linked pages is merely large. Also, if you know that the search system gives importance to the number of linked pages, you can intentionally increase the importance of your page by adding meaningless page divisions or pages with only useless links. There is also a problem.

[0008] Also, http: //www.elseviernl/cas/tree/s
In the WWW page that can be browsed at tore / comnet / free / www7 / 02 / com02.htm, in addition to the intuition that "documents with a large number of linked pages have high importance", "links from documents with high importance" The intuition suggests that the linked document has a high degree of importance "and that" a page linked from a page with a small link destination has a high degree of importance. "

The second intuition is based on the discovery that the former is more important between a WWW page introduced in a famous directory service and a WWW page introduced in an unnamed personal link collection. is there. A third intuition is based on the idea that documents linked from a link collection with 50 links are more important than documents linked from a link collection with 1000 links, for example. . In the importance judgment algorithm based on these intuitions, first, the provisional importance is calculated based on the number of links, and the importance is updated based on the link relation from the provisional importance.
.. is repeated until the operation converges.

However, even with this algorithm, a site having a large number of pages is advantageous because the number of links is large, and as a result, when the importance of a page is calculated, it is similar to the top of the page with high importance. There is the above-mentioned problem that pages of such sites are lined up.

When a user performs a search, an interface for accessing a keyword used for the search is required. As a conventional technique for accessing a keyword, it is closely related to an interface of kana-kanji conversion.

For example, according to Japanese Patent Application Laid-Open No. 03-241456, kana-kanji conversion is realized by a touch panel type device. According to this technique, all the readings of the keywords are entered by a software keyboard on the screen, and then the characters are converted into kanji characters by pressing a "convert" key.

Further, for example, Japanese Patent Application Laid-Open No. H10-15414
4, JP 10-154033 and http: //www.csl.son
According to the WWW page that can be browsed at y.co.jp/person/masui/POBox/index.htm, a pen-oriented text input system is realized. According to this technique, reading is input from a software keyboard on the screen. When a part of the reading is input, candidates are output as needed based on the usage history of the user.

JP-A-03-241456, JP-A-10-104
154144, JP-A-10-154033 and the aforementioned W
According to the WW page, it is necessary to always input the reading and spelling of keywords one by one in order to convert kana-kanji characters. Therefore, there is also a trivial reading input interface for inputting a long character string, for example. As an example of this technique, there is a method in which a keyword list is first created for each of the first characters such as "A", "I", and so on, and a user selects from the keyword list. However, in this case, when the number of keywords in the list starting from a specific reading is large, it becomes difficult for the user to select a specific keyword from this keyword list. An example of this method is a bank transfer machine.

As a further example of a self-explanatory reading input interface technology, reading characters are sequentially input (clicking in the case of a pointing device, etc.), and kana-kanji converted characters are displayed when the number of keywords starting from the reading is determined. There is also a method to do it. FIG. 32 shows a method of sequentially inputting the reading characters and displaying the Kana-Kanji converted characters when the number of keywords is determined. FIG. 32 shows an example in which “Akihabara” is displayed. As shown in FIG.
The user sequentially inputs the reading of the keyword from the zero sound. When displaying "Akihabara", the user first inputs "A", and then sequentially inputs "ki", "ha", "ba", and "ra". When all the sounds of "Akihabara" have been input and the number of keywords has been determined, the kana-kanji converted character "Akihabara" is displayed. Even with this method,
You need to input a lot of sounds until you reach a long-reading keyword.

The present invention solves the above-mentioned problem in the selection of important documents, that is, the problem that the importance is biased toward a specific site having many pages, and allows a malicious individual to control the importance. It is an object of the present invention to provide a document search apparatus and a method for making document search difficult.

Further, according to the present invention, when a search keyword is input based on the reading of a keyword, the keyword can be reached with less input, and the number of candidate keywords and documents to be displayed is set to a certain number or less. By
An object of the present invention is to provide a document search apparatus and method that facilitate selection of keywords and documents.

Still another object of the present invention is to provide an apparatus and a method for generating a link collection having a directory service-like interface and allowing quick access to important documents related to keywords, for example, WWW pages.

[0019]

According to the present invention, there is provided a document retrieval apparatus for retrieving document data from a document data group having a link relationship, wherein a link importance which is a weight given to the link relationship is determined. A link importance assigning means for assigning to the document data and an access means for accessing the document data based on the link importance are provided.

[0020] Pages linked from many pages are considered important, and links from pages that link to few pages are more important than links from pages that link to many pages. The link importance assigning means calculates the link importance by assigning weight to the link relation based on what is considered to be important, etc., and assigns it to the document data. The access means accesses the document data based on the calculated link importance. This makes it possible to automatically search for important document data.

Further, the link importance assigning means includes a URL
A similarity calculating means may be provided. The URL similarity calculating means calculates a URL (Uniform R) indicating the document data.
U, which is the similarity between source locators.
RL similarity is calculated, and the link importance assigning means calculates the URL
The link importance is calculated using the similarity between them and the link relation between the documents, and is added to the document data.

For example, document data in the same site often has a link relationship. In general, the URLs of the document data in the same site have a high URL similarity. By making the link weight from the document data having a high URL similarity lower than the link weight from the document data having a low URL similarity, the importance of a site having a large amount of document data is overestimated. prevent. After all, more accurate
It is possible to search for important document data. Also,
In order to consider the URL similarity when assigning the link importance, a third party may intentionally increase the importance of a specific document by increasing the number of links of the document data in the site. There is also an effect that it becomes difficult. In addition,
The URL similarity may be determined based on URL face information including a server address, a path, and a file name.

Further, the document search device may further comprise a keyword extracting means for extracting a keyword from the document data. In addition, the document search device further includes a document relevance calculating unit, wherein the keyword extracting unit calculates the frequency of appearance of the extracted keyword in the document data, and the document relevance calculating unit calculates the link importance and the keyword. The degree of association between the keyword and the document data is calculated using the appearance frequency.

The relevance is calculated using the link importance and the appearance frequency of the keyword in the document, and the document data having a high relevance is searched. It is possible to search for important documents that are likely to be present.

Further, the document search device further comprises a monitor for monitoring access from a user and creating an access log. The keyword document relevance calculating means includes an access log in addition to a keyword appearance frequency and a link importance. May be used to calculate the degree of association. By using the access log when calculating the degree of relevance, it is possible to search for document data that is more relevant to the keyword and has higher importance.

Since the link importance, the keyword appearance frequency, and the access log are used for calculating the relevance, even if the importance of specific document data is intentionally increased, such document data is difficult to be searched. There is also an effect.

Further, the document search device further comprises a document type discriminating means for discriminating the document type of the document data by using the URL similarity and the number of links and the number of links of the document data. Document data may be selected based on the document type, and the degree of relevance may be calculated for the selected document data.

There are several types of document data, such as link pages and content pages. These can be determined based on the number of links and the number of links. Based on this document type, certain types of document data, for example,
By selecting a content page and calculating the degree of relevance of the selected document data, it is possible to search more accurately.

[0029] The document search device may further include index generation means for generating an index that can access the keyword from the reading or spelling of the extracted keyword. further,
The document search device may further include a selection unit that allows a user to select a reading or spelling fragment of the keyword from the index. Then, the index generating means generates the index,
Document data of a certain number or less is searched for in descending order of the calculated relevance. The access means accesses the document data based on the reading and spelling fragments of the selected keyword. The user can easily select from the index because the document data included in the index is limited to a certain number. Further, since the document data included in the index is limited to a certain number, it is useful for a mobile terminal such as a mobile phone when the size of the display screen is limited.

Further, the document search device may further include a collection unit for collecting document data from the network. Further, according to the present invention, in a link collection generation system having a link relationship, a collection unit, a link importance assignment unit,
A URL character string determining unit and an index generating unit are provided.

The collecting means collects the document data from the network, the link importance assigning means assigns the link importance calculated using the link relation of the document data to the document data, and the URL character string discriminating means assigns the URL character string from the document data. A URL having a characteristic on a specific character string is determined. The index generation means includes:
A link collection in which links to the document data are arranged in order within a certain number based on the link importance and the characteristic of the URL in a specific character string is generated. A characteristic of a specific character string of the URL of the document data may indicate the content of the document data. For example, a character string such as JAVA or Java may be used for the URL of document data related to JAVA (registered trademark). Therefore, the features on the specific character string of the URL can be used for estimating the contents of the document data. Therefore, by creating a link collection to the document data based on the link importance and the characteristic of the specific character string of the URL, a link that allows the user to search for the document data including the content desired to be viewed. The collection can be automatically generated.

Further, the link collection generating system may further include a document type discriminating means. The document type determining means determines the document type of the document data as described above, and the index generating means selects the document data based on the document type, and specifies the link importance and URL of the selected document data. Generate a link collection based on the features on the character string. As a result, a collection of links to higher-quality document data can be generated.

