JP5810046B2 - Document search keyword presentation apparatus, method, and program - Google Patents

Description

  The present invention relates to a document search keyword presentation apparatus, method, and program, and more particularly, to a document search keyword presentation apparatus, method, and program for searching a document from a collection of documents collected from the web or the like.

  Conventionally, document retrieval that retrieves a document related to a given keyword and position by giving a keyword and a position coordinate of a target of interest for a large-scale document such as information obtained from the Internet is known. (For example, refer to Patent Document 1).

  Such a search (geographic information search) can be used for searching information related to a keyword of interest within a range while browsing a map. However, the user may not only seek information related to a specific keyword in the currently viewed map range, but also simply want to obtain characteristic information in that map range. In such a case, if only the conventional geographic information search is used, it is necessary to recall an appropriate search keyword. However, a user with the purpose of obtaining information characteristic of the knowledge in the first place also recalls the search keyword. can not help it.

  As a method in such a case, there is a method of recommending a keyword when it can be said that the region is characteristic based on statistics (see, for example, Non-Patent Document 1). This recommendation was based on the statistical bias between place names, in which a certain keyword appears more frequently in place names within a certain range. For this purpose, the frequency of keyword / place name pairs is counted in advance. However, the criterion for determining “appeared in pairs” in such a case is not necessarily self-evident. In the conventional technology, determination is made based on whether a place name and a keyword appear in the same document, that is, whether they co-occur in the document.

JP 2009-134463 A

Nobuaki Takashima, Yoshihito Yasuda, Naoki Fujita, Ryoji Kataoka, "Presentation of candidate expressions for query input support in geographic information retrieval", 26th National Congress of the Japanese Society for Artificial Intelligence, 2012.

  However, when co-occurrence in a document is targeted, there is a problem that pairs with no semantic connection or with a weak semantic connection are extracted. For example, consider the following sentence:

“Yes, yesterday there was a Sepak Takraw match in Yokosuka. It was a defeat. However, I was satisfied because the yakisoba that I did at home was delicious!”
In this sentence, the {Yokosuka-Yakisoba} pair is not connected semantically, but it is extracted as a pair when co-occurrence in the document is targeted.

  On the other hand, it is limited to co-occurrence in a narrow range such as in the same sentence, or when the place name and keyword are connected by the particle "no", such as "Kagawa no udon". It is also possible to extract. However, in such a case, the number of pairs handled is extremely reduced, and there is a problem that the frequency of the pairs may not be sufficient for statistical processing. This is because place names do not appear frequently in documents, but are often used to relate over a long range with a single appearance. For example, consider the following sentence:

“I have really liked this city since I moved to Yokosuka 3 years ago. It has mountainous topography, many tunnels, and roads tend to be congested. It ’s a mild climate that does n’t require much heating even in the winter.
In this sentence, {Yokosuka-warm} and {Yokosuka-Keihin Electric Railway} pairs are considered to have strong semantic connections, but they are not treated as pairs if they are limited to co-occurrence within a sentence. . For this reason, a sufficient number cannot be obtained to process the frequency of pairs as a statistical bias.

  In summary, if you try to collect high quality pairs, you may not get a sufficient number of pairs for statistical processing, but if you give priority to the number of pairs, There was a problem that would end up collecting low pairs.

  The present invention has been made in view of the above points, and provides a document search keyword presentation apparatus and method capable of calculating recommended keywords based on a sufficient number of statistics without sacrificing the effectiveness of the recommended keywords. And to provide a program.

