JP4935243B2 - Search program, information search device, and information search method - Google Patents

Description

The present invention relates to a search program, an information search device, and an information search method for text such as a web page, and in particular, a search program that automatically generates a candidate query and makes it possible to search again when a search fails due to a query input by a user, The present invention relates to an information search apparatus and an information search method.

  Conventionally, when searching for a document such as a web page, an interactive information search is performed in which a user creates a query according to a search purpose and performs a search.

  Here, a query represents a query such as data extraction or update for a database management system as a character string. For example, a query such as “What is the weather in Tokyo today?” It is also used for searching web pages.

According to the information search such as a web page using such a query, the content that the user wants to search can be expressed more accurately than the search that uses one or more words as keywords, and the content that the user expects Narrowed search results are obtained, and it is possible to eliminate the trouble of the user searching for a desired search result while opening a page again from a large number of search results as in keyword search.

JP-A-11-053382 JP 2004-139154 A JP 2005-251092 A

  However, in such interactive information search using conventional queries, skill is required to create an appropriate query, and if the query is not appropriate, the number of search results is very small. Alternatively, there is a problem that it takes time and labor to reach the required page because there are no search results.

  The reason why no query is found in the query input in this way, that is, the cause of the so-called zero hit is that the query is too long. In general search engines, pages containing the query string entered are returned as search results. For example, when a long query such as "What is the weather in Tokyo today?" When there is no document containing the same character string as that character string, 0 hits are found.

  The present invention makes it possible to control the number of search hits by presenting an appropriate candidate query to the user when the query inputted by the user is not appropriate, and to guide the user to an appropriate page with less effort and time. A search program, an information search device, and an information search method are provided.

The present invention also provides a search program, an information search device, and an information search method that improve the skill of creating a user query by presenting candidate queries to the user based on past query conversion history.

(program)
The present invention provides a search program. The search program of the present invention is stored in a computer.
A query input step for entering a query specified by the user;
A search step to search using a query;
If the search fails in the search step, a candidate query generation step that presents to the user a plurality of candidate queries generated by decomposing the query and causes the search step to search using the query selected by the user;
And execute
Furthermore, the candidate query generation step sets in advance any one of extraction of nouns, extraction of only nouns and verbs, and extraction of noun phrases as a decomposition method of the pre-decomposition query, and pre-decomposition based on the set decomposition conditions. A combination of a phrase and a phrase extracted from the query is generated as a candidate query.

  The candidate query generation step executes a candidate query generation process by determining that the search has failed when no search result is obtained by the search step and no hit is found.

The search program of the present invention further includes a computer,
For the set of the pre-decomposition query input in the query input step and the post-decomposition query selected by the user in the candidate query generation step, the step of extracting each feature quantity and registering it in the query decomposition knowledge database is executed,
In the candidate query generation step, each feature amount is extracted for a set of pre-decomposition query and post-decomposition query to be processed, and each feature amount of the pre-decomposition query and post-decomposition query pair registered in the knowledge database is extracted. A score is calculated from the similarity and given to the post-decomposition query to be processed, and the post-decomposition query sorted in order of score is presented to the user as a candidate query for selection.

The search program of the present invention is further stored in a computer.
The feature of the post-decomposition query that was successfully searched in the search step is extracted and registered in the query selection knowledge database.
In the candidate query generation step, the feature quantity of the decomposed query to be processed is extracted, the score is calculated from the similarity to the feature quantity with the decomposed query registered in the query selection knowledge database, and the processing target is decomposed A post-decomposition query that is given to the post-query and sorted in the order of score is presented to the user as a candidate query for selection.

The feature amount of the query before decomposition includes one or a plurality of combinations of the constituent element subjected to morphological analysis, the number of words, and the content word having a specific meaning,
The feature value of the query after decomposition is the decomposition method of extracting nouns only, extracting nouns and verbs only, or extracting noun phrases, the position of extracted words on the query before decomposition, the position of extracted words, the number of extracted words 1 or a combination of one or more semantic labels indicating the meaning of the extracted word, or the ratio of the number of elements after decomposition to the number of words in the query before decomposition.