The scope of the present invention also includes a method comprising a procedure of a process realized by the above-described apparatus. Also,
The scope of the present invention also includes a recording medium that records a program that allows the computer to execute the above-described processing.

[0034]

Embodiments of the present invention will be described below in detail with reference to the drawings. FIG. 1 shows a configuration of a document search device according to the first embodiment. 1 includes a processing device 11, an input device 12, and a display device 13. The processing device 11 is, for example, a CPU (Centra).
l Processing Unit) and a memory, the input device 12 corresponds to a keyboard, a mouse and the like, and the display device 13 corresponds to a display and the like.

The processing apparatus 11 includes a link importance assigning unit 21, a keyword extracting unit 22, a keyword document relevance calculating unit 23, an index generating unit 24, an index accessing unit 25, and an access analyzing unit 26. These correspond to, for example, software components described by a program, and are stored in a specific program code segment of the processing device 11.

The link importance assigning unit 21 is, for example, a WWW.
Link information is extracted from document data 30 such as a page. For example, in the case of a WWW page, HTML is analyzed and <a
href = “http://www.fujitsu.co.jp/”> Top Fujitsu </
Take out the anchor (a) tag part like a>. Then, based on the extracted link information, the link importance level 31
Is calculated, and the calculated link importance 31 is output to the keyword document relevance calculator 23. Further, the link importance assigning unit 21 calculates a URL similarity calculator 2 that calculates a URL similarity which is a similarity between the face of the link source and the URL of the link destination.
7 is provided. The URL similarity calculator 27 calculates the URL similarity between the link source and the link destination and calculates the link similarity calculator 21.
Calculates the link importance 31 based on the extracted link relationship and the URL similarity.

The keyword extractor 22 extracts a keyword from the document data 30, and outputs the result as a page keyword 3
Output as 2. Further, the keyword extractor 22 may totalize all occurrence frequencies of the extracted keywords in the document data 30. Keyword extractor 2
2, for example, when the document data 30 is described in Japanese, morphological analysis (word segmentation) is performed, and nouns (columns)
Is extracted as a keyword. In addition, the fluctuation of simple notation (such as "computer" and "computer") should be unified with rules and small rules. The synonym information is provided, for example, externally by a dictionary or the like.

The keyword document relevance calculator 23 calculates a keyword document relevance, which is a relevancy between a keyword and a document, based on the link importance 31, the page keyword 32, and an access log 34 described later, and indexes the calculation result. Output to the generator 24.

The index generator 24 generates index data 33 based on the keyword document relevance, and outputs the generated index data 33 to the index access unit 25. The index data 33 is generated using, for example, hypertext.

The index access unit 25 displays the contents of the index data 33 on the display device 13 and outputs information indicating the access status of the user to the access analyzer 26 in accordance with a user's instruction input from the input device 12. I do.

The access analyzer 26 analyzes information indicating the access status of the user, creates an access log 34 in which documents accessed by each keyword within a certain period are created, and uses the created access log 34 as a keyword. Output to the document relevance calculator 23.

Next, the data structure of various data will be described with reference to FIGS. FIG. 2 shows a table collection for storing document information. The table collection for storing document information includes a document information table 41 and a referenced document table 42. The document information table 41 includes a document ID, a UR
L, a title, a link to the referenced document table 42, and a link importance of the document data. The document ID is a numerical value uniquely assigned to the document data, the URL is information indicating the document data on the Internet, the title is the title of the document data, and the referenced document table 42 is the document data of the link source document data linked to the document data. Stores a set. The referenced document table 42 stores the document ID and pair information of the URL similarity, which is the similarity of the character face on the URL between the link destination and the link source document data. At most one per document
There are three referenced document tables. Document information table 41
And the referenced document table 42 are stored in the link importance assigning device 2.
1 corresponds to the link importance 31 created.

FIG. 3 shows a table collection for storing keyword information. The table collection for storing the keyword information includes a keyword table 51, a keyword correspondence table 52, and an appearance document table 53. Keyword table 51
Consists of a keyword ID, a representative word, and a link to the appearing document table 53. The representative word is information indicating which one is representative when there are a plurality of keywords having the same keyword ID. The keyword correspondence table 52 is
It consists of pair information of an actual keyword, reading (or spelling), and keyword ID. Here, the same keyword ID (kwID) is assigned to keywords representing the same concept (for example, “computer”, “computer”, “computer”, etc.). In addition, when reading Japanese keywords, the notation is unified by removing "-" in the reading and changing the repellent sounds like "ぁ" and "や" to "A" and "I". Keep it. For English keywords, capitalize them. As a result, it is possible to prevent keywords indicating the same concept from being treated differently due to variations in notations such as “computer” and “computer”, and to unify keywords in a generated index.
The appearance document table 53 includes a document ID of the document data including the keyword and an appearance frequency of the keyword.
The keyword correspondence table 52 and the representative words of the keyword table are provided in the processing device 11 in advance (not shown). The appearance document table 53 corresponds to the page keyword 32 created by the keyword extractor 22.

FIG. 4 shows a table collection for storing index information. The table collection for storing index information includes an index information table 61, a related document table 62, and a related keyword table 63. The index information table 61 includes an index character string, a sequence of continued character strings, and a related keyword string. In the index information table 61, the index data generation unit 24 generates a character string graph by a method described later based on the keywords stored in the keyword correspondence table 52 in FIG. Is created by degenerating. In FIG. 4, for example,
The top is "A""I" ... When "A" is selected, "Aibo", "Ao", ... appear next.
In addition, the keyword that can be considered from the character string “Aibo” indicates that “Aibo” and “Ivory”. Here, the keyword is a keyword included in the keyword correspondence table 52 of FIG. The related document table 62 is
It is a table for obtaining a related document ID which is an ID of a related document from a keyword ID. The keyword document relevance calculator 23 calculates the document relevance, and based on the calculation result, stores the related document ID strings within a certain number in the descending order of the document relevance in the related document table 62. The related keyword table 63 is a table for obtaining a related keyword ID from a document ID. The information in the related document table 62 and the related keyword table 63 are the same,
It is transposed. The detailed information of the related document ID is stored in the document information table 41 of FIG.

FIG. 5 shows an access log which is a table for storing access log information. The access log 71 is
Stores information on when the user selects document data from a keyword information screen (described later), that is, access date and time, keyword ID, pair information of the document ID of the selected document data, and the like. The access log 71 is stored in the access analyzer 26.
Corresponds to the access log 34 created. By counting logs within a certain period, it is possible to obtain how many times a specific document has been accessed from a specific keyword.

Hereinafter, the operation of the entire document search apparatus will be described with reference to FIG. FIG. 6 shows the procedure of the index generation process. First, the link importance level assigner 21 extracts link information, a URL, and the like from the document data. Subsequently, the extracted information is stored in the document ID and URL fields of the document information table 41, and a link (pointer) to the referenced document table 42 and the referenced document table 42 are generated (step S1).

UR provided in the link importance assigning unit 21
The L similarity calculator 27 calculates the extracted link information and UR
Based on L, the URL similarity between the link source and the link destination is calculated and stored in the field of the referenced document table 42.

Next, the link importance assigning unit 21 calculates the link importance based on the extracted link information and the URL similarity, and stores it in the field of the document information table 41 (step S2). The calculation of the URL similarity and the link importance will be described later.

The keyword extractor 22 extracts keywords from document data and stores them in the keyword and keyword ID fields of the keyword correspondence table 52, all fields of the keyword table 51, and the document ID and frequency fields of the appearing document table 53. (Step S
3). The keyword is extracted, for example, when the document data 30 is described in Japanese,
Performs morphological analysis (word segmentation) and performs the noun (sequence) obtained by word segmentation. Also, the fluctuation of simple notation (such as "computer" and "computer") should be standardized in rules and small dictionaries. Synonym information is provided externally by a dictionary or the like.

Next, the keyword extractor 22 reads the extracted keywords based on the unified rules described above, and stores the result in the reading field (or spelling) of the keyword correspondence table 52. (Step S4). In the keyword correspondence table 52, keywords are unified and stored, so that the keywords of the generated index can be unified.

Next, the keyword extractor 22 counts all the appearance frequencies of the extracted keywords in the document data, generates a pointer to the appearing document in the keyword table 51 based on the result, and generates an appearing document table 53. Is stored in the document ID and frequency fields (step S5). Furthermore, the keyword extractor 22
The total appearance frequency of the keyword IDs is totaled, a fixed number of keywords (for example, 10,000 words) are determined as target keywords for the index, and entries other than the keyword ID are excluded from the keyword table 51 and the keyword correspondence table 52.

Next, the keyword document relevance calculator 23
Calculates the keyword document relevance indicating the relevance between the keyword and the document based on the link importance of the document information table 41, the URL similarity of the referenced document table 42, and the access log 71, and calculates the keyword document relevance from the top. A predetermined number of documents are determined as related documents, and information is stored in the related document ID column fields of the related document table 62 and the related keyword table 63 based on the determination (step S6).