In order to solve the above problems, the present invention is a document search keyword presentation device that presents keywords according to a map range input by a user,
A document analysis means for extracting a candidate expression A of a keyword from the document set, a position of the candidate expression A in the document, and a place name expression in the document;
For a mesh that is a rectangle obtained by dividing the candidate representation A and the latitude and longitude of the reference map range by a predetermined frequency, the candidate representation A in the co-occurrence range corresponding to the extraction criterion for each of a plurality of extraction criteria having different qualities. And a frequency calculation means for obtaining the candidate expression A or the appearance frequency of the mesh in the co-occurrence frequency or the appearance range of the mesh and storing in the storage means;
Based on the pair of the mesh and candidate expression B included in the attention map range input from the user, the co-occurrence range or the appearance range is applied in order from the extraction criterion that is the narrowest, and the co-occurrence frequency or appearance of the storage means Refer to the frequency, calculate whether the number of pairs has reached a sufficient quantity derived from the value stored in the storage means, and if the score has reached the sufficient quantity, A score calculation unit that adopts a co-occurrence range or the appearance range and, if not reached, widens the co-occurrence range or the appearance range and calculates a score; a pair of the top N scores of the score calculated by the score calculation unit Output means for outputting.

The document analysis means of the present invention includes:
Means for extracting a heading position, a sentence break position, a paragraph break position, a candidate expression in the document, a position of the candidate expression in the document, and a place name expression described in the document of the document set; Including.

Moreover, the frequency calculation means of the present invention includes:
As the extraction criteria,
Co-occurrence of the candidate expression A in the same sentence (first extraction criterion), co-occurrence in the same paragraph (second extraction criterion), one in the headline of the document and the other somewhere in the text In the order of co-occurrence of cases (third extraction criterion), co-occurrence in the same document (fourth extraction criterion),
Based on the first extraction criterion, for a document obtained from the document analysis means, a place name in the same sentence and a co-occurrence frequency in the same sentence of the candidate expression are calculated, and a mesh included in a range implied by the place name; The candidate expression and the co-occurrence frequency are associated with each other and stored in the mesh-candidate expression frequency table of the storage means, the appearance frequency of the candidate expression in a sentence is stored in the candidate expression frequency table of the storage means, and the mesh A frequency calculation means by co-occurrence in the same sentence for storing the appearance frequency in the sentence in the mesh frequency table of the storage means;
Based on the second extraction criteria, for the paragraphs in the document obtained from the document analysis means, the place name in the same paragraph and the co-occurrence frequency in the same paragraph of the candidate expression are calculated and included in the range implied by the place name And the candidate expression and the co-occurrence frequency in the same paragraph are stored in the mesh-candidate expression frequency table, and the appearance frequency in the paragraph in which the candidate expression appears in the same paragraph is stored in the candidate expression frequency table. And a frequency calculation means by co-occurrence in the same paragraph for storing the appearance frequency of the mesh in the paragraph in the mesh frequency table;
Based on the third extraction criterion, a candidate expression that appears in the document heading obtained from the document analysis means and a place name that appears in the document, or a place name that appears in the heading and the document appears in the document. A candidate expression pair heading-text co-occurrence frequency is calculated and stored in the mesh-candidate expression frequency table;
Based on the fourth extraction criterion, for the document obtained from the document analysis means, the co-occurrence frequency in the same document of the mesh included in the range implied by the candidate expression in the document and the place name is the mesh-candidate expression frequency. A frequency calculation means by co-occurrence in the same document storing in the table, storing the appearance frequency of the candidate expression in the document in the candidate expression frequency table, and storing the appearance frequency of the mesh in the document in the mesh frequency table;
Including
The score calculation means includes
As the sufficient quantity, referring to the mesh frequency table, the sum (k) of the weighted sentence appearance frequencies of the meshes included in the attention map range input by the user is obtained, and the extraction criteria are sequentially set from the highest level. Apply the mesh-candidate expression frequency table, the candidate expression frequency table, and the score obtained by referring to the mesh frequency table to output the mesh and candidate expression pair and the score when the score exceeds the k Means to do.

  According to the present invention, it is possible to present a keyword according to the range of the map displayed by the user, so it is possible to present an appropriate keyword according to the current geographical range. In this case, if a sufficient number for repeated statistical processing is obtained from high quality pairs, the result is output at that point, so it is possible to output keywords without depending on low quality pairs Become. In addition, if sufficient numbers are not available, the recommended keywords should be calculated with a statistically sufficient basis for the geographical range in order to apply relatively low quality standards until sufficient numbers are obtained. Is possible.

