JP3460728B2 - Document search method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document search method, and more particularly, to a document search method that allows a search word to be regarded as a corresponding document even if the search word is not exactly the same as the word in the document. For example, it is applied to a document management device, an image management device, and the like.

[0002]

2. Description of the Related Art The following are known documents which describe the prior art relating to the present invention. The “similar document retrieval apparatus” proposed in Japanese Patent Laid-Open No. 2-2458 / 1990 automatically extracts a keyword from a document that does not have a keyword by performing a morphological analysis, etc. This is a document search function. When a search word is input, a document with a high degree of similarity is output, and independent words are extracted from the document even if keywords are not added to the document in advance, and the frequency is high. The keywords are used in order from the one, and the similarity is determined by comparing with the search word. However, if the document does not contain the exact same word as the search term, it will not be considered as a relevant document, and independent words will be extracted from the document and used as keywords in descending order of frequency and compared with the search term. The method has the drawback that the more frequently appearing words are more important, and an accurate search cannot be performed.

In addition, "text-based retrieval method based on semantic attributes" (Hiroshi Matsuo, 1)
Vol32, No9, Sep.1991 p1172-1179) deals with similarity relations of various expressions by searching the search instruction sentence by the semantic similarity with the headline sentence of each text based on the semantic attribute of the word. It is a thing. That is, the target card is retrieved from a DB (database) that stores a large number of cards with headwords based on the headline sentences. The headline sentences are treated as an index instead of the entire document, and the search words and the headline sentences are used. The partial agreement of is also recognized. However, since the headline sentence is the search target, there is a drawback that the entire document cannot be searched.

[0004]

[Purpose] The present invention has been made in view of the above circumstances, and it is possible to regard a search word as a corresponding document even if the search word is not exactly the same as the word in the document. Since the score is given to the keyword of, the purpose is to provide an accurate search, and to provide a document search method for searching the entire document (that is, not only the headline sentence). .

[0005]

In order to achieve the above object, the present invention comprises:
(1) Morphological analysis is performed on the input search word to divide it into word strings to which parts of speech are added, and keyword features are added to the word strings obtained by the morphological analysis means.
For each constituent word that has been
Set importance by gradually increasing points from this point
However, the keyword feature of the word string is not added
For each constituent word, the key is
The degree of importance set for each constituent word with word features
Try to increase the points in order from the points greater than the total points
An importance degree setting means for setting an importance degree and setting an importance degree for each constituent word of the search word, and each constituent word of a word string forming a keyword given to a registered document and the search word When the words match the constituent words of the word string, the matching score calculation means for calculating the matching score for each keyword from the importance set in the constituent words of the search word, and the document score for each document from the matching score of each keyword. The document score calculating means for calculating and the document output means for outputting the information about the document in the order of the document score, and further, (2) the (1 ) the importance setting means, the keyword setting means. When the feature is a compound word base, the keyword feature in the word string is assigned.
Not less than the importance of the constituent words that are not
(3) In the above (1), the importance setting unit, when the keyword feature is proper noun constituent word, the single
Making the keyword features in the word string less than the importance of the constituent words to which the keyword features are not added; and ( 4 ) in ( 1 ), the importance degree setting means adds the keyword features as prefixes. If it is a modification, the keyword features in the word string are
Be less than the importance of constituent words that are not assigned ,
Furthermore, (5) In the above (1), the importance setting unit, when the keyword feature is place name identification word, before
Constituent words that are not given keyword features in a word string
To be smaller than the importance of, further, (6) in the (1), the importance setting unit, when the keyword feature is era identifier words, keywords in the word sequence
To be smaller than the constituent word of importance that are not granted over de feature, furthermore, (7) in (1), said matching degree calculation means, keywords constituent word order of the search term constituent word The matching degree is increased according to the number of constituent words that match in the arrangement order of, and ( 8 ) in (1), the matching degree calculation means causes the keyword and the constituent word at the end of the search word to match. If you, increasing the degree of coincidence, further, (9) in the above (1), said matching degree calculation means, if the head of the constituent word of a keyword and the search word matches, increasing the degree of coincidence, Furthermore, ( 10 ) In the above (1), the document score calculating means calculates the document score by weighting the matching degree of the keyword in accordance with the appearance position of the keyword in the document, and further ( 1)
1 ) In the above item (1), the document score calculating means calculates the document score by weighting the matching degree of the keyword according to the subsequent word of the keyword. Hereinafter, description will be given based on examples of the present invention.

FIG. 1 is a block diagram for explaining an embodiment of a document search system according to the present invention. In the figure, 1 is a search word input means, 2 is a document score giving means, 3 is a document ranking means, 4 Is a document output unit, and 5 is a document to which a keyword is added. First, a user inputs a search term.
Next, the document score assigning means 2 assigns a score to each document according to the input search word. Note that, here, it is assumed that the document 5 is prepared in which words are preliminarily sectioned and the keywords are added. Next, the document ranking means 3 sorts the documents to which the scores have been added in descending order of score, and the document output means 4 outputs the sorted documents.

