JPH1074210A - Method and device for supporting document retrieval and document retrieving service using the method and device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a retrieving method for enabling a user to have a look at the whole image of a retrieved document group and to attain retrieval as service. SOLUTION: A feature word displaying means 22 is displayed on a display means 2, a word group characteristically appearing in a document group retrieved by a user's request is extracted, mutual relation among feature words is checked, a graph setting the feature words as nodes is prepared, and the whole image of retrieved results is displayed on the means 22. When a user selects his (or her) interested word or uninterested word while observing the displayed feature word graph, succeeding retrieval strategy can be effectively prepared.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document search support method and apparatus having a user interface for realizing an interactive guidance function in document search, and a document search service using the same.

[0002]

2. Description of the Related Art In document retrieval, various interfaces between a document retrieval apparatus and a user have been devised and developed so that a user can quickly and easily reach a desired document set. The main ones are feedback and guidance. Feedback is a mechanism by which, when a user makes a "hit / miss" determination for some items in a search result, a search result reflecting the determination is obtained. Guidance is a function that provides information that appears to be relevant to the search operation at each stage of the search operation, and thus is useful for users to devise or improve search conditions. .

[0003] Conventionally, a guidance function is generally used.
A method of presenting related information with respect to input search conditions has been performed. For example, there is a method in which a database indicating the relevance between words such as a thesaurus is held, and words related to the word input as a search condition are extracted from the database and presented. In the case of a thesaurus, the data is mainly tree-structured data indicating the superordinate-subordinate relationship between words. However, there is also a method of automatically generating related word data using co-occurrence statistics and using the data (for example, BR Schatz et al. , Inte
ractive term suggestion for users of digital libra
ries: Using subject thesauri and co-occurrence lis
ts for information retrieval. Proc. ACM DL96.p.126
-133). Also, a method of displaying a search word and its related words in a network based on co-occurrence statistical data between words has been proposed (for example, RH Fowler, DW Dearholt, Inf).
ormation Retrieval Using Pathfinder Networks. In P
athfinder Associative Networks, Ablex, article 12,
Edited by RW Schvaneveldt (1990)).

However, in the method of presenting related information with respect to a search condition, it is difficult to cope with a case where a plurality of search words are used or a case where negation is used, and a search for a document that does not use a keyword (associative search). Etc.) are also difficult to deal with. As a method of overcoming this, there is a method of automatically extracting relevant information from a search result and providing it to a user. For example, the scatter-gather method (D. Cutting, etc.)
(1992). Scatter / Gather: A Cluster-based Approach
to Browsing Large Document Collections. Proc. ACM
In SIGIR '92, p.318-329), the retrieved documents are automatically classified (clustered) and characteristic words for each class are displayed. However, in clustering, when the number of documents increases, the computational complexity increases in the order of the square or cube, making it difficult to react in real time. In general, as the search operation progresses, the difference between classes becomes more subtle, There is a problem that it is difficult to grasp the character of the class from the characteristic words of the class.

[0005]

SUMMARY OF THE INVENTION The present invention solves the above-mentioned problem and removes the characteristic words of the words that appear characteristically in the documents so that the topics included in the retrieved documents can be listed in real time. A screen display in a graph format or a list format, and a document search support method and apparatus capable of extracting a word group characteristically appearing in a document group from low-frequency words to high-frequency words in a well-balanced manner. An object of the present invention is to enable a user who wants to search a document to be able to perform the search from a remote place.

[0006]

For this reason, a word group characteristically appearing in a document group is used as a node so that topics included in the searched document group can be listed in real time, and strong words between characteristic words are used. If there is an origin relationship, that is, if the degree of occurrence in the same document is high, a graph is constructed by linking the word pair, and the graph is displayed on the screen. The vertical axis represents the document frequency of the characteristic word so that general words and highly specific words can be distinguished at a glance. In the case of a list, for example, characteristic words are classified by frequency class, and words having a high document frequency are arranged in the upper row so that the words having a high degree of specialty can be identified at a glance. When selecting characteristic words from a group of retrieved documents, in order to extract characteristic words from low-frequency words to high-frequency words in a well-balanced manner, the characteristic words are classified according to the frequency of appearance, and the frequency is calculated from each class. Ratios are extracted in descending order of the ratio of the document frequency in the document group to the document frequency in the entire search target.

[0007]

BEST MODE FOR CARRYING OUT THE INVENTION

Embodiment 1 Hereinafter, a first embodiment of the present invention will be described with reference to FIGS. The present embodiment is an example of a configuration of a search device using an independently used computer. In the present embodiment, a case where a search result is displayed as a graph will be mainly described. FIG. 1 shows the overall configuration of a document search apparatus according to the present embodiment. 1 is an input means, 2
Is a display means, 3 is a CPU, 4 is a calculation program holding means, 5 is a work area for operating the calculation program, and 6 is a database holding means. These means or devices are used for mutually transmitting signals. They are linked by a bus 100 for exchange. The input means 1 includes a keyboard 11, a mouse 12, a pen input means 13, and the like. The display means 2 displays a search interface 21 and a characteristic word display means 22 for guiding a search. The calculation program holding unit 4 stores a search interface operation routine 41, a morphological analysis routine 42, a search routine 43, and a characteristic word display unit operation routine 44 necessary for operating the document search apparatus of this embodiment. The characteristic word display means operation routine 44 extracts characteristic words from the retrieved document group and displays them on the characteristic word display means 22. Therefore, the characteristic word extraction routine 441, the co-occurrence relation analysis routine 442, the graph arrangement routine 443, and the graph The display routine 444 is used as a subroutine. Details of the work area 5 will be described later with reference to FIG. The database holding means 6 includes a document database 61 to be searched, an index database 62 used for search, a database 63 for word frequency, and an exclusion word database 64. These databases are generally prepared in advance,
The user selects and uses data suitable for his / her search purpose as search target data. For example, when searching for newspaper articles, a database issued by a newspaper company is purchased and used. However, the exclusion word database 64 may be provided as ancillary data by the device supplier.

FIG. 2 shows the configuration of the work area 5 in detail. The work area 5 is a calculation program holding means 4
Are areas for holding parameters and temporary data required for the operation of various routines in the above. A search work area 51, a characteristic word extraction work area 52, a co-occurrence relation analysis work area 53, a graph layout work It is composed of an area 54. Each area has a further subdivided data area, the details of which will be described when the respective associated routines operate. When a user attempts to search a document, first, a command for starting the document search system is input from the keyboard 11.
In response, the search interface operation routine 41 is activated, and the search interface 21 for interactively performing the search operation is displayed on the display unit 2. FIG. 3 is an example of an initial screen of the search interface 21. The search interface 21 includes a search request input unit 211, a keyword display / operation unit 212, a hit count display unit 213, a title display unit 214, a document display unit 215, and a search execution button 216.
And a feature word display button 217.

In this embodiment, three types of keywords for document search are used: essential keywords, additional keywords, and deducted keywords. The search is performed by taking an AND of the essential keywords, and if no essential keyword is specified, the search is performed by OR of the additional keywords. One point is added when the document searched for the essential keyword includes the point keyword, and one point is deducted when the document searched for the essential keyword includes the point keyword. Regardless of how many times the same keyword appears in the same document, one point is added or subtracted. The keyword display / operation unit 212 is composed of three parts corresponding to these three types of keywords. Since the configuration is the same for all three, the following description will be focused on the leftmost essential keyword. The essential keyword portion in the keyword display / operation unit 212 includes a keyword display unit 2121, a move button 21211, and a clear button 21212. The move button 21211 is used to transfer another type of keyword to the required keyword, and the clear button 21212 is used to remove the keyword from the required keyword. That is, when a keyword displayed as an additional keyword or a deducted keyword is selected and the move button 21211 for essential keyword is pressed, the selected keyword is moved to the essential keyword. When a keyword displayed as an essential keyword is selected and the clear button 21212 is pressed, the selected keyword is removed from the essential keywords. Similarly, a keyword displayed as a required keyword is selected, and for example, a move button 21
If the user presses 221, the selected keyword is moved to the additional keyword. In addition, select a keyword displayed as a required keyword, and click a deduction keyword move button 2
If the user presses 1231, the selected keyword is moved to the deduction keyword.