Next, the index generator 24 generates a character string graph based on the entry keywords in the keyword correspondence table 52 and their readings (or spellings), degenerates them, and stores the result in the index information table 61. It is stored (step S7). The degeneration will be described later.

Next, the index generator 24 creates an index based on the index information table, the related document table, the related keyword table and the document information table (step S).
8). The index is created by, for example, hypertext. The created index may be displayed on the display device 13.

The generated index is output to the display device 13 via the index access unit 25, and the user inputs using the displayed index. The index access unit 25 outputs information indicating the access status of the user to the access analyzer 26. The access analyzer 26 analyzes information indicating the access status and creates an access log 34 (not shown).

The procedure by which the link importance assigning device of the document retrieval apparatus calculates the link importance will be described below. First, in the present embodiment, when assigning the link importance, the link importance assigning device 21 uses the link relation, the URL, and the keyword of the document data. The importance of the document data determined based on the link relationship is referred to as link importance. The basic concept for determining the link importance is as follows. 1. Document data (pages) linked from URLs with low similarity are important.

For example, generally, a plurality of WWW pages (pages) provided in the same site are linked to other pages in the site, but the URLs of those pages are linked.
Are similar to each other. Therefore, it is possible to estimate that the importance of a page linked from a URL having a high similarity is low. 2. Pages that are linked from many pages are important pages, and are linked from important pages.
Pages with low URL similarity are important.

For example, famous directory services and government offices are linked from many pages, and documents linked from such important pages are more likely to be pages opened by individuals and entries of their contents. This is because it is considered that the importance is higher than the document linked from the page. In addition, many pages and pages provided in a service (site) having a mirror site are often linked in the site. Therefore, in the related art, there is a problem that many pages of the same site are searched. However, since the URLs of the pages in the site are almost the same, for example, in the same domain, it is possible to solve this problem by introducing the idea that "pages with low URL similarity are important". Becomes 3. The similarity of the URL is the smallest when the server address, path, and file name are all different, and the URL is similar so that the mirror site or the page in the same server has the higher similarity.
Defined from L face information.

By introducing the above three concepts, all link relationships are not treated equally, and a weight corresponding to the link importance is given to the link relationships. More specifically, the link weight is given as the reciprocal of the similarity between the URLs of the link source and link destination documents.
This is a problem in the prior art where the importance of a document is simply determined based on the number of linked pages and all links are treated equally, that is, a server (site) or an individual having a large number of pages or mirror sites can be used. It is possible to solve the problem that the importance is high simply because the amount is large. Also, even if the number of pages within the site is intentionally increased and links are set to increase the importance of the page, since the URL similarity of the page of the same site is high, increasing the importance of the page is more difficult than before. Also has the effect of becoming difficult.

Hereinafter, the calculation of the link importance by the link importance assigning device 21 will be described in more detail. The set of pages for which the link importance is calculated is DOC = ｛p ₁ ,
p ₂ , .... p _N ｝, the link importance of page p is W _p , the set of linked pages of page p is Ref (p), the set of linked pages of page p is Refed (p), the page URL of p and q
Assuming that the similarity is sim (p, q) and the difference is diff (p, q) = 1 / sim (p, q), when it is assumed that a link is provided from page p to q, the link The weight lw (p, q) is defined by the following equation (1).

[0061]

(Equation 3)

As can be seen from the equation (1), lw (p, q)
Is larger as the similarity sim (p, q) of the URLs of p and q is lower and the number of links from p is smaller. The link importance of each page is C _q for each p∈DOC.
As a constant (the lower limit of importance, which may be different for each page)

[0063]

(Equation 4)

Is defined as the solution of the simultaneous linear equation. The link importance assigning unit 21 assigns link importance to each page by solving this simultaneous linear equation.
In addition, since there exist many existing algorithms about the solution of such a simultaneous linear equation, description is abbreviate | omitted. The URL similarity sim (p, q) in the equation (1) is calculated by a URL similarity calculator 27 provided in the link importance assigning unit 21 (described later). From equations (1) and (2),
It can be seen that the above concept has been realized. That is, if the degree of similarity is low from the equation (1), the weight lw becomes large, so that the document page linked from a URL with a low degree of similarity from the equation (2) becomes important. From equation (2), pages linked from more pages are more important pages.

Furthermore, the important page (W
A page with low URL similarity (high link weight lw) linked from q) is important. Hereinafter, FIG.
With reference to FIG. 8 and FIGS. 8A and 8B, the concept represented by Expressions (1) and (2) will be described in more detail.

FIG. 7 shows the contents of the expressions (1) and (2).
Is schematically shown. In FIG. 7, each page is an arrow.
Is link relation, arrow direction is link direction, arrow thickness
Indicates the link weight. As shown in FIG.
p₁, P_TwoAnd p_ThreeLink to page q from
You. Page p₁Is two pages r other than page q₁as well as
r _TwoLink to the page p_ThreeIs the other two pages₁
And s_TwoLinked from.

Here, the URL similarity of each page is as follows: sim (p1, q) = sim (p1, r1) = sim (p2, r1) = 1 1 sim (p2, q) = 2, ie, page p URL of ₂ and q
Sim (p3, q) = 1, sim (s1, p3) = sim (s2, p3) = 3 (that is,
The URLs of pages s ₁ , s ₂ and p ₃ are similar).

When the equations (1) and (2) are applied to the case as shown in FIG. 7, each page p ₁ , p ₂ , p ₃ , s ₁ and s
The weight of link ₂ is as follows.

Lw (p1, q) = 1 / {1 × (1 + 1 + 1)} = 1/3 lw (p2, q) = 1 / {2 × (1/2)} = 1 lw (p3, q) = Thus 1 lw (s1, p3) = lw (s2, p3) = 1/3, (1) and formula weight lw (p1, q) links many pages p ₁ of the destination from the above calculation is that smaller I understand. Similarly, it can be seen from equation (1) and the above calculation that the greater the URL similarity, the smaller the link weight.

The link importance W _q of the page _q is as follows. _{W q = C q + {lw} (p1, q) × W p1 + lw (p2, q) × W p2 + lw (p3, q) × W p3} = C q + {(W p1 / 3) + W p2 + W p3 ＷW _p1 = C _p1 W _p2 = C _p2 W _p3 = C _p3 + ｛lw (s1, p3) × W _s1 + lw (s2, p3) × W _s2 ｝ = C _p3 + (W _s1 + W _s2 ) / 3 , the link importance W _p3 page p ₃ linked from more pages than the page p ₁ and p ₂ is larger. Maximum Also, look at the link importance W _q of the page q, because seems important pages, low URL similarity pages, i.e. link weight lw (p3, q) = 1 , in other words when the link weights (the link weight It can be seen that the greater the value of =), the greater the link importance.
Also, from equation (2) and the above calculation, the weight of a link from a page in the same site with a similar URL can be treated lighter than the weight of a link from a page with a similar URL with another URL. I understand. As a result, it can be understood that the problem that many pages of a site having a large number of pages appear in the search results can be solved.

FIG. 8 schematically shows the contents of the expressions (1) and (2). In FIG. 8 as well, as in FIG. 7, each page, an arrow indicates a link relationship, an arrow direction indicates a link direction, and a thickness of the arrow indicates a link weight. Further, FIG.
The shaded circle in indicates a page having a high URL similarity. 8 (a) and 8 (b), page q
Are linked from pages p ₁ , p ₂ and p ₃ . Further, in FIG. 8B, page q is pages p ₁ , p
The URLs of ₂ and p ₃ are similar, and the URL similarity sim
(pi, q) = n + 1 (n is an integer). Equations (1) and (2) are applied to FIGS. 8A and 8B, respectively. The case of FIG. 8A is as follows.

Link weight of each page lw (pi, q) = 1 / sim (pi, q) = 1 (URLs are dissimilar) Link importance W _q W _q = C _q + (W _p1 + W of page q) _p2 + _Wp3 ) In the case of FIG.

The link weight of each page lw (pi, q) = 1 / sim (pi, q) = 1 / (n + 1) (URL
The link importance of page q W _q W _q = C _q + (W _p1 + W _p2 + W _p3 ) / (n + 1) Therefore, when comparing the calculation results of FIGS. 8A and 8B, the URL is When the similarity sim (p, q) is high, it can be seen that the link importance W _{q of the} page _q is small even if the number of links is large. By introducing URL similarity, it can be understood that the problem that a server (site) having a large number of pages has a high importance simply because the number of pages is large is solved. .

Next, the page p in the equations (1) and (2)
And the URL similarity sim (p, q) between q and q will be described. URL
The similarity is calculated using the URL provided in the link importance level assigning unit 21.
It is calculated by the similarity calculator 27.

Generally, the URL of a page is composed of three types of information: a server address, a path, and a file name.
For example, URL of WWW page, http://www.flab.fujitsu.co.jp/hypertext/news/1999/
product1.html is the server address (www.flab.fujits
u.co.jp), path (hypertext / news / 1999), and file name (product1.html).