FIG. 2 is a flow chart for explaining the operation of the document score giving means in FIG. step1 ； The search word is subjected to morphological analysis and a part of speech is added to each word. step2 ； Give importance to each of these words according to the rules. step3 ； If the word of the keyword that each document has and the word of the search word partially match, the degree of importance given to the word of the search word is given, and the degree of importance is summed up for each keyword, and the degree of matching of the keyword is calculated. calculate. step4 ； The degree of coincidence is summed up for each document, and it is set as the score of the document.

In FIG. 2, "importance" is a value given to each word by morphologically analyzing the search word. The “coincidence degree” is a value obtained by adding the degree of importance of a search word corresponding to a keyword and a search word (part) in the document, and summing up for each word. The “score” is a value when the degree of coincidence is finally summed up for each document.

FIG. 3 is a flow chart for explaining the importance giving rule for a search word. As described above, it is assumed that the search word is morphologically analyzed and decomposed into parts of speech. First, the first important thing is to put the pointer at the end (step 1). In other words, the processing is performed while returning backward from the end of the word string. First, set n as the base point and sum as 0 (step
2). Next, it is determined whether or not the word has a keyword feature (step 3). Here, it is divided into those that are given and those that are not given, but those that have been given are the processing of the part above the broken line in FIG. 3 (herein called phase 1), and those that have not been given are broken lines. The processing of the lower part (here called phase2) will be performed. The keyword features will be described later.

First, the processing for phase 1, that is, a keyword feature is given will be described. First,
It is judged whether the keyword feature is "prefix modification" (step 4). The “prefix modification” is a prefix having a function of modifying a subsequent word, which will be described later. If there is no "prefix modification", set n to that word (step
5). Then, n is added to the value of sum and 1 is added to the value of n (step 6). Then, it is judged whether or not the word is the beginning of the word string (step 7), and if it is not the beginning, the process returns to the previous word (step 8), returns to step 3, and repeats the same processing. In other words, the values of n and sum increase as the position moves forward in the word sequence . If it is at the beginning, the processing for the keyword feature added ends here, and returns to the end (st
ep11) Start processing of phase2. When the keyword feature is "prefix modification" in step 4, the basic point of the word is set (step 9), and the basic point is added to sum (step 10).

Next, the process of phase 2 is performed. step11
Then, when I returned to the end, I added 1 to the sum I summed up in phase 1.
Is added (step 12). Next, as in phase 1, the presence or absence of keyword features is checked (step 13). In fact, features with no features have already been processed in phase 1, so here we will target those with no features. If there is a feature, it is confirmed whether it is the head of the word string (step 16), and the process is terminated. By the way, if there is no feature in step 13,
Set sum (step 14). Then, sum is added again to the total sum up to now, and 1 is further added (step 15). Then, it is judged whether or not the word is the head of the word string (step 16), and if it is not the head, the process returns to the previous word (step 17), returns to step 12, and repeats the same processing. In other words, in phase 2, sum is added as it goes to the front of the word string. In other words, a keyword feature is assigned a higher importance as it is positioned before the word string, and a keyword feature is assigned no matter how many keyword features are added (even if they are consecutive). The importance of one word that is not read is higher.

Here, the keyword features used in the above description will be described. There are five types of keyword features: compound word base, proper noun constituent words, prefix modification, place name identification, and era identification. Table 1 below summarizes the parts of speech, the characteristics, and the roles that can be given to each feature.

[0013]

[Table 1]

Other than the "prefix modification", when it appears alone, it is less likely to be a keyword, or has a characteristic that the distinctiveness is low. Even if we look only at "devices," this alone is not a characteristic word. The same applies to "place name identifiers" and "era name identifiers". "Tokyo" is "University of Tokyo"
There are many matching words such as "Tokyo company", "Tokyo school", "Tokyo company Tokyo branch", etc., and there are many matching words in the document for "Tokyo" alone. From such an intention, even if these words to which the keyword features are added are located in front of the word string, the importance is increased by one point at a time. On the other hand, the general noun or proper noun without a keyword feature is s
um makes it more important. "Prefix modification" is a little different from other features. Usually, the prefix is not considered to be a keyword, but the prefix that is considered to have a great effect on modifying subsequent words such as "new" and "large" is "prefix modification". For these, I decided to give only the basic points.

Next, the above rule will be described by taking the case where the following words are search words as an example. Example 1 Keio University Institute of Medical Science . Morphological analysis is performed and decomposed into parts of speech. (Morphological analysis results) Keio University Institute of Medical Science → Keio / University / Medicine / Science / Research / Place. Assign importance to each word according to the rules.