It is a block diagram of the document search keyword presentation apparatus in one embodiment of this invention. It is an example of the mesh-candidate expression frequency table in one embodiment of the present invention. It is an example of the candidate expression frequency table | surface in one embodiment of this invention. It is an example of the mesh frequency table | surface in one embodiment of this invention. It is a flowchart of the pre-process in one embodiment of this invention. It is a flowchart of this process in one embodiment of the present invention. It is a 2 * 2 contingency table used by the score calculation part in one embodiment of the present invention.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 shows the configuration of a document search keyword presentation device according to an embodiment of the present invention.

  This device may not obtain a sufficient number of pairs for statistical processing when trying to collect the above-mentioned high quality pairs (place names, keywords), while giving priority to the number of pairs. In response to the problem of collecting pairs of low quality, the recommended keywords are calculated based on a sufficient number of statistics without sacrificing the effectiveness of the recommended keywords.

  The apparatus has four pair extraction units with different pairs of high quality. Specifically, the frequency calculation unit 6 by co-occurrence in the same sentence, the frequency calculation unit 7 by co-occurrence in the same paragraph, and the co-occurrence in the headline-text relationship (one is in the headline of the document and the other is the text The frequency calculation unit 8 by co-occurrence in the case of somewhere) and the frequency calculation unit 9 by co-occurrence in the same document are four. The present invention is not limited to the four types of pair extraction criteria, and a plurality of extraction criteria having different qualities can be used. However, in cases other than the extraction criteria shown in the present embodiment, it is desirable to be able to predict in advance whether the quality of a pair that can be extracted based on the criteria is better or worse than other extraction criteria.

  Further, after obtaining the frequency of the pair, a statistical technique for determining whether or not the characteristic is characteristic from the frequency can use a technique known to those skilled in the art. In this embodiment, the bias of two factors is used. The method using the chi-square value as the method according to is described. In addition to this, it is also possible to use a method based on Poisson probability as a method based on a bias in the ratio of one factor as shown in Non-Patent Document 1.

  In order to determine the minimum unit of the range that is the basis for keyword calculation, let us consider a rectangle that divides the entire Japan (or the world) into latitude and longitude that are determined in degrees. This rectangle is hereinafter referred to as “mesh”. The size of the mesh can be set arbitrarily according to the target document set and the nature of the map. In this specification, approximately 500 m square is assumed. It is also conceivable to create a plurality of the apparatuses having different mesh sizes as required and present candidate expressions according to several granularities.

  The apparatus includes a document DB 1, a document analysis unit 2, a mesh-candidate expression frequency table storage unit 3, a candidate table frequency table storage unit 4, a mesh frequency table storage unit 5, and a frequency calculation unit 6 based on co-occurrence within the same sentence. , A frequency calculation unit 7 by co-occurrence in the same paragraph, a frequency calculation unit 8 by co-occurrence in the headline-text relationship, a frequency calculation unit 9 by co-occurrence in the same document, a score calculation unit 10, and an output unit 11.

  The document DB 1 is a database that stores documents that serve as a basis for calculating recommended keywords. The document to be stored is preferably the same as the document searched by the search keyword output by this apparatus. However, for example, when 10 years of newspaper articles are to be searched, the document DB 1 of this apparatus may be a small document set reflecting the entire search target document group, such as 1 year of newspaper articles.

  The mesh-candidate expression frequency table storage unit 3 stores a mesh-candidate expression frequency table. As shown in FIG. 2, the mesh-candidate expression frequency table describes each mesh number, the candidate expression co-occurred with the mesh and any of the co-occurrence extraction criteria, and the weighted co-occurrence frequency with each criterion. It is a table. These values are the frequency calculation unit 6 by co-occurrence in the same sentence, the frequency calculation unit 7 by co-occurrence in the headline-text relationship, the frequency calculation unit 8 by co-occurrence in the same document, and the same document at the time of pre-processing of the apparatus. A value is stored by each unit of the frequency calculation unit 9 by internal co-occurrence.