Further, the server address is further hierarchized by ".", And becomes wider as it goes backward. For example, if the server address is www.flab.fujitsu.c
o.jp, from behind, Japan (jp), company (co), Fujitsu
(fujitsu), laboratory (flab), machine (www).

In this embodiment, the URL similarity between two given pages p and q is defined by the above three combinations. As similarity sim (p, q), for example, domain similarity sim_domain (p, q) and fusion similarity described below
sim_merge (p, q) can be considered.

The domain similarity sim_domain (p, q) is calculated based on the domain similarity. The domain is the latter half of the server address and represents a company or organization. For servers in the United States ending with .com, .edu, .org, etc., from the end of the server address to the second, and for servers in other countries ending with .jp, .fr, etc., from the end of the server address Up to the third corresponds to a domain. For example, the domain of www.fujitsu.com is fujitsu.
com and the domain of www.flab.fujitsu.co.jp is fuji
tsu.co.jp.

The domain similarity between page p and page q is defined by the following equation (3). sim_domain (p, q) = 1 / α (when p and q are in the same domain) = 1 (when p and q are in different domains) (3) where α is a constant and from 0 It is assumed that a real value that is greatly smaller than 1 is taken. Figure 9 shows about 3 collected from the Internet.
Domain similarity sim_do to link relation between million URLs
Here is a case where the concept of main (p, q) is introduced to calculate the link importance. In FIG. 9, the horizontal axis indicates the number of top pages in descending order of link importance, and the vertical axis indicates the number of pages having different domains included in the top pages. Α = 0.1, 0.2, 0.
Cases of 3, 0.5, 0.7 and 1.0 are shown. As shown in FIG. 9, the number of pages having different domains included in the top 100,000 pages having high link importance is α = 1.
(Corresponding to the conventional case where the URL similarity is not introduced) is 4000, and when α = 0.1, 5500
Case. Therefore, it can be seen that as α decreases, pages having different domains rank higher in link importance. This is because the smaller α is, the more similarity sim_doma
in (p, q) increases and the URL similarity sim_domain (p,
q) is larger, the link weight lw (p, q) is smaller and, consequently, the link importance W _q is smaller.
This is because the higher the RL similarity is, the smaller the link importance is given to the page. That is, sim_domain
It can be seen that by introducing the concept of (p, q), pages having different domains are easily searched, in other words, pages having the same domain are hard to be searched.

As sim (p, q), a similarity sim_merge (p, q) obtained by fusing the above three types of information is defined as follows. sim_merge (p, q) = (similarity of server address) + (similarity of path) + (similarity of file name) Hereinafter, a method of calculating each term on the right side will be described.

As for the similarity of the server address, the hierarchy of the address is viewed from the rear, and when the levels match up to n levels, the similarity is set to 1 + n. For example, www.fujitsu.co.jp and ww
w.flab.fujitsu.co.jp is 4 because it matches up to 3 levels. www.fujitsu.co.jp and www.fujitsu.com are 1
Since the levels do not match (match 0 level), the similarity is 1.

The path similarity is compared for each element delimited by "/" in the path from the head, and the degree of similarity up to the matching level is determined. For example, /doc/patent/index.html and / doc / paten
Since t / 1999/2 / file.html matches up to two levels, the similarity is 2.

The file name similarity is set to 1 when the file names match. The above is based on the following concept. 1. Often, similar documents are placed in the same directory, so that URLs with the same path on the same server often have similar contents. 2. Mirror sites provided to distribute access have a high degree of similarity. These differ only in the server address portion, and the remaining paths and file names are often the same. 3. URLs having different server addresses, paths, and file names have low similarity.

The URL can also be obtained by this sim_merge (p, q).
Can be prevented from being searched for pages similar to. Therefore, by introducing the concept of sim (p, q) or diff (p, q) into lw (p, q), only servers (sites) and individuals with a large number of pages have a large amount Thus, the conventional problem that the link importance is high can be solved.

Note that the above-mentioned link importance W _p is also used for calculating the degree of association described later. Hereinafter, a procedure in which the keyword document relevance calculator 23 of the document search device calculates the relevance will be described.

First, in creating an index from a keyword to a document, the degree of relevance between the keyword and the document is required. The relevance is considered as follows. 1. A document that includes many keywords, that is, a document that has a high degree of keyword, has a high degree of relevance. 2. The higher the importance of a document, the greater the relevance. 3. It is desirable that the number of related documents of a certain keyword is not more than a certain number. (It's bad for one keyword to get 1,000 related documents.)

In the present embodiment, the following concept is introduced in addition to the above concept in order to keep the number of related documents of a certain keyword within a certain number. 4. Relevance based on user access log analysis: The more frequently a document is accessed from the keyword within a certain period, the higher the relevance. 5. Relevance based on link importance of documents: Among documents including the keyword, those having high link importance have high relevance.

By introducing the above idea, the relevance of the page p with respect to a certain keyword w is given by the following equation (4). Rel (p, w) = TF (p, w) * log _Wp * log (AC (p, w) +2) (4) where TF (p, w) is the number of occurrences of w in p , W _p is the link importance of _p and corresponds to W _p in the above equation (2).

AC (p, w) is the number of times that the keyword w was accessed from the keyword w during a certain period (for example, one month or one week). For each keyword, a certain number from the largest value of Rel (p, w) is set as a related page of the keyword.

In addition to the degree including the keyword, the URL
The link importance W _P and the user's access log analysis that introduce the concept of similarity are used for calculating the relevance. As a result, conditions for increasing the relevance of the page are increased, so that it is more difficult to intentionally manipulate the page contents to increase the relevance of a certain page.

Hereinafter, an index generated by the index generator 24, that is, an interface for selecting a keyword for search will be described. The selection interface according to the present embodiment is an interface that allows the user to reach the keyword by simply clicking the reading fragments of the keyword sequentially, and has the following features. On one screen, a fragment (character or character string) of the keyword reading and a part of the keyword corresponding to the keyword reading selected so far are displayed. -The user can click on the reading fragment (character or character string) of the keyword on the screen one after another to reach the keyword.・ Keywords displayed on one screen can be limited to a certain number.

In the above-described prior art, one "character" is always clicked to select the reading, but in the present embodiment, it is not necessarily a character but may be a "character string". Thereby, it is possible to reduce useless clicks until reaching the keyword. Further, since the number of keywords displayed on one screen is within a certain number, it is ready to select a keyword. The fact that the number of keywords displayed on one screen is within a certain number is useful for keyword selection in a mobile terminal such as a mobile phone in which the size of the screen of the display device is limited. To achieve these, the index generator 24 performs the following. 1. Unified keyword reading (or spelling). In some cases, the removal of "-" (long sound) and unsounding sounds such as "ぃ" are unified to "a" and "i". 2. From the keywords and their readings (or spellings), a directed graph (character string graph) is created with the reading characters as nodes and the keyword set as leaves. 3. Perform the following degeneration operation in the above graph (see below).

(A) Degeneration of a path to a leaf (b) Removal of an intermediate path (c) Renormalizing the keyword of a child node into the parent node and deleting the child node Create a reading input interface based on the reduced graph.

Hereinafter, a description will be given of how the index generator 24 performs the following steps for creating a keyword character string graph. The keyword character string graph is a graph with a directed label expressing the reading of a keyword. FIG.
Shows an example of a keyword character string graph.

The keyword character string graph is represented by a set of six (N, C, KW, t, nk, yomi). Here, N is a set of nodes, C is a set of kana characters, KW is a set of keywords, and t is N * C ⁺ → N node transition function. Also, C ⁺ is a label, that is, an arrangement of one or more kana characters (indicated by a solid arrow in the character string graph of FIG. 9A and the like), and nk is a keyword assigned to the N → W ⁺ node (FIG. ami etc. are indicated by dotted lines) yomi is “reading” of N → C ⁺ node.

Taking FIG. 10A as an example, the following is obtained. Note that yomi is self-evident and will be omitted.) N = ｛Top, “A”, “Ai”, “Aibo”, “Aibo”, “Aibori”, “Ao”, “Aozo”, “Aozora”｝ C = ｛'A',. . . , 'N''KW = ｛' blue ',' blue ',' blue sky ',' ivory 'ｔ t (top, a) = “a”, t (a, i) = “a”,
t (Aio) = "Ao", t (Aibo) = "Aibo", t (Aibo, U) = "Aibo", t (Aibo,
Ri) = "aibori", t (blue, zo) = "blue", t
(Aozora) = "Aozora", nk (Aibou) = ｛'Aibo'｝ nk (Aibori) = ｛'Ivory'｝ nk (Ao) = ｛'blue' blue 'ｎ NK (Aozora) = ｛'Blue sky' 索引 index generator 24, given keyword and reading,
Based on them, an initial keyword character string graph is generated. FIG. 11 shows a procedure for generating an initial keyword character string graph. A procedure for generating the initial keyword character string graph in the index generator 24 will be described with reference to FIG. FIG.
2 shows an example of an algorithm for realizing a procedure for generating an initial keyword character string graph.