[0016]

[Table 2]

As described above, the importance (score) is calculated as follows.
The base point (here, 2 points) is given to the last word of. Key
For words with word features,
The process of adding one point to each word is repeated. Key
Words that do not have features (here, "Keio") are
Add one more to the total of all priorities. this
Includes the keyword "University Institute of Medical Science"
Even if there is a document, the keyword "Keio"
This is because it is considered that the document including is more important. Example 2Research and development of new materials ． Morphological analysis is performed and decomposed into parts of speech. (Morphological analysis result) New material R & D → New / Material / Research / Development ． Assign importance to each word according to the rules.

[0018]

[Table 3]

An example of the handling of the prefix and the handling of the word to which the keyword feature is not given is at a position other than the beginning of the word string. The prefix with the keyword feature "prefix modification" is
A basic point (2 points) is given in order to make a difference in score from the prefix not given. In Example 1, those without a keyword feature were at the beginning of the word string, and were therefore calculated in order from the importance of the last word string. In this example 2, the word having no keyword feature (in this case, "material") is in the middle of the word string, but the flow is the same. I want to give the most importance to that word, so
The importance of "material" was added by adding 1 to the total importance of other words.

Up to this point, the processing of step 2 in FIG. 2 has been completed. In this way, the search terms are given importance. Next, a score is given for each document using this importance. As described in step 3 and step 4 of FIG. 2, when the word of the keyword of each document and the word of the search word match (even if they partially match), the score is given to the importance of the word of the search word. It is obtained by finding the degree of coincidence and finally summing them. The method of scoring will be described using the above-mentioned Example 2, “New Material Research”. That is,
This is the case when "new material research" is used as the search term. The importance of each word of this search term is shown once again.

[0021]

[Table 4]

Next, it is assumed that the following keywords are described in a certain document. At this time, the degree of coincidence of each keyword in the document is calculated as follows.

[0023]

[Table 5]

When the degree of coincidence is calculated, the degree of coincidence for each document is summed up. This value is the score for the document. For example, if example have in this document will be referred to as 13 + 11 + 10 = 34, score it comes to 34 points. In this way, when the scores are given to all the documents, the scores are sorted by the document ranking means, and the documents having the highest scores are outputted by the document output means.

FIG. 4 is a diagram for explaining another embodiment of the document search system according to the present invention. In the figure, 11 is a search word input means, 12 is a document score giving means, 13 is a document ranking means, and 14 is a document ranking means. Is a document output means. Search term input means 11
Enter the user's search term. Document score giving means 12
Is given to all registered documents according to the input search word. It should be noted that keywords registered in advance in units of words are added to each registered document. The document ranking unit 13 sorts the registered documents in descending order of document score. The document output unit 14 outputs the search result to the user.

FIG. 5 is a block diagram of the document score assigning means in FIG. 4, in which 21 is a morpheme analyzing means, 22 is an importance assigning means, 23 is a matching score calculating means, 24 is a document score calculating means, and 25 is a document score calculating means. Is a registered document. Calls to the morphological analysis means 21 morphological analysis of the search terms, to grant a part of speech for each word. In the importance assigning means 22, the importance is a value representing the importance of each word given to each word obtained as a result of morphological analysis of search words. The importance is calculated for each word according to the rule described later. In the coincidence calculation means 23,
The degree of matching is a value indicating the degree of matching between each keyword assigned to the registered document 25 and the search word. In the document score calculation means 24, the document score is a value indicating the degree of coincidence between the registered document and the search word. It is calculated from the degree of coincidence between each keyword assigned to the registered document and the search term.

Each means will be specifically described below. Importance degree giving means First, a method of calculating the importance degree given to each word obtained as a result of morphological analysis of a search word will be described. Importance is calculated according to the following rules. The importance of the word in the part-of-speech group 1 which is the closest to the end without any search word is the basic point. The degree of importance of the part-of-speech group 1 other than that is set to a value obtained by adding an increment to the degree of importance of the part-of-speech group 1 that is closest to the position and rearward. The importance of the prefix with the keyword feature "prefix modification" is the basic point. The importance of the prefix without the keyword feature "prefix modification" is 0. The importance of part-of-speech group 2 is the sum of the importances of part-of-speech group 1 in (1), (2) the importance of a prefix with prefix modification, and (3) the importance of part-of-speech group 2 behind the position. Three of the totals is the value obtained by adding the increment to the total. The importance of words other than the above is 0.

However, the part-of-speech group 1 is (1) a general noun with the keyword feature “compound word base”, (2) a proper noun with the keyword feature “proper noun constituent word”, and (3) a keyword feature “place name”. It is a proper noun with "identifier", and (4) keyword feature proper noun with "era identifier." What is part-of-speech group 2?
(1) General nouns without keyword features, (2) Proper nouns without keyword features, (3) Numbers, (4) Suffix,
(5) An unregistered word or the like. The keyword features are summarized in Table 6 below.