  The candidate expression frequency table storage unit 4 stores a candidate expression frequency table. As shown in FIG. 3, the candidate expression frequency table is a table in which the number of appearing documents, the number of appearing paragraphs, and the number of appearing documents are described for each candidate expression. Each value in the table is stored by the frequency calculation unit 6 due to co-occurrence in the same sentence, the frequency calculation unit 7 due to co-occurrence in the heading-text relationship, and the frequency calculation unit 9 due to co-occurrence within the same document at the time of pre-processing of the apparatus. Is done.

  The mesh frequency table storage unit 5 stores a mesh frequency table. As shown in FIG. 4, the mesh frequency table describes the weighted frequencies for the number of appearing sentences, the number of appearing paragraphs, and the number of appearing documents for each mesh number. Note that the weighted frequency here refers to the area implied by the place name. For example, if a place name that includes M meshes appears once, the weighted frequency of each mesh is Count as 1 / M times. Each value in the table is stored by the frequency calculation unit 6 due to co-occurrence in the same sentence, the frequency calculation unit 7 due to co-occurrence in the heading-text relationship, and the frequency calculation unit 9 due to co-occurrence within the same document at the time of pre-processing of the apparatus. Is done.

  This apparatus can be divided into pre-processing and main processing for outputting recommended keywords after a geographical range of interest is given.

  FIG. 5 is a flowchart of the preprocessing in the position embodiment of the present invention.

Step 101) The document analysis unit 2 receives a summary target document group from the document DB 1 as an input,
A list of candidate expressions that appear in the document, their position in the document,
A list of place names that appear in the document, their position in the document,
The latitude and longitude of the place names that appear in the document, the geographical range to imply,
The position of the document header,
The position of sentence breaks in the document,
Outputs the paragraph break position in the document. The procedure is shown below.

  First, the heading position, sentence break position, and paragraph break position in the input document group are determined. The position of the headline can be determined by an HTML tag when the input is HTML. In the case of plain text, the head document of each document is regarded as a heading. The sentence break position can be regarded as the position of a punctuation mark (“.”). Alternatively, a determination method based on machine learning known to those skilled in the art may be used. The paragraph break position can be regarded as the position where the line break follows the sentence break. Alternatively, a determination method based on machine learning known to those skilled in the art may be used.

  Next, the candidate expression in the document is specified, and the position of the candidate expression in the document is specified. The method for selecting the candidate expression is arbitrary, but as an example, the specific expression can be used as a candidate expression by using a specific expression extraction technique known to those skilled in the art. Or it is good also considering the word in the list which appeared in the document as a candidate expression by having a brand name and agricultural products as a list, and collating with a document.

  Next, an expression that seems to be a place name described in the document is specified, and the position of the place name in the document, the latitude / longitude of the representative point of the place name, and the range implied by the place name are determined.

  Non-patent document 2 “Toru Hirano, Yoshihiro Matsuo, Genichiro Kikui,“ Disambiguation of place names using geographical distance and famousness ”, The conventional method shown in “National Conference Annual Conference Proceedings, 2008”, etc. can be used.

  As for the implications of the names of various places, Non-Patent Document 3 “Yoshihito Yasuda, Hiroyuki Toda,“ Retrieval of Geographic Information Targeting the Very Near Search Location ”, Journal of the Japanese Society for Artificial Intelligence, Vol. 23, No.5 , pp. 364-373, July 2008 ”, it is possible to obtain the minimum circumscribed rectangle that includes the place name or to use an existing numerical map.

  Step 102) The frequency calculation unit 6 based on co-occurrence within the same sentence performs the following processing for all sentences with the document analyzed by the document analysis unit 2 as an input.