First, a set KW of keywords is created (step S11). Next, it is determined whether or not the created set KW is empty (step S12). If the inside of the set KW is empty (step S12, yes), there is no need to create a character string, and the process ends. If the inside of the set KW is not empty (step S12, no), the process proceeds to the next step.

Next, a certain keyword u is extracted from the set KW (step S13). And this keyword u
Is set, and a node nk {yomi (u)} of the read yomi (u) is set, and the node nk {yomi (u)} is added as a terminal node (step S14).

It is determined whether or not the processing of step S14 has been repeated for the character string length of the keyword u, that is, whether or not the keyword u has become empty (step S15). When the keyword u becomes empty (step S15, yes), the necessary processing for the keyword u has been completed, so the process returns to step S12, another keyword is taken from the set KW, and the procedure from step S13 is repeated. If the keyword u is not empty (step S15, no), the last character of the keyword u is cut out (step S16). Subsequently, the node is shifted to the immediately preceding parent node and added (step S17). Furthermore, attention is paid to the character before the cut-off last character of the keyword u (step S18), and the process returns to step S15.

According to this procedure, a keyword list assigned to a node is obtained as a set nk, and a list of lower nodes located below a certain node is obtained as t. FIG. 10A shows an initial keyword character string graph created by the above procedure. FIG. 10A is an initial keyword character string graph generated from the keywords Ao: Ao, Blue: Ao, Aozora: Aozora, Aibo: Aibo, Ivory: Aibo, and reading data. Similarly, FIG.
In the init_kw_graph () algorithm shown in FIG. 12, @KW = {blue, blue, blue sky, buddy, ivory} yomi blue = blue, yomi blue = blue, yomi blue =
Aozora, yomi ｛buddy｝ = aibo, yomi ｛ivory｝ = aibo.

After creating the initial keyword character string graph, the index generator 24 degenerates the character string. Hereinafter, degeneration of a character string will be described. Degeneration is: 1. Degeneration of intermediate nodes It consists of two kinds of operations, renormalization of terminal nodes.

First, the degeneracy process of the intermediate node in the index generator 24 will be described. FIG. 13 shows the procedure of the degeneration process of the intermediate node. Hereinafter, the procedure of the intermediate node degeneration processing will be described with reference to FIG. In addition, FIG.
4 shows an example of an algorithm for realizing an intermediate node degeneration process.

First, a set N of nodes is created (step S21). Subsequently, it is determined whether or not the inside of the set N is empty (step S22). If the set N is empty (step S
22, yes), since there is no need to degenerate the nodes,
The process ends. If the set N is not empty (step S
22, no), and acquires the node n in the set N (step S23). For the acquired node n, the node n
It is determined whether or not there is only one successor node following, and there is no keyword in node n (step S24). If the two conditions are satisfied (step S24, yes), node n is deleted from the keyword character string graph because node n can be degenerated, and the process returns to step S22 (step S25). If both of the two conditions are not satisfied (step S24, no), the process returns to step S22 without deleting the node n because the node n cannot be degenerated.

As described above, in the keyword character string graph, an intermediate node satisfying the two conditions of “there is no keyword assigned to the node” and “only one successor node (child node)” is degenerated. I have to do that. In the initial keyword character string graph of FIG. 10A, the nodes “Ai” and “Aozo” are “there is no keyword assigned to the node” and “there is only one successor node (child node)”. Intermediate node that satisfies these two conditions. FIG. 10B shows FIG.
The result of degenerating the intermediate nodes of the initial keyword character string graph shown in FIG. In FIG. 10B, the intermediate nodes “Ai” and “Aozo” have been deleted. Similarly, in the proc_shrink_middle () algorithm shown in FIG. 14, t 、 ″｝ = a + t あ a ｛= aibo + ao + t ＋ aibo ｛= aibo + aibori + t ｛aoi｝ = Blue + nk {blue} = partner + nk blue = ivory + nk blue = blue + blue + blue blue = blue sky +

Next, the degeneracy process of the terminal node in the index generator 24 will be described. FIG. 15 shows the procedure of the terminal node degeneration processing. Hereinafter, the procedure of the terminal node degeneration processing will be described with reference to FIG. In addition, FIG.
5 shows an example of an algorithm for realizing a terminal node degeneration process.

First, a set N of all nodes is created (step S31). Next, the nodes are sorted in the order of the number of keywords belonging to the node (step S32).
The integer i is set to 1 (step S33). Next, it is determined whether the integer i is smaller than the number of nodes in the set N (step S34). If the integer i is not less than the number of nodes in the set N (step S34, no),
Further, it is determined whether or not the terminal node is degenerated (step S35). If there is no degeneracy of the terminal node (step S35, no), the process ends. If there is degeneration of the terminal node (step s35, yes), step S33
Return to

If the integer i is smaller than the number of nodes in the set N (step S34, yes), the ith node n of the set N is obtained (step S36). It is determined whether or not the acquired node n is a terminal node (step S
37). If the acquired node n is a terminal node (step S37, yes), the process proceeds to step S38.
If the acquired node n is not a terminal node (step S37, no), the node i is not a node to be degenerated, so the integer i is incremented by one (step S4).
1) Return to step S34.

Next, the parent node p of the node n is obtained (step S38). Next, it is determined whether or not the sum of the number of keywords belonging to the parent node p and the number of keywords belonging to the child node n exceeds a predetermined upper limit of the number of keywords (step S39).

If the sum of the number of keywords belonging to the parent node p and the number of keywords belonging to the child node n does not exceed the upper limit (step S39, yes), the child node n is deleted (degenerated), and The belonging keyword is set as the keyword belonging to the parent node p (step S40). Subsequently, the integer i is incremented by one (step S4).
1) Return to step S34.

When the sum of the number of keywords belonging to the parent node p and the number of keywords belonging to the child node n exceeds the upper limit, if the child node n is degenerated, the number of keywords of the parent node becomes too large. The process proceeds to step S41 without degenerating the node n.

By moving the keyword information at the terminal node to the parent node immediately above, the depth of the tree (chain of nodes) is reduced, so that the user can reach the keyword with few clicks. Become. However, if you transfer too many child node keywords to the parent node,
Since a large amount of keywords are assigned to one node, it becomes troublesome for the user to select a keyword from among them. To work around this problem,
Given a parameter words_max, the number of keywords belonging to one node is less than the parameter words_max.

FIG. 10 (c) shows the result of the terminal node degeneration processing of the keyword character string graph of FIG. 10 (b) when the parameter words_max = 4. In FIG. 10B, the terminal nodes “Aibo” and “Aibo” each have only one keyword. The parent node "Aibo" of the terminal nodes "Aibo" and "Aibo" has two child nodes and has no keyword. Therefore, the sum of the keywords of the parent node “Aibo” and the child nodes “Aibo” and “Aibo” is less than words_max = 4. Therefore,
Since the child nodes “Aibo” and “Aiboro” can be degenerated, the child nodes “Aibo” and “Aibo” are deleted in FIG. 10C, and the keywords of these child nodes are the parent node “Aibo”. Has been moved. Similarly, in the proc_shrink_leaf () algorithm shown in FIG. 16, t｝ ″｝ = a + t ｛a｝ = aibo + ao + t ｛aoi｝ = aozo + nk ｛ai｝ = a partner + ivory + Nk {blue} = blue + blue + nk {blue} = blue sky +

FIG. 17 shows another example of a keyword character string graph that has undergone terminal node degeneration processing. In FIG. 17, the parent node “Kaise” has three terminal nodes “Kaisei”, “Kaisei”, and “Kaisei”. Since the keywords belonging to the terminal nodes “kaisei” and “kaisei” are “clear” and “analysis”, respectively, the terminal nodes “kaisei” and “kaisei” can be degenerated. Further, since the keywords belonging to the terminal node “Kaisesetsu” are two, “commentary” and “open”, the terminal node “Kaisesetsu” can also be reduced. Therefore, in the case of FIG. 17, two methods of degeneration are conceivable. However, the former can reduce the total number of nodes. In the present embodiment, the terminal nodes are sorted in the order of the number of keywords belonging to the nodes. This enables efficient degeneration as in the former case.

Next, an example of an index created by the index generator will be described with reference to FIGS. FIG. 18 shows the transition of the index screen from the index top screen to the branch page through the index intermediate screen and the keyword information screen. The transition of the index screen displayed on the display device will be described with reference to FIG. In FIG. 18, first, an index top screen is displayed. When the user selects the key word reading or the first part of the spelling from the index top screen, the screen transitions to the index intermediate screen. When the user selects the next part of the reading or spelling of the keyword from the index intermediate screen, the screen transitions to a further index intermediate screen. As a result of repeating this selection, the user arrives at the keyword desired to be searched, and when the user selects the keyword, the screen is switched to a keyword information screen.
When the user selects another keyword, the screen transitions to a further keyword information screen. When the user selects the title of the document page that he / she wants to view, the user links to the document page and transits to the document page. The user's selection operation can be performed by clicking with a mouse or pointing with a pen or the like. Each screen may be created by, for example, hypertext.