[0029]

[Table 6]

The processing flow for assigning importance according to this rule is shown in FIG. 3 described above. Hereinafter, examples with importance are shown in Tables 7 and 8 below. There are 2 basic points here,
The increment is 1 point.

[0031]

[Table 7]

[0032]

[Table 8]

Matching Calculator Next, a method of calculating the matching between the keyword and the search word based on the importance of each word of the search word will be described. Figure 1 above
In the embodiment of FIG. 2, the degree of coincidence is the sum of the degrees of importance of the search words that match the words included in the keyword. In contrast, the product of the importance of the words in the search word matches a word included in the keyword and the degree of coincidence. Below, as a search term, "new material research and development" shown in the example 2 of assigning importance,
Keyword "new material research""materialresearch""researchmaterial"
Table 9 below shows an example of calculation of the degree of coincidence in the case of changing the above.

[0034]

[Table 9]

[0035] In addition, the matching degree calculation means ing a large degree of coincidence in the case where the word line that is included in the word as well as search words included in the keyword matches in. Therefore, a new "adjacent point" is introduced, and the adjacency point is multiplied by the degree of coincidence for each match between the word sequence included in the keyword and the word sequence included in the search word. Again, Table 10 below shows an example of calculation of the degree of coincidence when the keyword is changed and the search term is “new material research and development”. Here, the adjacent points are two.

[0036]

[Table 10]

When the keyword is “new material research”, the search word and the keyword are completely the same, and the sequences of “new” and “material” and “material” and “research” are the same in the constituent words. There is. Therefore, adjacent points (2 points) are multiplied twice in the calculation of the degree of coincidence (indicated by underlining in Table 10). In Table 9 , the agreement for “material research” and “research material” was the same, which was 24.
However, in order to take into account the search terms of word order and keyword Table 10, search word and "new materials research and development" partially match of word order match the degree of "material research" is two-fold, and has a 48 .

[0038] In addition, so as not to change depending on the number of words in which the degree of coincidence when the keywords and search terms match completely contained in the search term. Therefore, a "normalization coefficient" is newly introduced so that the degree of coincidence becomes the normalization coefficient when the keyword and the search word completely match. First, the score of the search word is calculated from the importance of the constituent words of the search word. The search word score is the degree of matching when the keyword is equal to the search word. For example, if the matching degree calculation method is the method shown in Table 10 , the search word score of the search word “new material research and development” is 2 × 3 × 2 × 2 × 2 × 2.
= 768. Normalization is performed by dividing the degree of coincidence between the keyword and the search word by the search word score document and applying a normalization coefficient. For example, if the normalization coefficient is 1000 points and the search word and the keyword match, the degree of matching is as shown in Table 11 below.

[0039]

[Table 11]

When the normalization is not performed, the degree of coincidence differs depending on the search word, but the normalization process makes the degree of coincidence equal regardless of the search word. In addition, the search term is "new material research and development"
As an example, Table 12 below shows an example of calculation of the degree of coincidence when the keyword is changed.

[0041]

[Table 12]

Document Score Calculation Means Finally, a method for calculating the document score based on the degree of coincidence between the keyword and the search word will be described. In the embodiment shown in FIGS. 1 and 2,
The sum of the degree of coincidence between each keyword assigned to the registered document and the search word for all the keywords in the registered document was used as the document score. Therefore, although there is a drawback that the document score and a large number of keywords that have been granted in the registration document becomes large, to describe difficult method depends on the number of keywords
It In the first method, the average value of the degree of coincidence between each keyword of the registered document and the search word is used as the document score. That is, a value obtained by dividing the sum of the degrees of coincidence between each keyword of the registered document and the search word by the number of keywords of the document is used as the document score. As an example,
Table 13 below shows a case where “new material research and development” is used as a search word and keywords assigned to documents are “new material research”, “material research”, “research material”, and “Ricoh”.

[0043]

[Table 13]

Since the number of keywords in this document is 4, the sum of coincidences is divided by 4. In the second method, the document score is a value obtained by dividing the sum of the degrees of coincidence between each keyword of the registered document and the search word by the number of keywords whose degree of coincidence is 1 or more.

[0045]

[Table 14]

The second method differs from the first method in that
Since the number of keywords for which the degree of coincidence is 1 or more is 3, the sum of the degrees of coincidence is divided by 3. In the third method, the maximum value of the matching degree between each keyword of the registered document and the search word is used as the document score.