[0010] These move buttons are also used as copy buttons for copying the displayed characteristic word to the keyword, as described later. That is, whether to move or copy is used depending on where the target word is located. When inputting a search request, the search request input window 211
Then, a search request such as an essential keyword, an additional keyword, and a deducted keyword is input by using the keyboard 11 to wait for an input by clicking or the like. Subsequently, when the input completion button 2112 is pressed, the character string input to the input window 2111 is passed to the morphological analysis routine 42, divided into word strings, and registered with reference to the excluded word database 64. The result of removing the word is stored in the default keyword storage area 5111 or 5112 (FIG. 2) of the keyword storage area 511.
Here, the default keyword type is a required keyword. Each content is displayed in the form of a list on the keyword display unit 2121 or 2122.
In this case, as will be understood from an example described later, the word division state is determined according to the dictionary of the morphological analysis routine 42.

Here, when a search execution button 216 is pressed, a search routine 43 is started, and a required keyword is ANDed with reference to a search index database 62 (that is, data indicating which document contains a certain word). Are searched, and a row of the obtained document identification numbers is stored in the search result storage area 512. Note that the search routine 43 adds one point to one of the additional keywords when the additional keyword is included in the document searched by the indispensable keyword, and adds one to the additional keyword when the additional keyword is included. The work of deducting one point is performed, and this score is stored in the search result storage area 512 together with the document identification number. If the essential keyword is not specified, the search routine 43 performs a search for the additional keyword by OR of the additional keyword, and calculates a score in the same manner. If there are no required keywords and no additional keywords, search execution button 2
No search is performed even if 16 is pressed.

The essential keyword is processed by AND in the search, so it is indispensable to narrow down the search result more strictly. However, if it is desired to search without any omission in the search result, the additional keyword is added. It is preferable to set a deduction keyword when a search is performed in this OR as only an item, and when it is possible to assume a matter that may include a matter that the user does not want to enter into the search result. Further, the search routine 43 calculates a score distribution from the search results stored in the search result storage area 512, and stores the result in the search result score distribution storage area 513. The score distribution is data indicating how many documents have a score of addition or deduction.

An example in which a search request having "electronic publishing" as a required keyword is input will be described below. After inputting the character string “electronic publishing” as a required keyword in the search request input window 2111, the input completion button 2112 is pressed. “Electronic publishing” is divided into “electronic / publishing” by the morphological analysis routine 42 and the essential keyword storage area 51
11 and a required keyword display unit 2121
Are displayed on the first line and the second line. FIG. 4 shows the state of the search work area 51 when the search execution button 216 is pressed at this stage. In this example, since the essential keywords are “electronic” and “publishing”, they are stored in the essential keyword storage area 5111. Since no other additional keywords or additional keywords are added to the search request input window 2111 by the searcher, the additional keyword storage area 5112 and the additional keyword storage area 5113 are left blank. The searched document number and its score are stored in the search result document number storage area 512. In this case, the points are all 0 because there are no additional keywords and no deduction keywords. Data obtained by counting the number of cases for each score is stored in the search result score distribution storage area 513. In this case, the score was only 0, indicating that there were 77 cases.

FIG. 5 shows the state of the search interface 21 displaying the search result. The required keywords “electronic” and “published” are displayed in the required keyword display unit 2121, the contents of the search result score distribution storage area 513 are displayed in the hit number display unit 213, and the searched document identification number is displayed in the title display unit 214. And an appropriate number of titles are displayed on each line. If the user wants to know the document identification number and its title that are not displayed, the part appearing on the display may be shifted by a so-called scroll bar. If there is something that the user wants to read and see from the displayed title, pointing to the title with a mouse or the like causes a part of the content of the text to be displayed on the document display unit 215. If the user wants to know the contents of the document not displayed, the user can shift the part appearing on the display using the scroll bar in the same manner.

Thus, 77 documents related to “electronic publishing” have been searched. As the next stage, if it is desired to further narrow down the search to a specific target, or if this is not the case, the 77 documents are searched. You may want to get an overview of what topics are included. In such a case, the feature word display button 21 on the search interface 21 (FIG. 3)
When 7 is pressed, the characteristic word display means operation routine 44 is started, and the characteristic word display means 22 is displayed on the display screen 2.

FIG. 6 shows details of an example of the characteristic word display means 22. The characteristic word display means 22 is an operation unit 22
1. Keyword display / operation unit 222, number of hits display unit 223, characteristic word display unit 224, parameter setting unit 22
5 is comprised. The keyword display / operation unit 222 and the number of hits display unit 223 are linked with the keyword display / operation unit 212 and the number of hits display unit 213 of the search interface 21, respectively. Is changed, the respective displays of the search interface 21 are also automatically changed. But in the opposite direction, search interface 2
If the keyword or the number of hits is changed by the above operation, it is not automatically reflected on the characteristic word display means 22. In order to capture the contents, when the reset button 2214 of the operation unit 221 is pressed, the contents of the search interface 21 are copied to the characteristic word display unit 22. By pressing the characteristic word display button 217 on the search interface 21, the keyword on the search interface 21 and the number of hits are automatically copied on the initial screen of the characteristic word display means 22 displayed on the display screen 2. In the case of this example, “Electronic” and “Publishing” are displayed in the essential keyword display section 2221, and “Score 0:77” is displayed in the hit number display section 223.
Cases "are displayed.