(apparatus)
The present invention provides an information retrieval apparatus. The information retrieval apparatus of the present invention is
A query input part for inputting a query specified by the user;
A search part that searches using a query;
If the search unit fails to search, a candidate query generation unit that presents a plurality of candidate queries generated by disassembling the query and causes the search unit to search using a query selected by the user;
With
Further, the candidate query generation unit presets any of noun extraction, noun and verb extraction, or noun phrase extraction as a decomposition method of the pre-decomposition query, and the pre-decomposition query based on the set decomposition method A phrase extracted from the above and a combination of phrases are generated as candidate queries.

(Method)
The present invention provides an information retrieval method. The information retrieval method of the present invention is
A query input step for entering a query specified by the user;
A search step to search using a query;
If the search fails in the search step, a candidate query generation step of presenting a plurality of candidate queries generated by decomposing the query to the user and causing the search step to search using the query selected by the user;
With
In addition, the candidate query generation step presets any of noun extraction, noun and verb extraction, and noun phrase extraction as a decomposition method of the pre-decomposition query, and the pre-decomposition query based on the set decomposition conditions A phrase extracted from the above and a combination of phrases are generated as candidate queries.

  According to the present invention, when a query input by a user whose search result is 0 hits is not appropriate, the query input by the user is decomposed and decomposed based on, for example, a decomposition method of extracting only nouns and verbs. A word or a combination thereof is generated as a candidate query and presented to the user, and a search result that leads the user to an appropriate page, for example, is obtained with less effort and time (less cost) by re-searching with the candidate query selected by the user. be able to.

Also, the combination of the pre-decomposition query and post-decomposition query selected by the past user and the history of the post-decomposition query that was successfully searched are saved as a knowledge database, and when a new candidate query is generated, By assigning the cost according to the similarity to the candidate query and presenting it to the user, the candidate query can be evaluated by the cost, and the user's query is created by the learning effect through the process of selecting the high-cost candidate query. Skills can be improved.

  FIG. 1 is a block diagram of a functional configuration showing an embodiment of an information retrieval apparatus according to the present invention. In FIG. 1, the information search apparatus of the present embodiment includes a query input unit 10, a search control unit 12, a search engine 14 that functions as a search unit, a search result display unit 18, a candidate query processing unit 20, and a decomposition candidate display designation unit 22. A knowledge recording unit 24, a query decomposition knowledge database 30, and a query selection knowledge database 32.

  Further, the candidate query processing unit 20 is provided with functions of a decomposition candidate generation unit 34, a score assignment unit 36, and a decomposition candidate presentation unit 38. The knowledge recording unit 24 includes a query decomposition knowledge recording unit 26 and a query selection knowledge recording unit 28.

  The query decomposition knowledge recording unit 26 performs registration processing for the query decomposition knowledge database 30 and includes a pre-decomposition query feature extraction unit 40, a search success query feature extraction unit 42, and a database registration unit 44. The query selection knowledge recording unit 28 performs registration with respect to the query selection knowledge database 32 and includes a search success query feature extraction unit 46 and a database registration unit 48.

  The query input unit 10 inputs a query that is a character string for search specified by the user. The search control unit 12 outputs the query input from the query input unit 10 to the search engine 14, and in response to this, the search engine 14 searches a web page existing on the Internet 16, and the search result is displayed as a search result display unit. 18 will be displayed.

  When the search result by the search engine 14 using the input query from the query input unit 10 fails, the search control unit 12 hits 0 hits without obtaining any search results in this embodiment. If it does, the candidate query processing unit 20 operates.

  The candidate query processing unit 20 generates a plurality of candidate queries by decomposing the query input from the query input unit 10 when the search engine 14 hits 0 hits, and the candidate query is displayed as a decomposition candidate display designation unit. The search engine 14 performs a search using the query selected by the user and selected by the user.