[0047]

[Table 15]

Next, another embodiment will be described. The method of assigning importance and the method of calculating the degree of coincidence are the same as those in the above-mentioned embodiment, and therefore their explanations are omitted. The document score calculation method will be described below. The document score calculation is to calculate a document score from the degree of matching between each keyword added to the registered document and the search word. Although a plurality of calculation methods have been proposed in the above-described embodiments, the average value method will be used for the description below. However, the method proposed in the present invention can be applied to the maximum value method and the like. In the average value method, the average value of the coincidence between each keyword of the registered document and the search word is used as the document score. As an example, "new material research and development" is used as the search term, and the keywords added to the document are "new material research", "material research", and "research material".
The case of "Ricoh" is shown.

[0049]

[Table 16]

[0050] In this method of the present invention, the calculation result of the document score by the appearance position of the keyword is changed. Generally, the importance of the keyword varies depending on the appearance position in the document, and it is effective to find the search result that meets the user's request by changing the calculation result of the document score depending on the appearance position. When the occurrence position of the keyword is the title, using the matching degree obtained by the matching degree calculating means a value obtained by multiplying the title coefficient (the original match degree) (weighted degree of matching) the document scoring. In the above example, assume that the appearance positions of the keywords are as shown in the following table. If the coefficient for the title is set to 2, the "material research" that appears in the title
The weighted coincidence is calculated as 61 × 2 = 122. As a result, the document score also differs from the previous value.

[0051]

[Table 17]

[0052] In addition, the first paragraph first sentence occurrence position of the keyword, respectively, first paragraph second sentence later, the second paragraph after 1
Sentences, the second paragraph onward, the second sentence onward, the weighted coincidence degree multiplied by the coefficient is used to calculate the document score. In the example above,
The first paragraph first sentence coefficient is 1.5, the first paragraph second sentence and subsequent coefficients are 1.2, the second paragraph and subsequent first sentence coefficients are 1, and the second paragraph and subsequent sentence subsequent coefficients are Table 18 shows the calculation of the document score when the score is 0.8.

[0053]

[Table 18]

[0054] In addition, changing the calculation result of the document score by subsequent language of keywords. Generally, the importance of a keyword varies depending on the succeeding word of the keyword. Therefore, it is effective to find the search result that meets the user's request by changing the calculation result of the document score depending on the succeeding word. If subsequent word keyword is case particle "ga", used match degree obtained by the matching degree calculating means a value obtained by multiplying the (original degree of coincidence) the "ga" coefficient a (weighted degree of matching) the document scoring . In the above example, assume that the succeeding words of each keyword are as shown in the following table. Here, when the coefficient for “ga” is set to 2, the weighted degree of coincidence of “new material research” whose subsequent word is “ga” is 2
It is calculated as 49 × 2 = 498. As a result, the document score also differs from the previous value.

[0055]

[Table 19]

[0056] subsequent word each sub-particle "is" in the keyword, case particle "wo", rated Case Markers "but" sub-particle "is" /
In the case other than the case particle "wo" (Other) uses a weighted degree of matching multiplied by the coefficient for the calculation of the document score. Table 20 below shows the calculation of the document score when the coefficient for "ha" is 1.5, the coefficient for "wa" is 1 and the coefficient for other is 0.5 in the above example.

[0057]

[Table 20]

Applying the above introduced together,
It is also possible to calculate the document score. In the previous example, the document score is calculated as follows.

[0059]

[Table 21]

The document retrieval method described above is characterized by the following. The search can be performed even when the search word input by the user partially matches the keyword assigned to the document. At the time of search, the degree of matching (matching degree) between the search word and the keyword is calculated. Therefore, the search process is performed according to the following steps. S1: The search word is divided into words by morphological analysis. S2: The importance of each word is set. S3: The degree of coincidence is calculated from the degree of importance of the words having the search word and the keyword in common. However, this method has some improvements. (A) In the importance setting of S2, it is necessary to scan the search word twice from back to front. Therefore, the importance setting is complicated. (B) In the calculation of the degree of coincidence in S3, in paragraph numbers (0022) to (0024), since the order of the keyword and the word of the search word is ignored, the degree of coincidence is the same even for a keyword having a different word order. It becomes a value. For example, with this method, it is not possible to distinguish between "material research" and "research material" with different word order consisting of the same constituent words. (C) As shown in the paragraph number (0035), keywords having different word orders can be distinguished by introducing adjacent points, but a product operation is used to calculate the degree of coincidence. Generally, in a computer, the product operation is slower than the sum operation, so that the document search is slow in this method.

In the embodiment described below, for the improvement point (a), the search word is scanned once. Regarding the improvement point (b), when the word order matches in the calculation of the matching score, a bonus score is given according to the word order match. Regarding the improvement point (c), the product operation is not used for the coincidence calculation. Figure 6
Is a block diagram for explaining still another embodiment of the document retrieval system according to the present invention, in which 31 is a document retrieval means and 32 is a document retrieval means.
The search term input means, the document score applying means 33, document sorting means 34, 35 document output means, is 36, which is the index word file, 37 document files, document registration means 38.