FIG. 19 shows an example of the index top screen. On the index top screen, characters (strings) below the top of the index information table 61 shown in FIG. 4 are displayed. In FIG. 19, the Japanese syllabary, the alphabet, and the numbers from 0 to 9 are displayed. The user can go to the next screen by clicking on the reading or spelling of the keyword to be searched.

FIG. 20 shows a further example of the index top screen. In FIG. 20, as a result of unifying notation and / or degeneracy of nodes when reading keywords, for example, "ぢ" and "Zu" in the "Da" line of the Japanese syllabary are removed from the index. I understand. Similarly, it can be seen that the letters “Y” and “Z” have been removed.

FIG. 21 shows an example of the index intermediate screen. FIG.
Reference numeral 1 denotes an index intermediate screen when “A” is selected on the index top screen. The upper half shows the character string that follows "a", and the lower half shows other keywords. The index intermediate screen includes an index information table 61 of FIG. 4 and a keyword correspondence table 51 of FIG.
Keyword ID).

In FIG. 21, character strings following "a" include "wart", "e", "o", and the like. For example,
When "Ibo" is selected, it moves to "Aibo". Other keywords such as "love" and "love dog"
Within 20) All keywords that start with the displayed "A" and cannot be read continuously are displayed here. Conversely, if no keyword is displayed and there is no continuous reading, it is understood that the keyword is not included in the index.

FIG. 22 shows a further example of the index intermediate screen. FIG. 22 is an index intermediate screen when “i” is selected on the index top screen. The upper half shows the character string that follows "i", and the lower half shows other keywords. In FIG. 22, a character string that follows "A" includes "wart, ero" and the like. As other keywords, "ion", "enable", and the like are displayed in a fixed number (for example, within 20).

FIG. 23 shows a further example of the index intermediate screen. FIG. 23 shows an index intermediate screen when “event” is selected. Since there is no child node of the node "event", a character string continuing after "event" is not displayed, but a keyword is displayed. The user selects a keyword to be input from this screen. It can be seen that the keywords including "event" not included in this screen are not included in the index.

In the prior art, since only one character can be selected for reading, it is necessary to repeat the selection operation many times to input a long character string. However, according to the present embodiment, it is not always necessary to select a continuous character string one by one in order to degenerate a node. For example, FIG.
It is also possible to select two characters at a time, such as the character string "wart" of FIG. Therefore, the number of times the user selects can be reduced.

Furthermore, all keywords for which continuous reading cannot be selected are displayed here. Conversely, if the keyword is not displayed and there is no continuous reading, the keyword is not included in the index, so the user only selects one character at a time and performs the last character. There will be no problem that the keyword to be input is found not to be included in the index.

Further, when the terminal node is degenerated,
Since a keyword having a value greater than the value set as the parameter word_max is not displayed, it is relatively easy for the user to search for a desired keyword from the displayed keywords when selecting a keyword. In addition, the fact that the displayed keywords are limited to a certain number or less is useful in keyword selection in a mobile terminal such as a mobile phone having a limited display screen size.

Further, as a search interface, a set of fixed keywords can be reached with as little labor as possible, and the following points are different from the kana-kanji conversion in the prior art.・ Conversion key operation is not required.
It is only necessary to provide the minimum information that can specify the keyword.

Therefore, for example, in the keyword set,
If only "Knowledge management" is included as a word starting with "Nare", "Knowledge management" is displayed only when the user specifies "Nana" or "Re".

FIG. 24 shows an example of the keyword information screen. The keyword information screen when the keyword “compression” is clicked on the intermediate screen is shown. In FIG.
First, a representative word and a synonym ("compression""Compress") are displayed. This can be obtained from the information in the keyword table 51 and the keyword correspondence table 52 in FIG. “A” in the upper right of the screen indicates the path that has been followed. Thus, it is possible to prevent the user from going back or getting lost in the hypertext. The screen displays the title, link information, and other keywords of each document. Since a certain number (for example, 20 documents) is displayed in descending order of priority, it is not difficult for the user to select a document from among them. The list of document IDs is shown in FIG.
From the related document table 62. Information on each document ID is contained in the document information table 41 of FIG. Other keywords are obtained from the related keyword table 63 of FIG. When the user selects the document information desired to be viewed, the user can link to the document and display the document on the screen.

FIG. 25 shows a further example of the keyword information screen. The keyword information screen when the keyword “event calendar” is clicked on the intermediate screen is shown. "Top"-"I"-"Event" at the upper right of the screen indicates the path that has been followed. By selecting each path, the user can return to the previously viewed screen.

FIG. 26 shows the configuration of a network document search device according to the second embodiment. 26, in addition to the configuration of the first embodiment shown in FIG. 1, a collection device 81 and a synonym dictionary 82 are further provided. The collection device 81 is, for example, a Web robot that collects a large amount of documents from an intranet (and / or the Internet). The synonym dictionary (synonym data) 82 stores a part of the information of the synonym correspondence table. The input device and the output device may be, for example, the WWW browser 83.

The collection device 81 automatically collects documents from the network. The keyword extractor 22 of the processing device 11 uses the synonym dictionary 82 to extract keywords from the collected pages, and determines the frequency of occurrence of keywords in the pages. Tally. That is, a document having high importance is automatically selected using the synonym dictionary 82. This allows the intranet (and
Or the Internet).

FIG. 27 shows the configuration of a network document search apparatus for a specific type document according to the third embodiment. FIG.
In addition to the configuration of the second embodiment shown in FIG.
A document type discriminator 91 is provided in the processing device 11. The document type determiner 91 determines the document type of the document data based on the link relationship and the URL of the document data collected from the intranet (and / or the Internet). More specifically, the document type discriminator 91 calculates the URL similarity calculated in the link importance assigning device 21 and the number of links / links indicated by the link relationship extracted by the link importance assigning device 21 based on the URL similarity. Determine the content type of the document data (without understanding its contents). The document type discrimination in the document type discriminator 91 is performed as follows. 1. A page having a certain number or more of links to document data whose URL similarity is equal to or less than a certain value (low) is a link collection. 2. A page having a certain number or more links to document data having a URL similarity of a certain value or more (high) is a menu (entry).
Page. 3. A page referred to by a certain number or more from document data having a URL similarity of a certain value or more (high) is a menu (entry) page. 4. In addition to the above, a page in which the link to the document data whose URL similarity is equal to or higher than a certain value (high) is a certain number or less is a content page.

By making such a determination, the document type determiner 91 can classify the document type (eg, menu page, link collection, content page) of the document data (WWW page) with a considerable probability.

The document type discriminator 91 discriminates the document type, and outputs the discriminated document type 92 to the keyword document relevance calculator 23. The keyword document relevance calculator 23 selects document data of a specific type based on the determined document type 92, and calculates the document relevance of the selected document data based on the link importance, page keyword, and access log. I do. For example, the keyword document relevance calculator 23 may select document data whose document type is determined to be a content page and calculate the relevance of these content pages.

As described above, the network document search apparatus shown in FIG. 27 restricts the pages to be included in the index to any of the types determined by the document type discriminator obtained here, thereby achieving higher quality document organization. It is possible to do.

FIG. 28 shows the configuration of a link collection generation system according to the fourth embodiment. 28, the link collection generation system includes a collection device 101, a processing device 102, and an input / output device 107. The collection device 101 is, for example, a Web robot that collects a large amount of document data from the Internet (or / and an intranet). Processing device 1
02 includes a link importance assigning unit 21, a URL character string discriminator 103, an index generator 24, and a WWW server 106. The link importance assigning unit 21 calculates the link importance of the document based on the URL similarity and the link relationship, and outputs the calculated link importance 31 to the index generator 24.

The URL character string discriminator 103 discriminates the contents of the collected document data (without understanding the contents) based on the characteristics of the URL in the character string. The URL character string discriminator 103 discriminates the content of the document data, for example, from the characteristics of the URL in the character string as follows. 1. Document data in which the character string of the URL includes Y2K, y2k, year2000 is a page related to the year 2000 problem. 2. The document data that includes news, release, and press in the character string of the URL, and thereafter has a numeric string (often date and time information) is a news (press) release page. 3. Pages whose URLs include java and JAVA are Java-related pages. 4. A page including the download, dwnload, and dwnld in the URL character string is a page related to software download. 5. Pages that include LINUX, linux, or Linux in the URL string are Linux-related pages.