Here, when the characteristic word display button 2212 of the operation unit 221 is pressed, a characteristic word extraction routine 45 is started,
The document identification numbers with the highest scores are read from the data stored in the search result storage area 512, the contents of the documents corresponding to the document identification numbers are analyzed, and words (characteristic words) characteristically included in the documents are analyzed. The result of analyzing the relevance between these characteristic words and forming a graph is displayed on the characteristic word display unit 224. The process will be described in detail below. FIG. 7 shows an example of “electronic publishing”, which shows data stored in the graph storage area 543 (FIG. 2). The graph includes nodes and links, and is stored in a node storage area 5431 and a link storage area 5432, respectively. The stored node data is the characteristic word (character string) displayed at each node
And the coordinates indicating where to display it in the characteristic word display unit 224 are the center coordinates, and the number of horizontal and vertical characters of the character display area and the size of the display area. Are half their value, ie, the size from the center to the edge, for ease of use.) On the other hand, the stored link data is composed of start point coordinates and end point coordinates of a line segment to be displayed on the graph. In the figure, in addition to the start point coordinates and the end point coordinates stored in the link storage area 5432, corresponding character string data are added for reference, but in an actual device, this data is unnecessary. FIG. 8 is a diagram illustrating the characteristic word display unit 22 in a state where the characteristic word display button 2212 of the operation unit 221 is pressed and a graph of characteristic words is displayed. The graph display routine 444 determines that the graph storage area 5
According to the data 43, a graph composed of characteristic words and links connecting the characteristic words is expressed in the characteristic word display unit 224. For example, from the data of FIG.
The display area is a rectangular area having 3 characters in the horizontal direction, 2 lines in the horizontal direction, 27 pixels on both sides in the horizontal direction, and 18 pixels vertically in the vertical direction. In this embodiment, starting from the upper left of the characteristic word display unit 224, the coordinates are set to the right in the horizontal direction and to the down in the vertical direction. The link data is defined by the coordinates of the start point and the end point. The first of the link data means connecting the center coordinates of the feature words “publishing” and “electronic”, and the second data is from coordinates (203, 131) to (308, 4).
0) means a line segment. When displaying these words, a white opaque rectangle is displayed on the background of each node for character display in the display area of the node, and in the display area of the node, hiding the line segment is easy to see as a graph, ,
If the line of the graph indicating the link and the display area of the node overlap, the line of the graph does not appear, which may be misleading. For example, in the data of FIG. 7, the line of the graph connecting “Desktop Publishing” and “Publication” passes through the display area of “Nifty Serve”. At this point, the line of the graph does not appear as a line. As a result, "Nifty Serve" and "Publication" are connected by graph lines, and "Nifty Serve" and "Desktop Publishing" appear to be connected by graph lines. In FIG. 8, as a countermeasure, instead of displaying a white opaque rectangle on the background, the graph line is not displayed only near the starting point and the ending point of the graph line and enters the display area of the node. And the other display areas are displayed so as to indicate that they are passing. Even if a white opaque rectangle is attached, it is very difficult to arrange the graph so that the lines of the graph are not hidden. Particularly, if many characteristic words are to be displayed in a graph, it may be impossible to display the characteristic words in an easily viewable size.

The characteristic word display setting means 2251 of the parameter setting section 225 is for adjusting the number of words displayed on the characteristic word display section 224. The setting knob 22511 is moved right and left to set a desired numerical value. Display unit 2251
2, the set value is displayed, and the value is stored in the extracted word number storage area 5213 of the characteristic word extraction parameter storage area 521. This value is used as the characteristic word extraction routine 441.
Utilized by. Hereinafter, a process from when the characteristic word display button 2212 of the characteristic word display unit 22 is pressed to when the graph data as shown in FIG. 7 is created will be described. When the characteristic word display button 2212 is pressed, a characteristic word extraction routine 441 and a co-occurrence relation analysis routine 442 and a graph arrangement routine 443 stored in the calculation program holding unit 4 are sequentially activated. Feature word extraction routine 4
41 reads the highest score and the number thereof from the search result score distribution storage area 513 of the search work area 51. In the example of “electronic” and “publishing” shown in FIG. 4, the highest score (S) is 0.
It is a point and the number (K) is 77 cases. Further, the upper limit (M) 5211 of the number of scanned documents is read from the feature word extraction parameter storage area 521. (Here, it is assumed that M = 300.) Since it takes time to analyze all documents when the number K of retrieved documents is large, a certain limit M
In the case where the value exceeds, the parameter is for extracting M samples.

Next, the characteristic word extraction routine 441 refers to the search result storage area 512, reads the contents of all the document identification numbers whose scores match the highest score S from the search target document database 61, and performs morphological analysis. Words are divided by using the routine 42, and the number of documents in which the words appear in all types of words (hereinafter referred to as document frequency) is counted. In this example, since the number K of the highest scores is 77 and the upper limit value of the number of scanned documents M = 300 or less, all the documents are read. In the morphological analysis of the document, if there is enough space in the database holding means, it is possible to hold the result of morphological analysis of all documents in advance and read it. This eliminates the need to perform morphological analysis each time a search is performed, so that the analysis time can be significantly reduced, which is effective. The word and the document frequency data thus obtained are stored in the frequency data storage area 523 in the feature word extraction work area 52.
Is stored in Since the result of the morphological analysis of the relevant document is used later, the word-divided document storage area 52 is used.
2 is stored.

FIG. 9 shows a part of data stored in the frequency data storage area 523 in the example of “electronic publishing”. The data for each word is composed of five items: word name, document frequency, overall document frequency, frequency ratio, and frequency class. The document frequency is a frequency representing the number of occurrences of the word in the documents (77 in this case) searched in the above operation. The total document frequency is a frequency indicating how many documents are used in the entire search target document regardless of the search result by the keyword. The information is stored in the word frequency database 63, and frequency information of the corresponding word is extracted therefrom. Here, the word frequency database 6
Reference numeral 3 presupposes that all documents to be searched are scanned in advance, and document frequencies of all appearing words are counted and created. The frequency ratio is a value obtained by dividing the document frequency by the entire document frequency. For example, in the first “ROM”, the document frequency is 21 and the entire document frequency is 1183, so the frequency ratio is 21 ÷ 1.
183 ≒ 0.017.

Next, the frequency class will be described. In general, words characteristic of a certain document group can be determined from the magnitude of the frequency ratio, and the higher the frequency ratio, the higher the characteristic level. However, it is dangerous to compare two words having greatly different document frequencies in a frequency ratio. In the case of low-frequency words, since the overall frequency is low, there is a high probability that the frequency ratio happens to increase. For example, in FIG. 9, the frequency ratio of “desktop publishing” is 0.75, and it cannot be said that the frequency ratio is large and the feature level is high. This is because the document frequency is only 3 and the overall document frequency is only 4. Therefore, in order not to compare words having greatly different document frequencies, the document frequencies are divided into appropriate classes in advance and classified into classes, and words having a large frequency ratio in each class are taken as characteristic words. This makes it possible to extract characteristic words from low-frequency words to high-frequency words in a well-balanced manner. The following is an example of how to determine the frequency class. Feature word routine 4
Reference numeral 41 denotes a frequency class division number (C) 5212, which is a parameter indicating the number of frequency classes to be divided and set by the user. Here, C = 5 (generally, C is an integer of 1 or more). Assuming that the i-th frequency class is C [i], the document frequency for belonging to C [i] is f
It is assumed that it is not less than [i] and less than f [i + 1]. However, for the largest class, "f [i + 1] or less" is used instead of "less than f [i + 1]". The method of determining the value of the frequency threshold f [i] is, for example, here, assuming that K ′ is the number of relevant documents, f [i] = K ′ (i / (C + 1))
To the power. (K '= K when the number K of retrieved documents does not exceed the upper limit M of the number of scanned documents, and K' = M when K> M.) In this example, K '= 77. Since C = 5, f [1] = 77 to the (1/6) th power = 2.06,
Hereinafter, f [2] = 4.25, f [3] = 8.77, f
[4] = 18.10 and f [5] = 37.33. Therefore, Class 1: Document frequency 3 or more and 4 or less, Class 2: Document frequency 5 or more and 8 or less, Class 3: Document frequency 9 or more and 18 or less, Class 4: Document frequency 19 or more and 37 or less, Class 5:
The document frequency is 38 or more and 77 or less.

According to the classification condition, the frequency class of each word is determined from the document frequency of each word. In the case of “ROM”, the document frequency is 21 and the class is 4, and in the case of “interactive”, the document frequency is 5 and the class is 2. If the document frequency is lower than class 1 (in this case, the document frequency is 2 or less), it is excluded from the feature word extraction. The above assignment of the frequency class can also be directly calculated by the following equation. However, if the value matches C, 1 is subtracted. (Frequency class) = ｛log (document frequency) ÷ logK '×
(C + 1) Maximum integer value not exceeding -1 Subsequently, the feature word extraction routine reads the number of extracted words (p) 5213, and extracts from the respective frequency classes the ones with the highest frequency ratios so as to add up to this number. . As an example of a method for realizing this, the quotient obtained by dividing the number p of extracted words by the number C of frequency class divisions is n, and the remainder is r. From classes whose frequency classes are 1 or more and r or less, n + 1 are taken. There is a method of taking n classes from classes whose frequency class is larger than r.