  First, 1 is added to the column of the number of appearance sentences of the entry of the word of the candidate expression frequency table of the candidate expression frequency table storage unit 4 for the candidate expression excluding all duplication in the sentence.

  If at least one place name and at least one candidate expression appear in the sentence, the following processing is performed for all place name-candidate expression pairs composed of the place name and the candidate expression.

  First, a list of mesh numbers included in the range implied by the place name is calculated. 1 / M is added to the sentence appearance frequency in the mesh frequency table storage unit 5. Further, for each mesh, if there is no entry for candidate expression in the mesh-candidate expression frequency table of the mesh-candidate expression frequency table storage unit 3, it is added. When the number of included meshes is M, 1 / M is added to the co-occurrence frequency in the same sentence of the candidate expression in the mesh-candidate expression frequency table of the mesh-candidate expression frequency table storage unit 3.

  Step 103) The frequency calculation unit 7 by co-occurrence within the same paragraph receives the document analyzed by the document analysis unit 2 and performs the following processing for all the paragraphs in the document.

  First, 1 is added to the column of the paragraph appearance frequency of the entry of the word in the candidate expression frequency table of the candidate expression frequency table storage unit 4 for the candidate expression excluding all duplication in the paragraph.

  If at least one place name and at least one candidate expression appear in the paragraph, the following processing is performed for all place name-candidate expression pairs composed of the place name and the candidate expression.

  First, a list of mesh numbers included in the range implied by the place name is calculated. 1 / M is added to the paragraph appearance frequency in the mesh frequency table storage unit 5. For each mesh, if there is no candidate expression entry in the mesh-candidate expression frequency table of the mesh-candidate expression frequency table storage unit 3, it is added. If the number of meshes included at this time is M, 1 / M is added to the co-occurrence frequency in the same paragraph of the candidate expression in the mesh-candidate expression frequency table.

  Step 104) The frequency calculation unit 8 based on the co-occurrence in the headline-text relationship performs the following processing for all the documents analyzed by the document analysis unit 2.

  If the candidate expression appears in the document heading and the place name appears in the document, or the place name appears in the document heading and the candidate expression appears in the document Performs the following processing for all place name-candidate expression pairs composed of these place names and candidate expressions.

  First, a list of mesh numbers included in the range implied by the place name is calculated. For each mesh, if there is no entry for a candidate expression in the mesh-candidate expression frequency table of the mesh-candidate expression frequency table storage unit 3, it is added. If the number of included meshes is M, 1 / M is added to the headline-text co-occurrence frequency of the mesh-candidate expression frequency table.

  Step 105) The frequency calculation unit 9 due to co-occurrence in the same document receives the document analyzed by the document analysis unit 2 and performs the following processing for all the documents in the document.

  First, 1 is added to the column of the document appearance count of the entry of the word in the candidate expression frequency table of the candidate expression frequency table storage unit 4 for the candidate expression excluding all duplications in the document. If at least one place name and at least one candidate expression appear in the document, the following processing is performed for all place name-candidate expression pairs composed of the place name and the candidate expression.

  First, a list of mesh numbers included in the range implied by the place name is calculated. 1 / M is added to the document appearance frequency in the mesh frequency table storage unit 5. For each mesh, if there is no entry for a candidate expression in the mesh-candidate expression frequency table of the mesh-candidate expression frequency table storage unit 3, it is added. If the number of meshes included at this time is M, 1 / M is added to the co-occurrence frequency in the same document of the candidate expression in the mesh-candidate expression frequency table.

  Next, this process will be described.

  FIG. 6 is a flowchart of this processing in one embodiment of the present invention.

  Step 201) The score calculation unit 10 receives the attention geographic range indicated by the user as an input, and outputs a score indicating whether or not the candidate expression appears characteristically biased to the attention range to the output unit 11. The procedure is shown below.