The candidate query generation by the candidate query processing unit 20 sets, for example, the following method as a decomposition method of the query (input query) before decomposition.
(1) Extract nouns only (2) Extract nouns and verbs only (3) Extract noun phrases

  By setting one of the decomposition methods in advance for such a decomposition method, a phrase is extracted from the pre-decomposition query based on the set decomposition method, and a combination of the extracted word and phrase is generated as a candidate query. . This process is performed by the decomposition candidate generation unit 34.

  The score assigning unit 36 assigns a score to each decomposition candidate generated by the decomposition candidate generating unit 34 using the history knowledge stored in the query decomposition knowledge database 30 or the query selection knowledge database 32. The decomposition candidate presentation unit 38 sorts the decomposition candidates to which scores have been assigned in the order of the scores, and then displays them on the decomposition candidate display designation unit 22 to receive user designation.

  Here, in the query decomposition knowledge database 30, each feature amount is set for a pair (pair) of the query before decomposition input by the query input unit 10 and the query generated by the candidate query processing unit 20 and selected by the user. Extracted and registered as query decomposition knowledge. Registration with the query decomposition knowledge database 30 is performed by a query decomposition knowledge recording unit 26 provided in the knowledge recording unit 24.

  On the other hand, in the query selection knowledge database 32, the feature amount of the decomposed query that has been successfully searched by the search by the search engine 14 is extracted and registered as query selection knowledge. The registration of the query selection knowledge database 32 is performed by the query selection knowledge recording unit 28 provided in the knowledge recording unit 24.

  The score assignment by the score assigning unit 36 using the query decomposition knowledge database 30 is performed by extracting each feature amount for a set of a query before decomposition and a query after decomposition that are currently processed and registering them in the query decomposition knowledge database 30. A score is calculated from the similarity between the pre-decomposition query and the feature values of the post-decomposition query, and is given to the post-decomposition query that is a decomposition candidate.

  In addition, in the score assignment using the query selection knowledge database 32 by the score assignment unit 36, the feature amount of the decomposed query that has succeeded in the search to be processed is extracted, and the decomposition registered in the query selection knowledge database 32 A score is calculated and given from the similarity to the feature value of the post-query.

  FIG. 2 is a block diagram of a hardware environment of a computer in which the search program of this embodiment that implements the functions of the embodiment of FIG. 1 is executed. In FIG. 2, a RAM 54, a ROM 55, a hard disk drive 56, a keyboard 60, a mouse 62 and a device interface 58 for connecting a display 64 and a network adapter 66 connected to the Internet are connected to the bus 52 of the CPU 50. .

  The search program according to the present embodiment is stored in the hard disk drive 56. When the computer is started, the boot program is read from the hard disk drive 56 to the RAM 54 and executed by the operation of the BIOS. After the OS is read from the RAM 54 to the RAM 54 and executed, the search program of the present embodiment is read from the hard disk drive 56 to the ROM 54 and executed by the CPU 50.

  FIG. 3 is a flowchart of the information search process according to this embodiment. In FIG. 3, the process will be described with reference to the embodiment of FIG.

  First, in step S1, an input query created by the user is read from the query input unit 10, and in step S2, a search engine 14 searches a web page on the Internet 16 using the input query.

  For this search result, it is determined in step S3 whether or not there are zero hits. If it is not 0 hits, the process proceeds to step S9, the search result is output to the search result display unit 18, and the query selection knowledge recording process provided in the knowledge recording unit 24 in step S10 is used for query selection knowledge recording processing. This is performed on the knowledge database 32.

  On the other hand, if the search fails in step S3 and 0 hits are determined, the process proceeds to step S4, the candidate query generation process is performed by the candidate query processing unit 20, and the candidate query is displayed on the decomposition candidate display designating unit 22. To do.

  Subsequently, when the user's decomposition candidate selection is determined in step S5, the process proceeds to step S6, and a search using the candidate query that is the decomposition query selected by the user is performed by the search engine 14, and the search result is displayed in the search result display unit in step S7. 18 is output. Subsequently, in step S8, the query decomposition knowledge recording unit 26 provided in the knowledge recording unit 24 executes query decomposition knowledge recording processing, and registers the processing result in the query decomposition knowledge database 30.