The following description will be made with reference to the example of FIG. 9 on the assumption that the number of extractions p is 10. Since the number of divisions C is 5, the quotient n of p ÷ C is 2,
The remainder r is 0. Therefore, two pieces are equally taken from the classes 1 to 5. From the data in the frequency data storage area 523, two data of each frequency class are taken in ascending order of frequency ratio. From the data in FIG. 9, class 5
The words “published” (0.
027), "Electronics" (0.015), "Media"
(0.006) and “information” (0.001). Therefore, the top two “publishing” and “electronic” are taken as characteristic words. Similarly, from class 4, "ROM" and "compact" from class 4, "mail" and "publication" from class 3, "interactive" and "nifty serve" from class 2, and "desktop publishing" from class 1
And "publishing" are extracted as characteristic words. They are stored in the feature word list storage area 524.

FIG. 10 shows an example of data stored in the feature word list storage area 524. The characteristic words extracted by the above process and their document frequencies are stored. In the figure, the frequency class is shown for reference, but this need not be provided. As described above, the process exits the feature word extraction routine 441. Subsequently, the co-occurrence relationship analysis routine 442 analyzes the co-occurrence data relationship between the feature words, and stores the result in the co-occurrence data storage area 531.

The co-occurrence data storage area 531 is a two-dimensional array having a set of characteristic words stored in the characteristic word list storage area 524 vertically and horizontally. Each element represents the number of documents in which the corresponding word pair appears in common. Co-occurrence relation analysis routine 4
Reference numeral 42 reads a word-segmented document group from the word-segmented document storage area 522, and increments the corresponding element in the co-occurrence data storage area 531 for all characteristic word pairs that co-occur for each document. To go.

Next, the co-occurrence relation analysis routine 442 calculates the co-occurrence strength for each characteristic word pair. The co-occurrence strength is a value obtained by dividing the co-occurrence frequency counted in the above operation by the document frequency of the latter word in the table (the word corresponding to the column in the table). The value (FIG. 10) stored in the feature word list storage area 524 is used as the document frequency of the word. FIG. 11 shows the data stored in the co-occurrence data storage area 531 at this stage. Each cell is composed of two numerical values, and the upper row shows the co-occurrence frequency of the corresponding word pair, and the lower row shows the co-occurrence strength of the word pair (co-occurrence frequency / document frequency of the word on the column side). For example, the upper numerical value 6 in the sixth row and the third column indicates that the characteristic word “published” in the sixth row and the characteristic word “ROM” in the third column co-appeared in six documents. In this case, since the document frequency of the word “ROM” on the column side of the word pair is 21, the numerical value of the co-occurrence strength in the lower row is 6 ÷ 21 ≒ 0.
29. In the co-occurrence data storage area 531, the characteristic words are arranged in descending order of the document frequency. Only the lower half of the diagonal line in the table will be used in later work, so the rest is omitted.

Subsequently, the co-occurrence relation analysis routine 442 extracts word pairs having a high co-occurrence degree (pairs to be linked in the characteristic word graph) from the co-occurrence data. In this embodiment, a link indicating the relevance between characteristic words is set to a word having the largest value of co-occurrence strength among words having a higher document frequency than each word. Co-occurrence relation analysis routine 44
2 collects word pairs to be linked according to this criterion and stores them in the co-occurrence link storage area 532. Even if the co-occurrence strength is 2 or 3, it is also effective to link if the co-occurrence strength is not so small (for example, 0.9 times or more of 1). . FIG. 12 is a diagram showing the contents of the co-occurrence link storage area 532 at this stage. Figure 1 shows the process of extracting these links.
A description will be given based on the example of FIG. Looking at the second “publishing” in FIG. 12, since there is only “electronic” for a document whose document frequency is “publishing” or more, a link is provided from “publishing” to “electronic”. Looking at the third “ROM”, the two most frequent are “publishing” and “electronic”, and their co-occurrence strength is 0.27. In this case, a link is provided to "publishing" with a smaller number in the co-occurrence data storage area 531. Next, regarding the fourth “compact”, the co-occurrence strength with the third “ROM” is the largest at 0.81. Therefore, a link is provided from "compact" to "ROM". Thereafter, the same operation is continued to obtain link data as shown in FIG.

As described above, the co-occurrence relation analysis routine 442 is exited, and subsequently, the graph arrangement routine 443 is started.
Based on the data of the characteristic word list storage area 524 (FIG. 10) and the data of the co-occurrence link storage area 532 (FIG. 12), a work of actually arranging a graph having a group of characteristic words as nodes on a two-dimensional plane is performed. FIG. 13 shows details of the graph arrangement routine 443. Graph placement routine 443
Are the y coordinate calculation routine 4431 and the x coordinate calculation routine 4
432, a conversion routine to display coordinates 4433, an overlap avoidance routine 4434, and a link arrangement routine 4435, which are activated in this order. y coordinate calculation routine 44
31 and the x coordinate calculation routine 4432 calculate the coordinates at which each node is to be arranged, assuming that the display area is a square area of [-1, 1] × [-1, 1, 1]. These coordinates are called normalized coordinates. The calculated coordinate data is stored in the normalized coordinate storage area 541.

First, the y coordinate calculation routine 4431 is started, and the calculation formula is: y = (6 / π) × arctan (0.2 × log (f /
fm)), the normalized y-coordinate of the position where it should be displayed is calculated from the document frequency f of each characteristic word. That is, the higher the document frequency, the higher the position on the y-axis. Here, fm is the frequency of the characteristic word that comes exactly in the center when the characteristic words are arranged in document frequency order (however, in the case of an even number, the number is (number ÷ 2 + 1) th). In the embodiment, the document frequency 77 of “electronic” and “publishing” is at the top, and the document frequency 9 of “publication” is at the center position. π is the pi, log is the natural logarithm, arctan is the inverse of the tangent function, and the angle is in radians. For example, since the frequency of “compact” is 21, its normalized y coordinate is (6 / π) × arctan (0.2 × log (21 ÷
9)) ≒ 0.32. The normalized y coordinates of the other feature words are calculated in the same manner. Next, x coordinate calculation routine 4
432 is activated to calculate the normalized x coordinate of each characteristic word display position. FIG. 14 is a diagram showing details of the x coordinate calculation routine 4432. First, in step 44321, nodes without a parent node (link destination) are collected. In this case, only "electrons" correspond to it. Therefore, the value of the x coordinate is calculated by the expression xi = −1 + in step 44321.
Substituting i = 1 for 2i / (r + 1), -1+ (2 × 1)
/ (1 + 1) = 0 is calculated.

Subsequently, a loop 44322 is entered, and in a step 44323, one node linked to a node having a fixed x coordinate (in this case, only “electron”) is taken.
From the co-occurrence link data (FIG. 12), it can be seen that "publishing" satisfies the condition here. Subsequently, the process proceeds to step 44324, in which a set of parent nodes of the node selected in step 44323 is obtained, and further, the average value of their x coordinates is calculated. The set of parent nodes of “publishing” is {“electronic”}, and the average of the x coordinate is 0. Next, in step 44325, nodes whose parent node set matches {“electronic”} are collected. Here it is "publishing only.