The document registration means 38 saves the document input by the user and the keywords attached thereto in the document file and the index word file. A plurality of keywords can be set in one registration document, and one keyword may be a compound word composed of a plurality of constituent words (for example,
"Document search" is a compound word composed of two words, "document" and "search". The index word file 36 has a structure in which (a plurality of) keywords for each registered document can be identified. The document search means 31 searches the index word file 36 for a document that matches the search word input by the user, and presents the result to the user. The document search is performed by the search word input means 3
2, a document score assigning means 33, a document sorting means 34, and a document output means 35. The search word input means 32 inputs the search word of the user. The document score assigning means 33 calculates a score according to the input search word for all registered documents. The document sorting means 34 sorts the registered documents in descending order of document score. The document output means 35 outputs the search result to the user.

FIG. 7 is a block diagram of the document score giving means in FIG. 6, in which 41 is a morpheme analyzing means, 42 is an importance setting means, 43 is a document score calculating means, and 44 is a matching degree calculating means. . The morpheme analysis means 41 morphologically analyzes the search word, divides it into words, and determines the part of speech for each word. In the document search device of the present invention, a compound word composed of a plurality of words can be used as a search word input by the user. In the degree-of-importance setting means 42, the degree of importance is a value representing the degree of importance of each word given to each word obtained as a result of morphological analysis of search words. Details of the setting method will be described later. In the document score calculation means 43, the document score is a value representing the degree of coincidence between the registered document and the search word. It is calculated from the matching degree of the search word with each keyword given to the registered document. Here, the degree of coincidence is a value representing the degree of coincidence between each keyword assigned to the registered document and the search word. It is calculated from the importance of each word of the search word, and details of the calculation method will be described later. The document score calculation method is the method described above (paragraph number (0042) to
(0055)) is used.

The importance degree setting means and the coincidence degree calculating means will be described below. First, the importance setting means will be described. When the importance level is set, the search word input by the user is divided into words by morphological analysis. A search word Q composed of n (n> 0) words is written as q ₁ ... Q _n . For example, the search term "document search device" is "document"
It is composed of three words, "search" and "device", and q ₁ = document, q ₂ = search, and q ₃ = device. The importance of the word q included in the search word is written as w (q). In the present invention, the importance of a word is given as follows. - tail of the importance of the word the end of the search terms, the basic point α. Importance of the word other than-end tail is a value obtained by adding a value obtained by multiplying the position coefficient β to the distance from the end tail base point.

When this is written as an expression, it becomes as follows. w (qi) = α + β * (n−i) (1) In this method, it is not necessary to scan the search word twice as described in the prior art, and all the constituent words of the search word can be scanned once. The degree of importance can be set. The importance setting is shown as an example. The search term is “new material fiber development”. This search term is divided into four words, “new”, “material”, “fiber”, and “development”. When the parameters of the above equation are α = 10 and β = 2, the importance of each word is as shown in Table 22 below.

[0066]

[Table 22]

In the above-described method, as the number of constituent words of the search word increases, the importance of the word closer to the head increases, so that the number of constituent words of the search word is large even if the head words of different search words are the same. There is a problem that the degree of importance of the word becomes higher . By applying a bias corresponding to the number of constituent words of search words, avoid such problems
Can Ru. That is, the constituent word number coefficient γ is introduced to set the importance. w (qi) = α + β * (n−i) + γ * n ...
(2) Especially, if γ = −β, the importance of the first word can always be the same value independently of the number of constituent words. If the parameters are α = 12, β = 2, γ = -2 for the search term “new material fiber development” used in the previous example, the importance of each word is as shown in Table 23 below. Become.

[0068]

[Table 23]

In the above-mentioned method, the importance is set by the same expression regardless of the nature of the word. However, some do not important as the search word by the nature of words, what is important is desired to provide a high importance. For example, prefixes are generally less important than nouns because they play an auxiliary role. Therefore,
Setting the parameters of importance depending on the part of speech of a word (α, β,
It was to change the γ). For example, parameters for nouns (general nouns, sahen nouns, etc.) are α [noun] = 12,
β [noun] = 2, γ [noun] =-2, parameters for the prefix are α [prefix] = 4, β [prefix] = 0, γ [prefix]
= 0. At this time, the importance of each word of the search word “new material fiber development” is as shown in Table 24 below.