By making such a determination, the URL character string determiner 103 determines a specific URL and outputs the determined specific URL set 104 to the index generator 24. The index generator 24 determines the specific U based on the link importance 31.
The RL sets 104 are arranged in descending order of link importance, a certain number is taken out from the top, a link collection is created from the top URLs, and output to the WWW server 106 as the link collection 105. If the number of URLs obtained by the character string discriminator 103 is small, the number of links may be increased by adding another page to which the link source frequently refers (links) by checking the link relationship. This is because similar pages on the WWW are often referred to simultaneously from similar link collections. The WWW server 106 provides the link collection to the user, and the user displays the link collection via the WWW browser 107 and gives an input instruction.

As described above, based on the URL character string, the content is determined without looking at the content of the document page, and a link collection is created using the determination result. As a result, it is possible to easily create a high-quality link collection according to the content of the content.

FIG. 29 shows the configuration of a link collection generation system according to the fifth embodiment. The link collection generation system of FIG. 29 is configured to further include a document type discriminator 111 in the link collection generation system of FIG. The function and operation of the document type discriminator 111 are the same as those of the document type discriminator 91 of the document search device according to the third embodiment described with reference to FIG.

The collection device 101 is connected to the Internet (or
And the intranet), the collection device 81 collects a large amount of document data, the link importance assigning unit 21 calculates the link importance of the document based on the similarity of the URL and the link relationship, and converts the link importance 31 into an index generation period. 24. The URL character string discriminator 103 discriminates a specific URL based on the characteristics of the URL in a character string, and outputs the determined specific URL set 104 to the index generator 24. The document type discriminator 111 discriminates the document type of the document data (without looking at the content) based on the URL similarity and the number of links / links, and outputs the discriminated document type 112 to the index generator 24.
Output to

[0140] The index generator 24 selects document data of a specific document type from the specific URL set 104 based on the document type 112. Subsequently, based on the link importance 31 of the selected document data, the selected document data is arranged in descending order of the link importance, a certain number is taken out from the top, a link collection is created from the upper URL, and the link collection is created. The data is output to the WWW server 106 as 105. WWW
The server 106 provides the link collection to the user, and the user
A link collection is displayed via the WW browser 107, and an input instruction is given.

Thus, it is possible to easily create a high-quality link collection according to the content. By the way, the document search apparatus shown in FIGS.
The link collection generation systems 8 and 29 can be configured using an information processing device (computer) as shown in FIG. The information processing apparatus of FIG.
It includes a memory 122, an input device 123, an output device 124, an external storage device 125, a medium drive device 126, and a network connection device 127, which are connected to each other by a bus 128.

The memory 122 is, for example, a ROM (Rea)
d Only Memory), RAM (Random)
Access Memory), and stores programs and data used for processing. CPU121
Performs necessary processing by executing a program using the memory 122.

Each device and each unit constituting each of the document search device shown in FIGS. 1, 26 and 27 and each processing device of the link collection generation system shown in FIGS. 28 and 29 are stored as programs in specific program code segments of the memory 122. You.

The input device 123 is, for example, a keyboard,
It is a pointing device, a touch panel, or the like, and is used for inputting instructions and information from a user. Output device 12
Reference numeral 4 denotes, for example, a display, a printer, or the like, which is used for inquiring a user, outputting a processing result, and the like.

The external storage device 125 is, for example, a magnetic disk device, an optical disk device, a magneto-optical disk device, or the like. The above-described program and data are stored in the external storage device 125, and are stored in the memory 12 if necessary.
2 can also be used.

The medium driving device 126 is a portable recording medium 12
9 to access the recorded contents. Examples of the portable recording medium 129 include a memory card, a floppy (registered trademark) disk, and a CD-ROM (Compact Dis
k Read Only Memory), an optical disk, an optical disk, or any other computer-readable recording medium is used. The above-described program and data can be stored in the portable recording medium 129, and can be used by loading them into the memory 122 as needed.

The network connection device 127 is a LAN
(Local Area Network), WAN
An external device is communicated via an arbitrary network (line) such as (Wide Area Network) and data conversion accompanying the communication is performed. Further, if necessary, the program and data described above can be received from an external device, and loaded into the memory 122 for use.

FIG. 31 shows a computer-readable recording medium capable of supplying a program and data to the information processing apparatus shown in FIG. Portable recording medium 129
The programs and data stored in the external database 130 are loaded into the memory 122. And CPU
121 executes the program using the data and performs necessary processing.

[0149]

According to the present invention, when calculating the importance of a WWW page, not only the link relation but also the similarity of the URL is taken into consideration, so that a specific site having many pages or its mirror site can be calculated. It is possible to calculate the importance by making the importance difficult to be overestimated. This makes it possible to select an important document with higher accuracy.

Further, the importance calculated according to the present invention has a property that a malicious individual does not intentionally operate.
Further, according to the present invention, it is possible to efficiently access a keyword or document data simply by clicking one or more fragments of one or more characters of the reading (or spelling) of a keyword as an index.

Also, in the keyword index, the number of keywords and documents on one screen can be kept within a certain number, so that not only is it easy for the user to select the keywords, but the display area such as a portable terminal is limited. It is also effective for media use.

Further, according to the present invention, the document relevance is calculated based on the keyword appearance frequency in the document data and the importance of the document data, and the links for accessing the document data are arranged in the order of the relevance. As a result, it is possible to create a link collection that enables quick access to high-quality documents related to a specific keyword.

According to the present invention, it is possible to determine the type of a document (menu, link collection, content) using the similarity of URL and the number of references / references. Further, the document data is selected based on the result of the determination of the document type, and the selected document data is combined with the calculation result of the link importance and / or the frequency of appearance of the keyword to enable access to a higher quality document. A link collection can be created.

Further, according to the present invention, it is possible to automatically select document data of a specific field with high accuracy by determining a specific URL. Further, based on the specific URL determined as the above-mentioned link importance, a link collection enabling access to document data in a specific field can be easily and accurately created.

Further, based on the document type determined as described above, the document data of the specific document type is selected from the document data of the specific URL, and the above link important information of the selected document data is selected. By creating a link collection based on the URL, it is possible to create a link collection that enables access to higher-quality document data in a specific field.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a document search device according to a first embodiment.

FIG. 2 is a diagram showing a collection of tables for storing document information.

FIG. 3 is a diagram showing a collection of tables for storing keyword information.

FIG. 4 is a diagram showing a table collection for storing index information.

FIG. 5 is a diagram showing an access log.

FIG. 6 is a flowchart illustrating a procedure of an index generation process.

FIG. 7 is a diagram (part 1) schematically illustrating a calculation performed by the link importance level assigning device.

FIG. 8 is a diagram (part 2) schematically illustrating the calculation performed by the link importance assigning device.

FIG. 9 is a diagram showing the result of introducing the concept of URL similarity when calculating link importance.

FIG. 10 is a diagram illustrating an example of a keyword character string graph.

FIG. 11 is a flowchart illustrating a procedure for generating an initial keyword character string graph.

FIG. 12 is a diagram showing an example of an algorithm for realizing a procedure for generating an initial keyword character string graph.

FIG. 13 is a flowchart illustrating a procedure of an intermediate node degeneration process.

FIG. 14 is a diagram illustrating an example of an algorithm for realizing a procedure of an intermediate node degeneration process.

FIG. 15 is a flowchart illustrating a procedure of a terminal node degeneration process.

FIG. 16 is a diagram illustrating an example of an algorithm for implementing a procedure of a degeneracy process of a terminal node.

FIG. 17 is a diagram illustrating an example of a keyword character string graph in which a terminal node has been degenerated.

FIG. 18 is a diagram showing transition of an index screen.

FIG. 19 is a diagram illustrating an example of an index top screen.

FIG. 20 is a diagram showing a further example of an index top screen.

FIG. 21 is a diagram showing an example of an index intermediate screen.

FIG. 22 is a diagram showing a further example of an index intermediate screen.

FIG. 23 is a diagram showing a further example of an index intermediate screen.

FIG. 24 is a diagram showing an example of a keyword information screen.

FIG. 25 is a diagram showing a further example of a keyword information screen.

FIG. 26 is a configuration diagram of a document search device according to a second embodiment.

FIG. 27 is a configuration diagram of a document search device according to a third embodiment.

FIG. 28 is a configuration diagram of a link collection generation system according to a fourth embodiment.

FIG. 29 is a configuration diagram of a link collection generation system according to a fifth embodiment.

FIG. 30 is a configuration diagram of an information processing apparatus.

FIG. 31 is a diagram showing a recording medium.

FIG. 32 is a diagram illustrating an example of a trivial reading input interface.