Then, the process proceeds to a branch step 44326, but since the average value of the x-coordinate of the parent node is 0, a step 44327 is performed.
Is selected and the x-coordinate of “publishing” is calculated. By substituting s = 1, xp = 0, and i = 1 into the calculation formula of step 44327, the x coordinate of “publishing” is calculated as 0. Thus, the normalized x-coordinates of “electronic” and “publishing” have been determined. However, since the x-coordinates of all the nodes have not been determined yet, the loop 44322 is repeated. In step 44323, among the nodes for which the x coordinate has not been determined yet, one of the nodes to which a link is not set other than “electronic” and “publishing” is selected. In this case, “ROM” satisfies the condition.
In step 44324, a set of link destinations of "ROM" is obtained to obtain "" publish "". Also, the average value xp of the x-coordinate of the parent node {“publish”} is calculated to be 0.

In step 44325, nodes whose link destination set matches {“publish”} are collected. "RO
Other than "M", "email" corresponds to it.

Since the average value xp of the x-coordinate of the parent node is 0, the upper stage is selected at branch 44326, and step 4432 is selected.
7, x coordinates of "ROM" and "mail" are calculated as [-0.33, 0.33 by dividing [-1,1,] into three equal parts, respectively. In a similar manner, nodes having a common link destination are collected for nodes whose links are established only to nodes for which the x-coordinate has already been determined, and within the section [−1, 1] centered on the average of the parent x-coordinate. The x coordinate is determined so as to be evenly arranged so as to fit.

FIG. 15 is a diagram showing the coordinate data stored in the normalized coordinate storage area 541 at this stage in the example of “electronic publishing”. Then, graph arrangement routine 4
43 activates a conversion routine 4433 to display coordinates, and converts the coordinates normalized to the above [-1, 1] × [-1, 1] area to coordinates representing the actual position in the feature word display unit 224. The conversion is performed, and the result is stored in the center coordinate column (FIG. 16) of the node storage area 5431. The conversion is performed by the following linear expression. _{X = R x × (1 +} x) + O x, Y = R y × (ym
-Y) + O _y. Here, lowercase x and y are normalized coordinates, and uppercase X and Y are coordinates in the characteristic word display unit 224. ym represents the maximum value of y. In the example of FIG. 15, ym
= 0.774. The coefficients R _x , R _y , O _x , and O _y use the values stored in the corresponding areas of the graph arrangement parameter storage area 542 (FIG. 2). In this example, R _x = 20
0, R _y = 200, O _x = 60, and O _y = 40. For example, in the case of “compact” by the above-mentioned linear transformation, since the normalized coordinates are (−0.555, 0.320), X
= 200 × (1-0.555) + 60 = 149, Y = 2
It is calculated as 00 × (0.774−0.320) + 40 ≒ 131. In this way, the actual coordinates of all nodes on the characteristic word display unit 224 are calculated and stored in the node storage area 5431 (FIG. 16). At this time, as preparation for the next step, words are arranged in ascending order of x-coordinate. In addition, the number of horizontal characters h and the number of lines v as the size of the character display area, the horizontal size H and the vertical size V of the character display area are calculated and stored in the node storage area 5431.

The character display area size is calculated according to the following formula. Characters are written horizontally, and the size limit is W characters. The value of W is the upper limit 542 of the number of characters in the horizontal direction of the character display section.
Use the value stored in 6. Here, it is assumed that W = 3.
Assuming that the number of characters to be displayed is M, the number of characters h in the horizontal direction,
And the number of rows v is M ≦ W, h is M and v is 1. When M> W, h is W and v is a minimum integer equal to or larger than (M ÷ W). For example, for “Electronic”, the number of characters is two, which is smaller than the width limit W = 3, so that the number of lines v
Is 1 and the width h is 2. In the case of "interactive", since the number of characters is 8 and exceeds the width limit W = 3, the number of lines v is the minimum integer equal to or more than (8/3), that is, 3, and
The width h is W = 3. Further, a value H of a half of the horizontal size and a value V of a half of the vertical size of the character display area are calculated from the respective numbers of characters h and v by the following formula. Where 2
The reason for taking the half value is that the half value is mainly used in the subsequent processing. H = h × F / 2 + _mx , V = v
× F / 2 + m _y. Where F is the character font size, m _x
The magnitude of the x-direction of the margin, m _y is the magnitude of the y direction of the margin. m _x a m _y are such that the two nodes are not too close, it represents the distance should be kept minimal. F, m _x, m _y
Use the values stored in the character size 5425, the horizontal margin 5427, and the vertical margin 5428 (FIG. 2) of the character display unit, respectively. In this example, F = 16, _mx =
3, suppose that m _y = 2. For example, in the case of "compact", h =
3 and v = 2, H = 3 × 16/2 + 3 = 27, V = 2
X16 / 2 + 2 = 18 is calculated. The number of characters and the display area size as the character display size in the node storage area 5431 in FIG. 16 are calculated in this way.

Although the coordinates in the characteristic word display section have been obtained in this manner, there is a possibility that nodes may overlap at this stage. For example, in the example of FIG. 16, "electronic" and "publishing"
Have the same coordinates, so they overlap. For this reason, the overlap avoiding routine 4434 is started, and an operation is performed to shift the coordinates so that the overlap does not occur.

FIG. 17 shows the details of the overlap avoiding routine 4434. All nodes are sorted in ascending order of x-coordinate, and N [1],. . . , N [r]. Let the coordinates of N [i] be (X [i], Y [i]) and the value of the character display area size be (H [i], V [i]). i = 2,. . . , R
Perform the following operation for j = 1,. . . , I−1, take the maximum value of X [j] + H [j] for j such that | Y [j] −Y [i] | <V [i] + V [j].
And If there is no such j, there is no need to perform the operation of shifting the coordinates for this i. δ = ξ− (X [i]
−H [i]). When δ ≦ 0, there is no need to perform the operation of shifting the coordinates for this i. If δ> 0, an overlap occurs, so that N [i],. . . , N [r]
Are shifted δ to the right. That is, X [k] =
Let X [k] + δ (k = i,..., R).

As described above, coordinates are provided so that all nodes can be displayed without overlapping. For example, in the case of “interactive” with i = 2, | Y [2] −Y [1] | = | 240−131 | =
At 109, since V [2] + V [1] = 26 + 18 = 44, | Y [2] −Y [1] | <V [2] + V [1] does not hold. Therefore, for "interactive", the operation of shifting to the side is not performed. Next, i = 3, that is, “ROM” will be described. For j = 1, | Y
[3] -Y [1] | = | 131-131 | = 0, V [3] + V [1] = 10 + 18 = 28, and | Y
[3] −Y [1] | <V [1] + V [3]. That is, the overlap with the “compact” of j = 1 occurs.
Looking at the relationship with “interactive” with j = 2,
| Y [3] -Y [2] | = | 131-240 | = 10
9, V [3] + V [2] = 10 + 26 = 36 and | Y
Since [2] −Y [3] | <V [2] + V [3], there is no possibility of overlapping with “interactive”. Therefore, the x coordinate only needs to be considered for j = 1. ξ = X
[1] + H [1] = 149 + 27 = 176, and the shift width δ is δ = ξ− (X [i] −H [i]) = 176− (193−27) = 10. Therefore, j = 3,. . .
For X, all X [j] are incremented by +10. (X
[3], Y [3]) = (203, 131), and FIG.
To obtain the coordinates of the "ROM" in. Thereafter, by repeating this step, the same data as in the node storage area 5441 of FIG. 7 is obtained. Even in the operation of avoiding the overlap of the character display area, the overlap between the character display area and the line of the graph described above cannot be checked, and as a practical matter, when the display area is limited and the strict avoidance is required, an appropriate size is required. In the meantime, there is a possibility that the display cannot be performed. Therefore, in the embodiment, this is not checked.