[0070]

[Table 24]

In the above-mentioned method, if the word parts of speech are the same, the importance is set by the same formula. However, whether or not it is important as a search term is not determined only by the part of speech, but depends on the nature of the document targeted by the search system. In the above-mentioned embodiment, the keyword feature is proposed as a description of the grammatical / semantic features of a word smaller than the part of speech. For example, in a fiber-related document retrieval system, nouns related to fibers often appear in documents.
It is less important as a search term than general nouns. Therefore, a keyword feature “compound word base” is added to the noun “fiber” to distinguish this word from other general nouns. Therefore, even according to the keyword feature not only the part of speech of a word importance of setting parameters (alpha, beta, gamma) and to change the. For example, the parameter for the noun class is α depending on the presence or absence of the keyword feature “compound word base”.
[Noun / presence] = 12, β [Noun / presence] = 2, γ
[Nouns / features] =-2, α [Nouns / no features] = 1, β
[Noun / No Feature] = 1, γ [Noun / No Feature] =-1. Assuming that the parameters for the prefix are the same as before, the importance of each word of the search word "new material fiber development" is as shown in Table 25 below.

[0072]

[Table 25]

Next, the coincidence calculation method will be described. In the matching degree calculation, the degree of matching between one of the keywords added to the document and the index word is calculated using the importance set for the constituent words of the search word. Basically, the sum of the degrees of importance set for the constituent words common to the keyword and the search word is defined as the degree of coincidence between the keyword and the search word. For example, it is assumed that "new material fiber development" is used as the search term and the importance is set as shown in Table 25. Here, using the three words “new material”, “new development”, and “synthetic fiber” as keywords, the degree of coincidence is calculated.

1. Keyword: “new material” (“new” and “material” are constituent words) At this time, two words “new” and “material” are common with the search word. Consistency = w (new) + w (material) = 4 + 8 = 12 2. Keywords: "fiber material development"("fiber""material"
“Development” is a constituent word) At this time, the three words “fiber”, “material”, and “development” are common to the search word. Consistency = w (fiber) + w (material) + w (development) = 3 + 8
+ 4 = 15 3. Keywords: "Synthetic fiber sales"("Synthetic""Fiber"
At this time, only “fiber” is common with the search word. Consistency = w (fiber) = 3

In the above method, when a plurality of words are common to the search word and the keyword, it is impossible to distinguish whether or not they are different depending on the appearance order of the common words. In other words, the keyword is "material fiber" for the search term "new material fiber development".
However, the degree of coincidence is the same for "fiber materials". But,
Since the appearance order of the "material" and "fiber" is the same as that of the "fiber material", the "material fiber" should have a higher degree of coincidence than the "fiber material". Therefore, when there are a plurality of words common to the search word and the keyword, a bonus point is added when the order of the words (word arrangement) is the same in the search word and the keyword. The bonus point (hereinafter, referred to as “adjacent point”) is proportional to the number of matching word sequences, and the adjacency per word sequence is δ. When δ = 3, the degree of coincidence with the same search word and keyword as above is as follows.

1. Keyword: The sequence of "new material", "new" and "material" is common. Matching degree = w (new) + w (material) + δ = 4 + 8 + 3 = 15 2. Keyword: “textile material development” Three words are common, but there is no one with the same word sequence. Consistency = w (fiber) + w (material) + w (development) = 3 + 8
+ 4 = 12 In the above method, it is not possible to distinguish between the case where the search word and the keyword completely match and the case where the search word is included in the keyword. That is, the degree of coincidence with the search term “new material fiber development” is the same whether the keyword is “new material fiber development” or “new material fiber development center”. In order to solve this problem, the top δ if the first word of the search words and keywords match, when the tail of the words the end of the search words and keywords match, the δ [end tail] as a bonus point I tried to add it. δ [top] = δ [youngest tail]
= 2, the degree of coincidence with the same search word and keyword as above is as follows.

1. Keyword: The word sequence of "new material", "new", and "material" is common, and "new" is the search term.
Both of the keywords are at the beginning. Matching rate = w (new) + w (material) + δ + δ [start] = 4 + 8
+ 3 + 2 = 17 2. Keywords: "fiber material development", "development" is at the tail end either of the search terms keywords. Consistency = w (fiber) + w (material) + w (development) + δ [ end ] = 3 + 8 + 4 + 2 = 14

[0078]

As is apparent from the above description, the present invention has the following effects. (1) search word a morphological analysis, the results and words are part of speech degraded, word search term and the document by comparing the keywords stored in the part of speech units in the document is not completely match You can also search. (2) by calculating the degree of coincidence between keywords search words and in each document, can be given a score that meets the search term in each document. In this case, the basic point is given to the last word in the word string, and the score is increased as it goes back to the front of the word string. Therefore, the word located before the word string can be given a higher score. . (3) Regarding the calculation of the degree of coincidence between the search word and the document, a compound word base that is one of the keyword features , a proper noun constituent word, and a place name
By using the identification word or the era identification word , it is possible to avoid giving a high score to a word that is unlikely to be a keyword when giving a score to a document. (4) For the calculation of the degree of coincidence between the search word and the document, by using the prefix modification, which is one of the keyword features, it is possible to give a score to a prefix having a special meaning. (5) The degree of coincidence can be accurately calculated by taking the word arrangement order into consideration in calculating the degree of coincidence. (6) Since the importance of the word is set according to the position of the constituent word of the search word, the importance can be set accurately and the search accuracy is improved. In addition, the search speed is improved because the search word is scanned once. At this time, since the weighted coincidence degree and the document score are calculated based on the appearance position of the keyword in the registered document or the subsequent word, the document score is more accurate than the conventional one. (7) Since the order (word sequence) of the constituent words of the search word and the keyword is reflected in the degree of coincidence, the degree of coincidence calculation can be accurately performed and the search accuracy word is improved. Further, since the coincidence degree calculation is only a sum operation, the search speed is improved.