[Explanation of symbols]

11, 102 processing device 12, 123 input device 13 display device 21 link importance imparting device 22 keyword extractor 23 keyword document relevance calculator 24 index generator 25 index access unit 26 access analyzer 27 URL similarity calculator 30 document Data 31 Link importance 32 Page keyword 33 Index data 34 Access log 41 Document information table 42 Referenced document table 51 Keyword table 52 Keyword correspondence table 53 Appearance document table 61 Index information table 62 Related document table 63 Related keyword table 71 Access log 81 , 101 Collection robot 82 Synonym data 83, 107 WWW browser 91, 111 Document type discriminator 92, 112 Document type 103 URL character string discriminator 104 Specific URL 05 Links 106 WWW server 121 CPU 122 memory 124 output device 125 external storage device 126 medium drive 127 bus 128 a network connection device 129 portable recording medium 130 program data _{p, p 1, p 2,} p 3, q, r 1 , R ₂ , s ₁ , s ₂
Page S Step

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) G06F 17/30 350 G06F 17/30 350C 12/00 546 12/00 546B

Claims (33)

[Claims] 1. A document retrieval apparatus for retrieving document data from a document data group having a link relationship, wherein the link relationship is weighted to give a link importance which is an importance to the document data. A document search apparatus comprising: an assigning unit; and an access unit that accesses the document data based on the link importance. 2. The method according to claim 1, wherein the link importance assigning unit assigns a URL (Uniform Resource) pointing to the document data.
(URL Locator), and a URL similarity calculating unit that calculates a URL similarity that is a similarity between the URLs, and calculates the link importance using the similarity between the URLs and the link relationship between the document data. The document search device according to claim 1, wherein: 3. The document search apparatus according to claim 1, further comprising a keyword extracting unit for extracting a keyword from the document data. 4. The keyword extracting means calculates an appearance frequency of the keyword in the document data, and determines a relevance between the keyword and the document data using the link importance and the appearance frequency of the keyword. The document search apparatus according to claim 3, further comprising a keyword document relevance calculating means for calculating. 5. A monitor for monitoring access from a user and creating an access log, wherein the keyword document relevance calculating means uses the keyword appearance frequency, the link importance, and the access log. The document search device according to claim 4, wherein the relevance is calculated. 6. A document type discriminating unit for discriminating a document type of the document data using the URL similarity and the number of links and the number of links of the document data. The document search apparatus according to claim 4, wherein the document data is selected based on a document type, and the relevance is calculated for the selected document data. 7. The document search device according to claim 3, further comprising: an index generation unit configured to generate an index capable of accessing the keyword from reading or spelling of the extracted keyword. 8. The apparatus according to claim 1, further comprising a selection unit that selects a reading or spelling fragment of the keyword from the index, wherein the index generation unit is configured to output the keyword document relevance calculation unit within a predetermined number of the high relevance. The document search device according to claim 7, wherein the document data is put into an index, and the access unit accesses the document data based on a fragment of reading or spelling of the selected keyword. 9. The document search device according to claim 1, further comprising a collection unit that collects the document data from a network. 10. The document search apparatus according to claim 1, wherein the link importance assigning unit reduces the importance of document data that is linked from a URL having a low similarity. 11. The document search apparatus according to claim 1, wherein the link importance assigning unit regards a page linked from important document data having a low URL similarity as important. 12. The document search apparatus according to claim 2, wherein the similarity of the URL is determined based on URL face information including a server address, a path, and a file name. 13. The target document data set is defined as DOC =
{P ₁ , p ₂ ,..., P _N } The importance of the document data p is Wp The page set of the link destination of the document data p is _{Ref ( p)} The page set of the link source of the document data p is _{Refed ( p)} The similarity of the document data p and q on the URL is sim (p, q), the difference is diff (p, q) = 1 / sim (p, q), and the link from the document data p to q is When it is said that it is stretched,
The weight lw (p, q) of the link is defined as follows: The importance of each page is obtained by setting C _q as a constant (for the lower limit of the importance, 2. The document search apparatus according to claim 1, wherein the document is obtained as a solution of the simultaneous linear equation. 14. The method according to claim 1, wherein the document data is WWW (World).
14. The document search device according to claim 1, wherein the document search device is a Wide Web page. 15. A document index generation device that generates an index of a document data group having a link relationship, wherein: a link importance assigning unit that assigns a link importance to the document data using the link relationship; Keyword extraction means for extracting a keyword from data; index generation means for generating an index capable of accessing the keyword from reading or spelling of the extracted keyword; and when the reading or spelling of the keyword is selected from the index. A document index generating apparatus, further comprising: an access unit for accessing a keyword and document data having a high link importance related to the keyword. 16. The link importance assigning means includes a URL (Uniform Reso) pointing to the document data.
and a URL similarity calculating unit that calculates a URL similarity, which is a similarity between the source and the source data. The link importance assigning unit uses the similarity between the URLs and a link relationship between the document data to obtain the link importance. The document index generation device according to claim 15, wherein a degree is calculated. 17. A document index generation device for generating an index of a document data group having a link relationship, wherein a link importance is assigned to the document data based on whether or not the URLs of the document data are similar. Link importance assigning means, keyword extracting means for extracting a keyword from the document data, and index generating means for generating an index capable of accessing the keyword from reading or spelling of the extracted keyword based on the link importance. A document index generation device, comprising: 18. A link collection generation system for generating a link collection to a document data group having a link relationship, comprising: a collection unit for collecting the document data from a network; and an importance calculated using the link relationship. Link importance assigning means for assigning a link importance to the document data, and a UR having a characteristic on a specific character string from the document data group
URL character string discriminating means for discriminating L, and index generation for generating a link collection in which links to the document data are sequentially arranged within a certain number based on the link importance and features of the URL on a specific character string. Means, and a link collection generation system, comprising: 19. The apparatus according to claim 19, further comprising: a document type determination unit configured to determine a document type of the document data using the URL similarity and the number of links and the number of links of the document data. 20. The document data is selected based on the selected document data, and the link collection is generated for the selected document data based on the link importance and a characteristic of the URL on a specific character string. The link collection generation system described in 1. 20. A document retrieval method for retrieving document data from a document data group having a link relationship, wherein a step of assigning a link importance, which is an importance calculated using the link relationship, to the document data; Accessing the document data based on the link importance. 21. A URL (U) indicating the document data.
The method further comprises: calculating a URL similarity, which is a similarity between a plurality of document resources (niform Resource Locators); and calculating the link importance using the URL similarity and a link relationship between documents. The document search method according to claim 20. 22. The document search method according to claim 20, further comprising: extracting a keyword from the document data. 23. A step of calculating an appearance frequency of the keyword in the document data; and a step of calculating an association degree between the keyword and the document data using the link importance and the appearance frequency of the keyword. The document search method according to any one of claims 20 to 22, further comprising: 24. The method further includes: monitoring access from a user to create an access log; and calculating a relevance using the keyword appearance frequency, the link importance, and the access log. 24. The document search method according to claim 23, wherein: 25. A step of determining a document type of the document data using the URL similarity and the number of links and the number of links of the document data; selecting the document data based on the document type; 25. The document search method according to claim 23, further comprising: calculating the degree of relevance of the obtained document data. 26. The document search method according to claim 22, further comprising: generating an index that can access the keyword from reading or spelling of the extracted keyword. 27. A step in which a user selects a fragment of reading or spelling of the keyword from the index, a step of inserting a certain number or less of document data having a high degree of relevance into the index, 27. The method according to claim 26, further comprising: accessing the document data based on a fragment of reading or spelling of a keyword. 28. The method according to claim 20, further comprising a step of collecting the document data from a network. 29. An automatic link collection generation method for automatically generating a link collection to a document data group having a link relationship, comprising: a step of collecting the document data from a network; A process of assigning link importance, and a UR having a specific character string feature from the document data group
L, and generating a link collection in which links to the document data are arranged in order within a certain number based on the link importance and features of the URL in a specific character string. A link collection generation method characterized by the following. 30. A step of determining the document type of the document data using the URL similarity and the number of links and the number of links of the document data; selecting the document data based on the document type; 30. The link collection generation according to claim 29, further comprising: generating the link collection based on the link importance and the specific character string feature of the URL for the obtained document data. Method. 31. A recording medium recording a program for a computer that generates a link collection to a group of document data having a link relationship, wherein: a step of collecting the document data from a network; Assigning a link importance to the document data; and a UR having a specific character string feature from the document data group.
L, and generating a link collection in which links to the document data are arranged in order within a certain number based on the link importance and the characteristic of the URL in a specific character string. And a computer-readable recording medium storing a program for causing the computer to execute the above. 32. A program for causing a computer to perform control for searching for document data from a document data group having a link relationship, wherein a weight is assigned to the link relationship and a link importance, which is an importance, is assigned to the document data. And a step of accessing the document data based on the link importance. A program for causing the computer to execute processing including: 33. A program for causing a computer to control generation of a link collection to a document data group having a link relationship, comprising: collecting the document data from a network; and using the link relationship to generate the document data. Assigning a link importance to the UR, and UR having a characteristic on a specific character string from the document data group
L, and generating a link collection in which links to the document data are arranged in order within a certain number based on the link importance and the characteristic of the URL in a specific character string. For causing the computer to execute.