Finally, the graph arrangement routine 443 activates the link arrangement routine 4435. The link arrangement routine 4435 stores information on a word pair to which a co-occurrence link is to be established stored in the co-occurrence link storage area 532 in the co-occurrence relation analysis work area 53 and the node data storage area 5.
From the coordinate data of each node stored in 431, data of a line segment to be displayed on the characteristic word display unit 224, that is, the coordinates of the start point and the coordinates of the end point are created and stored in the link data storage area 5422. For example, the co-occurrence link storage area 532 in FIG. 12 has a link from “ROM” to “publishing”. From the data stored in the node data storage area 5431 of FIG. 7, the coordinates of “ROM” are (203, 13).
Since it can be seen that the coordinates of “publishing” are (308, 40), (3) with (203, 131) as the starting point,
08, 40) is stored in the link data storage area 5432. Thus, the data of the graph to be displayed (FIG. 7) is created. Hereinafter, an example of a usage form in which a search operation is advanced with reference to a graph display of the characteristic word displayed on the characteristic word display unit 224 of the characteristic word display unit 22 will be described.

FIG. 8 shows an example of the display of characteristic words relating to “electronic publishing”. Here, it is assumed that the user is interested in “desktop publishing”, which is one of the provisionally displayed words. In this case, the position of the word is indicated on the screen with the mouse 12 or the like, and then the point button 2
If you specify 2222, "Desktop Publishing"
Are stored in the additional keyword storage area 5112, and are displayed on the additional keyword display unit 2122 of the search interface 21 and the additional keyword display unit 2222 of the characteristic word display unit 22. Then, when the search execution button 216 of the search interface 21 or the search execution button 2211 of the characteristic word display unit 22 is pressed, the search is executed in a form in which “desktop publishing” is added to the additional keyword, and the search can be narrowed down. Also, the characteristic word display unit 2 of FIG.
If an interesting word cannot be found in the characteristic words displayed in 24, the number of display words can be increased by using the characteristic word display number setting means 2251. FIG. 18 shows an example in which the number of characteristic word display words is increased to 20. In this case, in the example of the data in FIG. 9, 20 words are selected from this data by the characteristic word extraction routine 441, and displayed in the same manner as described in the case of FIG. Here, if the user is interested in “information search” in “electronic publishing”, the words “search” and “information search)” are displayed in the displayed graph, and can be used. Click on "Search" and "Information Search" in the feature word display area with a mouse or the like and then move to the additional keyword 2
Pressing 2222 adds these words as additional keywords. By pressing the search execution button 2211, the search can be narrowed down. If the user wants to view a graph of characteristic words after narrowing down the search, the user can press the characteristic word display button 2212. Then, when the search and the graph of the characteristic word are continuously performed, if the search execution + characteristic word display button 2213 is pressed, the above steps are continuously performed.

Next, if the user is not interested in "information search", or has already read the document related to "information search" and wants to pay attention to other topics, the deduction keyword is used. If “search” and “information search” have already been added to the point-added keyword, these words displayed on the point-added keyword display section 2222 are indicated by a mouse or the like, and then the move to the point-reduced keyword button 2
If the user presses 2232, these words move from the point-added keyword to the deducted keyword. If the user wants to directly use the word displayed in the characteristic word display section 224 as a deduction keyword, click the corresponding word with a mouse or the like and press the move button 22232 to the deduction keyword, as in the case of the point-added keyword. Good. That is, in the present embodiment, a move operation is performed by the move button between the search keywords, and a copy operation is performed by the move button between the displayed characteristic word and the keyword.

When the search is executed after moving "search" and "information search" to the deducted keyword, the score of a document including these words is reduced, and the score of a document not including these words is relatively increased. Electronic publishing "
It is possible to focus on documents that are not related to “information search”. FIG. 19 includes a characteristic word display style selecting means 2171,
This is an example of a search interface 21 having a function of selecting display of a characteristic word in a graph form or a list form. Compared to the graph display, the list display displays many feature words, so the relevance of the feature words cannot be displayed, so the result of focusing on the relevance cannot be evaluated. By using the bar, many characteristic words appearing in the search result can be listed, so that there is an advantage that the possibility that the user can find a related word that matches the interest is high.

Therefore, using the characteristic word display style selecting means 2171 shown in FIG. 19, first, the retrieval result is displayed in a graph, and the whole image of the characteristic word is overviewed including the mutual relevance. If the related words that match the user's interest are not sufficiently expressed, a two-stage result evaluation can be performed in which a more detailed search is performed using a list display. Further, when a word of interest is obtained from the display using the list, the word can be used as a keyword and the search can be started again. If "graph" is selected by the characteristic word display style selecting means 2171 in FIG. 19, a graph of characteristic words is displayed as described in FIG. 8 or FIG. As shown in FIG. 19, if "list" is selected, the characteristic words are displayed in the form of a list on the characteristic word display section 224, as shown in an example in FIG. Even when "list" is selected by the characteristic word display style selecting means 2171, the method of extracting characteristic words from the retrieved document group is the same as in the case of the graph display described above. However, in the case of a list display, as shown in FIG.
It is considered that it is easier to see three classes of high, medium and low. Therefore, in the display example of FIG. 20, the number of divisions of the frequency class is three. In FIG. 20, in response to the selection of “list”, the characteristic word display section 224 displays the high frequency characteristic word display section 22.
41, a display frame with a scroll bar is set in each of the medium-frequency characteristic word display section 2242 and the low-frequency characteristic word display section 2243, and a characteristic word corresponding to the frequency class data of the characteristic word in the frequency data storage area 523 is displayed in each display frame. Will be displayed within The display order in each display frame is preferably arranged, for example, in the order of the magnitude of the frequency ratio. As a result, the user can list from characteristic words having higher generality to characteristic words having higher specificity such as proper names, and can search for a word that matches the interest from a wide range of options.

Embodiment 2 Hereinafter, a second embodiment of the present invention will be described with reference to FIG. Whereas the first embodiment is an example of a configuration of a search device by a computer used independently, in this embodiment,
The present invention realizes a search method that can respond to search requests from a plurality of users. FIG. 21 shows the overall configuration of another embodiment for realizing the document search method of this embodiment. In this embodiment, a plurality of clients can access one server via a signal transmission line, and can receive a search service for each client. Generally, a server does not use the server itself as a client. However, in the present embodiment, in consideration of the fact that it is necessary to use both the server and the client in response to a problem pointed out by the client, the server is used in the first embodiment.
A search device is provided in which the communication means 7 is added to the configuration substantially the same as that described above. The client attaches dashes corresponding to each of the input unit 1, the display unit 2, the CPU 3, the calculation program holding unit 4, the work area 5 for operating the calculation program, and the bus 100 in the configuration described in the first embodiment. It comprises the means shown, the communication means 7 for linking with the server, and the printer 81 as the output means 8. An interface IF1 is provided on the bus 100 of the server, and interfaces IF2 and IF3 are provided on the bus 100 of the client. Lines NET1 and NE connecting the server and the client are provided.
It is tied at T2. As for the client 2, only the bus 100 and the interface IF2 are shown, and the others are omitted.