  Co-occurrence within the same sentence, co-occurrence within the same paragraph, co-occurrence when one is a document heading and the other is somewhere in the body, co-occurrence within the same document, α, β, γ, and δ are set in advance. In addition, a frequency parameter sufficient for the basis is set as k and is determined in advance.

  First, it is determined whether there is a candidate expression that appears statistically biased by the following procedure for co-occurrence in the same document.

  1. The basis frequency number z is initialized (z = 0).

  2. Let n be the total number of sentences (n = 0).

  3. With reference to the mesh frequency table of the mesh frequency table storage unit 5, the sum (k) of the weighted sentence appearance frequencies for the meshes included in the noted geographic range given as input is obtained.

  4). With reference to the mesh-candidate expression frequency table in the mesh-candidate expression frequency table storage unit 3, a set I of candidate expressions i appearing in the mesh included in the geographic region of interest given as an input is obtained.

  5. The following is repeated for each candidate expression i in set I:

5-1. The sentence expression frequency s _i of _i is obtained by referring to the candidate expression frequency table in the candidate expression frequency table storage unit 4.

5-2. With reference to the mesh-candidate expression frequency table in the mesh-candidate expression frequency table storage unit 3, the sum r _i of the weighted sentence appearance frequencies of i within the geographic region of interest is obtained.

5-3. Additional n, _k, s i, by using the _{r i,} the value A of a 2 × 2 contingency table as in FIG. 7, to calculate B, C, by the following equation D.

  5-5. Let the score S (i) of i be S (i) = αX (i).

6). For all candidate i described above, after obtaining the X (i), is added r _i X (i) is the frequency within the target range of i which was maximal in z.

  In the above procedure, if z exceeds the parameter k, the process is terminated, a list of pairs of S (i) and i is output, and the process of the score calculation unit 10 is terminated.

  If z does not exceed k, the same processing as described above is performed for co-occurrence within the same paragraph, which is the next extraction criterion. However, the score S (i) of i is a value obtained by adding βX (i) to the previous score, that is, S (i) = S (i) + βX (i).

  Similarly, the process is repeated until z exceeds, and a list of S (i) and i pairs in which the final extraction criterion, i.e., co-occurrence in the same document, is output.

  In the case of co-occurrence in the headline-text relationship, the column of the document appearance frequency is used when referring to the candidate expression frequency table and the mesh frequency table.

  The output unit 11 outputs, as an output of the present apparatus, the number N obtained in advance in descending order of the candidate expressions obtained by the score calculation unit 10 and the scores in descending order.

  As described above, the present invention recommends that a pair of place names and candidate expressions has a small quantity but a high quality pair, a large quantity but a low quality pair, and recommends a keyword corresponding to the place name. Focusing on the fact that a sufficient number is necessary as a statistical basis, it is possible to determine high-quality recommended keywords by applying a plurality of criteria in stages.

  The present invention constructs the operation of the constituent elements of the document search keyword presenting apparatus shown in FIG. 1 as a program and installs and executes it on a computer used as the document search keyword presenting apparatus, or distributes it via a network. Is possible.

  The present invention is not limited to the above-described embodiments, and various modifications and applications can be made within the scope of the claims.

1 Document DB
2 Document analysis unit 3 Mesh-candidate expression frequency table storage unit 4 Candidate expression frequency table storage unit 5 Mesh frequency table storage unit 6 Frequency calculation unit 7 by co-occurrence in the same document 8 Frequency calculation unit 8 by co-occurrence in the same paragraph Frequency calculation unit 9 by co-occurrence in relation Frequency calculation unit 10 by co-occurrence in the same document Score calculation unit 11 Output unit

Claims (5)