[Brief description of drawings]

FIG. 1 is a configuration diagram for explaining an embodiment of a document search system according to the present invention.

FIG. 2 is a flowchart for explaining the operation of the document score giving means in FIG.

FIG. 3 is a flowchart for explaining a rule of giving importance to a search word according to the present invention.

FIG. 4 is a configuration diagram for explaining another embodiment of the document search system according to the present invention.

5 is a block diagram of the document score giving means in FIG.

FIG. 6 is a configuration diagram for explaining still another embodiment of the document search system according to the present invention.

7 is a configuration diagram of the document score giving means in FIG.

[Explanation of symbols]

1 ... Search word input means, 2 ... Document score giving means, 3 ... Document ranking means, 4 ... Document output means, 5 ... Documents given from keywords.

─────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-3-116377 (JP, A) Satoshi Imago, Document filing system using short unit keywords, Proc. Of the 43rd National Convention of IPSJ ( 4), 1991, p. 4-215 to p. 4-216 Masako Mouji, Introduction of features in Japanese morphological analysis, IPSJ 43rd Annual Meeting (3), 1991, p. 3-111 to p. 3-112 Nakamura, Sugiyama, On the Matching Technical Report on Partial Matching Search, Proc. p. 5-263 (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06F 17/30 JISST file (JOIS)

Claims (11)

(57) [Claims] 1. A morpheme analyzing means for morphologically analyzing an input search word to divide it into a word string having a part of speech, and a keyword feature of the word string obtained by the morpheme analyzing means.
For each constituent word given with, go back to the front of the word string.
Therefore, increasing the points from the basic point in order
Is set and the keyword features of the word string are assigned
For each constituent word, follow backward in the word sequence.
And set the weight for each constituent word to which the keyword feature is given.
The points will be gradually increased from the total score of the importance
Thus, the degree of importance is set, and the degree-of-importance setting means for setting the degree of importance to each of the constituent words of the search word, each of the constituent words of the word string constituting the keyword given to the registered document, and the search word When a match is found between the constituent words of the word string that constitutes the, the matching score calculation means for calculating the matching score for each keyword from the importance set in the constituent words of this search word, and the document score is calculated from the matching score for each keyword. A document retrieval method comprising: a document score calculation means for calculating each document; and a document output means for outputting information about the document in the order of the document scores. 2. The importance level setting means, when the keyword feature is a compound word base, a keyword in the word string.
Document retrieval system according to claim 1, characterized in that less than constituent word of importance not to de identity is assigned. 3. The key in the word string when the keyword feature is a proper noun constituent word.
Document retrieval system according to claim 1, wherein the small <br/> Kusuru it from constituent word of importance not a word identity is assigned. 4. The importance level setting means, if the keyword feature is prefix modification, a keyword in the word string.
Document retrieval system according to claim 1, characterized in that less than constituent word of importance not granted the identity. 5. The importance degree setting means, when the keyword feature is a place name identification word, a keyword in the word string.
Document retrieval system according to claim 1, characterized in that less than constituent word of importance not to de identity is assigned. 6. The importance level setting means, when the keyword feature is an era identification word, a keyword in the word string.
Document retrieval system according to claim 1, characterized in that less than constituent word of importance not to de identity is assigned. 7. The matching degree calculation means increases the matching degree in accordance with the number of constituent words that match in the arrangement order of the constituent words of the keyword and the arrangement order of the constituent words of the search word. Document search method described in 1. 8. The document retrieval system according to claim 1, wherein the coincidence degree calculating means increases the coincidence degree when the keyword and the constituent word at the end of the search term coincide with each other. 9. The document retrieval system according to claim 1, wherein the coincidence degree calculating means increases the degree of coincidence when the keyword and the first constituent word of the search term coincide with each other. 10. The document retrieval method according to claim 1, wherein the document score calculation means calculates the document score by weighting the matching degree of the keyword according to the appearance position of the keyword in the document. 11. The document retrieval method according to claim 1, wherein the document score calculation means calculates the document score by weighting the matching degree of the keyword according to the subsequent word of the keyword.