When the client 1 attempts to search for a document, first, a command for starting the document search system is input from the keyboard 11 of the input unit 1. In response, the communication means 7 on the client side and the communication means 7 on the server side communicate with each other via the communication path NET1, and the search interface operation routine 41 of the calculation program holding means 4 on the server side communicates with the client 1 side. It is transmitted and activated on the client 1 side. As a result, a search interface 21 for interactively performing a search operation is displayed on the display unit 2. After the search interface 21 is displayed, the client 1 may use this screen to input a word serving as a search key in the same procedure as described in the first embodiment. Note that the client may hold a copy of the search interface operation routine 41 in the calculation program holding means 4 and start it. Also, WW
It is also convenient to be able to receive this search support service using a hypertext browsing interface such as a W browser. In this case, the server prepares hypertext (HT) for transmitting the search interface operation routine 41 to the client. It is assumed that the client has an environment in which a general-purpose hypertext browsing interface can be used.

From the address input section of the hypertext browsing interface displayed on the display means 2, the address designated by the search support service (ie, the address of the server on the network and the search interface operation routine 41 are sent. Is specified, the specified hypertext HT is sent to the client via the communication means together with the search interface operation routine 41, and the sent search interface is sent. Operation routine 41
Is started on the client computer, and the search interface 21 is displayed on the display means 2 and becomes usable. In addition,
In the above description, the address of the hypertext HT is directly specified. However, if the address of the present hypertext HT is embedded as an anchor in the hypertext displayed on the browsing unit of the hypertext browsing interface, the anchor is used. The same operation can be performed by clicking the portion with a mouse or the like.

The search request input by the client 1 is transmitted to the server side via the communication means 7 and 7 and the communication path NET1, and the server executes necessary search, characteristic word extraction, and graph layout calculation. Is returned to the client 1 by the communication means 7, 7 again, and the client 1
Is passed to a search interface operation routine 41, which displays a characteristic word graph on the characteristic word display means 22 based on the data. As described in the first embodiment, the client 1 may input data corresponding to a further necessary search operation according to the search result, as described in the first embodiment. This data is transmitted to the server again, and a necessary search is executed on the server.
2 is displayed. The client 1 can use the output printed by the printer 81 if necessary. In this way, the client 1 can use only the result executed on the server side without having a substantial search program. Therefore, in the client 1, the work area 5 only needs to have the ability to hold the initial input data, the search result transmitted from the server, the characteristic word, and the data relating to the graph layout, and the like. Service is available.

[0048]

As described above, according to the present invention, the user can list from the more general characteristic words to the highly specific characteristic words such as the proper names, and can select from a wide range of options. You can search for words that match your interests.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an example of a configuration of a search device by a computer used independently as an embodiment of the present invention.

FIG. 2 is a diagram showing an example of data allocation arrangement in a work area.

FIG. 3 is a view showing an example of a search interface display screen between a user and a computer.

FIG. 4 is a diagram showing an example of data stored in a search work area when a search is performed.

FIG. 5 is a diagram showing an example in which the search interface display screen shown in FIG. 3 displays a search result after executing a search.

FIG. 6 is a diagram showing an example of a display screen at the time of starting a characteristic word display unit for giving a characteristic word as a search key by a user.

FIG. 7 is a diagram illustrating an example of data stored in a characteristic word graph storage area when a characteristic word display request is made by a user.

FIG. 8 is a diagram showing an example of a graph display of characteristic words in a retrieved document group.

FIG. 9 is a diagram showing an example of word frequency data in a retrieved document group.

FIG. 10 is a diagram showing an example of a feature word list in a retrieved document group.

FIG. 11 is a diagram showing an example of data representing a co-occurrence relationship between characteristic words in a retrieved document group.

FIG. 12 is a diagram showing an example of a list of characteristic word pairs having a particularly strong co-occurrence relationship in a retrieved document group.

FIG. 13 is a pad diagram (PAD diagram, Problem Analyst) showing an example of the configuration of a calculation routine for calculating the graph arrangement of characteristic words;
ysis Diagram).

FIG. 14 is a pad diagram showing an example of an x coordinate calculation method in a graph layout.

FIG. 15 is a diagram showing an example of coordinate data when display data is virtually arranged in a normalized region when a graph of a search result is displayed.

FIG. 16 is a view showing an example of coordinates of a graph before avoiding overlapping of display data when displaying a graph of a search result.

FIG. 17 is a pad diagram showing an example of details of a routine for avoiding overlapping display nodes of a graph.

FIG. 18 is a diagram showing an example of a characteristic word graph display when the number of characteristic words to be displayed is 20;

FIG. 19 is a diagram showing an example of a search interface display screen provided with a feature word display style selection unit.

FIG. 20 is a diagram showing an example of a display screen for displaying a list of characteristic words.

FIG. 21 is a block diagram showing a configuration example in a case where a subject of a search apparatus is provided on a server side and a plurality of clients access the server to perform a search.

[Description of sign]

1, 1: input means, 11, 11: keyboard, 12, 1
2: mouse, 13, 13: pen input means, 2, 2: display means, 21, 21: search interface, 7, 7: communication means, 8: output means, 81: printer 81, IF1, I
F2, IF3: Interface, NET1, NET
2: line, 211: search request input unit, 212: keyword display / operation unit, 2121: essential keyword display unit, 2
1211: Add button to required keyword, 2121
2: Erasing button for essential keywords, 2122: Additional keyword display area, 2123: Deduction keyword display area, 21
3: Search hit count display area, 214: Title display area,
215: document display section, 216: search execution button, 21
6: characteristic word display button, 2171: characteristic word display style selection means, 22: characteristic word display means, 221: characteristic word display means operation section, 222: keyword display / operation section of characteristic word display means, 223: characteristic word display Mean number of search hits display section, 224: characteristic word display section, 2241: high frequency characteristic word display section, 2242: medium frequency characteristic word display section, 2243: high frequency characteristic word display section, 225: parameter of characteristic word display means Setting unit, 2251: characteristic word display word number setting means, 3: calculation program execution means (CPU), 4: calculation program holding means, 41: search interface operation routine, 4
2: Morphological analysis routine, 43: Search routine, 44:
Characteristic word display means operation routine, 441: characteristic word extraction routine, 442: co-occurrence relation analysis routine, 443: graph arrangement routine, 4431: y coordinate calculation routine, 443
2: x coordinate calculation routine, 4433: conversion routine to display coordinates, 4434: overlap avoidance routine, 4435:
Link arrangement routine, 444: graph display routine,
5: Work area, 51: Search work area, 511:
Keyword storage area, 5111: Essential keyword storage area, 5112: Additional keyword storage area, 511
3: Deducted keyword storage area, 512: Search result storage area, 513: Search result score distribution storage area, 52:
Feature word extraction work area, 521: Feature word extraction parameter storage area, 5211: Scanned document number upper limit storage area, 5212: Frequency class division number storage area, 521
3: Extracted word number storage area, 522: Word divided document storage area, 523: Frequency database storage area, 52
4: characteristic word list storage area, 53: co-occurrence relation analysis work area, 531: co-occurrence data storage area, 532: co-occurrence link storage area, 54: graph layout work area,
541: normalized coordinate storage area, 542: graph arrangement parameter storage area, 543: graph storage area, 54
31: node storage area, 5432: link storage area, 6: database holding means, 61: search target document database, 62: search index database,
63: word frequency database, 64: exclusion word database.