A document search keyword presentation device that presents keywords according to a map range input by a user,
A document analysis means for extracting a set of keyword candidate expressions A from the document set, a position of each candidate expression A in the document, and a place name in the document;
For each of a plurality of extraction criteria with different corresponding document ranges , a rectangle obtained by dividing the latitude and longitude of the reference map range by a predetermined frequency for each candidate expression A and each place name in the document range corresponding to the extraction criteria. In the mesh, the co-occurrence frequency is obtained for each mesh implied by the name of each place, the appearance frequency of each candidate expression A in the document range is obtained, and the appearance frequency for each mesh implied by the place name in the document range is obtained. , A frequency calculation means for storing the co-occurrence frequency and the appearance frequency in a storage means;
Based on the pair of mesh and candidate expression B included in the attention map range input by the user, the document range is applied in order from the narrowest extraction criterion, and the co-occurrence frequency and appearance frequency of the storage means are referred to When the basis frequency number z indicating whether the number of pairs has reached a sufficient quantity derived from the value stored in the storage means is calculated, and the basis frequency number z has reached the sufficient quantity adopting the document range widely documents range has not been reached, widely in document scope, score indicates whether a candidate expression B has appeared biased in characteristic of the attention map range S Score calculating means for calculating
Output means for outputting a pair having a relatively high score S calculated by the score calculating means;
A document search keyword presentation device characterized by comprising: The extraction criterion with the narrowest document range is the occurrence or co-occurrence within the same sentence,
The frequency calculating means includes
In accordance with the number of meshes implied by the place name, the co-occurrence frequency associated with the mesh is weighted, and the mesh, the candidate expression A, and the weighted co-occurrence frequency are associated with each other in the mesh-candidate expression frequency table of the storage means. Store and
The candidate expression A and the appearance frequency are associated with each other and stored in the candidate expression frequency table of the storage unit,
According to the number of meshes implied by the place name, the appearance frequency associated with the mesh is weighted, and the mesh and the weighted appearance frequency are associated with each other and stored in the mesh frequency table of the storage means.
The document search keyword presenting apparatus according to claim 1 . A document search keyword presentation method for presenting a keyword corresponding to a map range input by a user,
Computer
Set of candidate expressions A keyword from the document set position in a document for each candidate representations A, and a document analysis step of extracting the place name in the document,
For each of a plurality of extraction criteria with different corresponding document ranges , a rectangle obtained by dividing the latitude and longitude of the reference map range by a predetermined frequency for each candidate expression A and each place name in the document range corresponding to the extraction criteria. In the mesh, the co-occurrence frequency is obtained for each mesh implied by the name of each place, the appearance frequency of each candidate expression A in the document range is obtained, and the appearance frequency for each mesh implied by the place name in the document range is obtained. A frequency calculating step of storing the co-occurrence frequency and the appearance frequency in a storage means;
Based on the pair of mesh and candidate expression B included in the attention map range input by the user, the document range is applied in order from the narrowest extraction criterion, and the co-occurrence frequency and appearance frequency of the storage means are referred to When the basis frequency number z indicating whether the number of pairs has reached a sufficient quantity derived from the value stored in the storage means is calculated, and the basis frequency number z has reached the sufficient quantity adopting the document range widely documents range has not been reached, widely in document scope, score indicates whether a candidate expression B has appeared biased in characteristic of the attention map range S A score calculating step for calculating
An output step of outputting a pair having a relatively high score S calculated in the score calculating step;
The document search keyword presentation method characterized by performing this . The extraction criterion with the narrowest document range is the occurrence or co-occurrence within the same sentence,
The frequency calculating step includes:
In accordance with the number of meshes implied by the place name, the co-occurrence frequency associated with the mesh is weighted, and the mesh, the candidate expression A, and the weighted co-occurrence frequency are associated with each other in the mesh-candidate expression frequency table of the storage means. Store and
The candidate expression A and the appearance frequency are associated with each other and stored in the candidate expression frequency table of the storage unit,
According to the number of meshes implied by the place name, the appearance frequency associated with the mesh is weighted, and the mesh and the weighted appearance frequency are associated with each other and stored in the mesh frequency table of the storage means.
The document search keyword presentation method according to claim 3 . Computer
Document retrieval keyword presentation program for functioning as each unit of the document search keyword presentation apparatus according to claim 1 or 2.

Images