Claims (17)

[Claims] Detecting a document having the keyword as a search result document from a search target document group according to a set keyword, and indicating how many documents in the search result document group appear in a certain word; Detecting the document frequency of a word, detecting the total document frequency of the word meaning how many documents the word appears in the entire search target document group, and determining the document frequency of the word and the total document frequency of the word. Deriving a frequency ratio meaning the ratio of, the document frequency is divided into frequency classes in a predetermined relationship, and each word corresponds to the frequency class according to the document frequency of each word,
Literature search support characterized by extracting an appropriate number of words from each frequency class in the order of the frequency ratio of words as feature words, and displaying the extracted feature words in a graph format or a list format Method. 2. The document search support method according to claim 1, wherein the extracted characteristic words are displayed in either a list format for each frequency class or a graph format showing the relationship between the characteristic words. 3. A means for detecting a document having the keyword as a search result document from a search target document group according to a set keyword, and means how many documents in the search result document group appear in a certain word. Means for detecting the document frequency of the word, means for detecting the number of documents in which the word appears in the entire search target document group, and the document frequency of the word and the entire document frequency of the word. Means for deriving a frequency ratio meaning a ratio of the words, means for classifying the frequency ratio into frequency classes according to a predetermined relationship, and means for associating each word with the frequency class according to the frequency ratio of each word; A document extractor comprising: means for extracting a number of words as characteristic words in order of magnitude of the frequency ratio of words; and means for displaying the extracted characteristic words in a graph format or a list format. Apparatus. 4. A means for displaying the extracted characteristic words in either a list form for each frequency class or a graph form showing the relationship between characteristic words, and means for selecting and specifying a characteristic word display form. 3. The document search device according to 3. 5. The relation between the characteristic words is determined on the basis of the co-occurrence relation between the characteristic words, and the graph form is formed by linking word pairs of characteristic words with characteristic words as nodes. The document search device according to claim 3 or 4, wherein the document search device is a graph. 6. The set keyword is a required keyword,
There are three types of keywords, an additional keyword and a deducted keyword. The search by the essential keywords is performed by an AND condition based on each essential keyword. For each document in the retrieved search result document group, the additional score is added when the additional keyword is included. Score high according to the number of keywords,
5. The document search apparatus according to claim 3, wherein when a deduction keyword is included, a deduction is performed according to the number of deduction keywords, and a characteristic word is extracted from a document group that has obtained a higher score. 7. The document search apparatus according to claim 6, wherein a search is performed by additional keywords when no essential keyword is set, and the search by each additional keyword is performed under an OR condition. 8. The type of keyword can be changed among three types of set keywords, a required keyword, an additional keyword, and a deducted keyword, and the displayed characteristic word can be changed to any one of the required keyword, the additional keyword, and the deducted keyword. 8. The method according to claim 6, wherein the crab can be copied.
Document search device described. 9. The document retrieval apparatus according to claim 4, wherein, in the characteristic word graph display, the vertical axis represents the document frequency of the characteristic word in the retrieved document group. 10. A method for detecting a document having a keyword as a search result document from a search target document group according to a keyword transmitted from a search source, and determining how many documents in the search result document group include a word. Detecting the document frequency of a word meaning the word, detecting the total document frequency of the word meaning how many documents the word appears in the entire search target document group, detecting the document frequency of the word and the entire word Deriving a frequency ratio meaning a ratio to a document frequency, classifying the frequency ratio into frequency classes according to a predetermined relationship, and associating each word with the frequency class according to the frequency ratio of each word, Extract an appropriate number of words from the class as feature words in order of magnitude of the word frequency ratio, and configure the extracted feature words as data that can be displayed in a graph format indicating the relationship between the feature words A document search service comprising: configuring the extracted feature words as data that can be displayed in a list format for each frequency class; and transmitting the feature words to a search source as data that can be displayed in a graph format or a list format. Method. 11. The search source includes at least means for transmitting a keyword for specifying a document having a keyword to be extracted, and a graph form or list showing the transmitted characteristic words and the relation between the characteristic words. 11. The document search service method according to claim 10, further comprising means for receiving and displaying data that can be displayed in a format and receiving a search service. 12. The search source according to claim 10, wherein the search source receives the search service by transmitting, together with display software, data that can be displayed in the form of a graph or a list showing the transmitted characteristic words and the relation between the characteristic words. Literature search service method. 13. The search source receives a search service by transmitting user interface driving software for receiving a search service at the start of a search operation or in advance from a search service provider and driving the user interface drive software to receive the search service. 10. A document search service method according to 10. 14. A computer-readable recording medium on which frequency data for calculating a characteristic degree of each word appearing in a search result is recorded, wherein data on each word is:
(A) a character string, (b) a document frequency indicating how many times the word appears in the searched documents, and (c) a number of documents used in the entire search target document regardless of the search result. (D) the document frequency of the word in the search result calculated from the document frequency in the search result and the entire document frequency in the whole database,
(E) A frequency class obtained by classifying according to the magnitude of the document frequency in the search result, and a word having a higher degree of characteristic from each of the frequency classes is set as a characteristic word in the search target document group. A computer-readable recording medium that records frequency data of words appearing in search results. 15. A computer-readable recording medium for recording co-occurrence data in which characteristic words co-occur in order to calculate the degree of relevance between characteristic words appearing in a search result, wherein data relating to each characteristic word pair is recorded. Is related to (a) the co-occurrence frequency at which both characteristic words in the search result document group co-occur, and (b) the relation between the co-occurrence frequency and the characteristic words calculated from the frequency data appearing in the search results of both characteristic words. And computer-readable recording which records co-occurrence data between characteristic words in a search result, wherein a link indicating that the characteristic word pair having a high degree of relevance is strong is provided. Medium. 16. A computer-readable recording medium for recording data for displaying a graph of a characteristic word pair appearing in a search result on a screen, wherein the data for displaying the graph of the characteristic word pair on a screen is: a) data for displaying a characteristic word at a node portion of a graph; and (b) data for displaying a link indicating the relevance between characteristic words. Character string, and the number and size of characters in the vertical and horizontal characters of the area to display the character string, the data of each link is composed of start point coordinates and end point coordinates, the two-dimensional display of the characteristic word graph by link and character string A computer-readable recording medium that records data for displaying a characteristic word graph on a screen, wherein the characteristic word graph is enabled. 17. A feature word of each word appearing in a search result is calculated, a feature word is derived, and a link is established to a feature word pair determined to be highly relevant based on the co-occurrence frequency of the feature word pair. A computer-readable recording medium which records data for displaying a graph of characteristic words obtained on a screen, wherein data relating to each word appearing in the search result is (a) a character string, and (b) a searched document. , The document frequency indicating the number of occurrences of the word, and (c) the document frequency throughout the database indicating the number of documents used in the entire search target document regardless of the search result. d) the characteristic degree of the word in the search result calculated from the document frequency in the search result and the entire document frequency in the entire database; and (e) the frequency class in the case of classifying according to the magnitude of the document frequency in the search result. In each of the frequency classes, a word having a higher degree of the characteristic degree is set as a characteristic word in the search target document group. In order to calculate the degree of relevance between the characteristic words, data relating to each characteristic word pair is represented by ( f) From the co-occurrence frequency at which both characteristic words co-occur in the search result document group and (g) the co-occurrence frequency and the relevance of both characteristic words calculated from the frequency data appearing in the search results for each of the characteristic words In order to link the characteristic word pair having a high degree of relevance, and to display the characteristic word graph with the link displayed on the screen, the data for displaying the characteristic word graph on the screen is (h) graph data. Data for displaying a characteristic word in a node portion, and (i) data for displaying a link indicating the relevance between the characteristic words, and the data of each node includes center coordinates, a character string to be displayed, and, The character string display area is composed of the number of characters in the vertical and horizontal directions and the size, and the data of each link is composed of start point coordinates and end point coordinates, and each of the characteristic word graphs can be displayed two-dimensionally by links and character strings. A computer-readable recording medium that records data for displaying a characteristic word graph on a screen.