  FIG. 4 is a flowchart showing details of the query decomposition knowledge recording process in step S8 of FIG. In FIG. 4, in the query decomposition knowledge recording process, the feature amount of the pre-decomposition query (input query) that is currently processed in step S1 is extracted, and the feature amount of the decomposition query selected by the user in step S2 is then extracted. In step S3, a pair of the feature quantity of the query before decomposition and the feature quantity of the query after decomposition is registered in the query decomposition knowledge database 30 as query decomposition knowledge.

  FIG. 5 is an explanatory diagram showing a specific processing operation of the query decomposition knowledge recording unit 26 of FIG. 1 corresponding to FIG. In FIG. 5, “What will happen to the weather in Tokyo today?” Is processed as a pre-decomposition query 68 created and input by the user. A search is performed by selecting “today + Tokyo region + weather” and the search is successful, and this is obtained as a search success query 70.

  In the query decomposition knowledge recording unit 26, first, the feature amount 72 of the pre-decomposition query 68 is extracted by the pre-decomposition query feature extraction unit 40. As the feature amount 72 of the query 68 before decomposition, elements included in the character string are classified by morphological analysis, and each part of speech is extracted.

  Further, the “number of words” constituting the character string of the query 68 before decomposition is extracted, and in this case, “number of words: 9”. Further, the “number of content words” is extracted for the content words having a specific meaning included in the pre-input query 68, and in this case, “number of content words: 3”. Further, “predicate” is extracted as a phrase at the end of the sentence of the query 68 before decomposition.

  Next, the search success query feature quantity extraction unit 42 extracts the feature quantity 74 from the search success query 70. Extraction of the feature quantity 74 in the search success query 70 is, as a decomposition method of the search success query 70, “noun only” in this example, the position of each word of the search success query 70 in the pre-decomposition query 68, in this embodiment. Since “Today” is the first, “Tokyo region” is the third, and “weather” is the fifth, “position (word): 1, 3, 5” is extracted.

  Further, “position (content word): 1, 2, 3” is extracted as the position of each word in the search success query 70. Further, “word count: 3” is extracted as the word count of the search success query 70. Further, “3/9” is extracted as the ratio of the number of elements in the search success query 70 (number of elements / number of words). Further, in this example, “LOCATION” is extracted corresponding to “Tokyo region” as the meaning label of the word in the search success query 70.

  Extraction of the feature amounts 72 and 74 shown in FIG. 5 is an example, and other appropriate features regarding the character strings in the pre-decomposition query 68 and the search success query 70 can be extracted as necessary.

  FIG. 6 is a flowchart showing details of the query selection knowledge recording process in step S10 of FIG. In FIG. 6, the query selection knowledge recording processing unit 28 first checks whether or not the search success query is a post-decomposition query in step S1 in a state where the search is successful. If the query is a post-decomposition query, the process proceeds to step S2, the feature quantity of the post-decomposition query is extracted, and the feature quantity extracted in step S3 is registered in the query selection knowledge database 32. On the other hand, if the search success query is a pre-decomposition query, that is, an input query in step S1, the feature amount extraction and registration in steps S2 and S3 are skipped, and the query selection knowledge is not registered.

  FIG. 7 is an explanatory diagram showing a specific processing operation by the query selection knowledge recording unit 28 of FIG. 1 corresponding to FIG. In FIG. 7, “Today + Tokyo region + weather”, which is a post-decomposition query, is targeted as the search success query 70, and the feature quantity is extracted by the search success query feature extraction unit 46. To get. The content of the feature quantity 74 is the same as the feature quantity 74 extracted by the search success query feature extraction unit 42 of FIG. Subsequently, the database registration unit 48 registers the feature quantity 74 in the query selection knowledge database 32.

FIG. 8 is a flowchart showing details of the candidate query generation process in step S4 of FIG. In FIG. 8, in the candidate query generation process, first, a candidate query decomposition method preset in step S1 is acquired. This decomposition method is either (1) noun extraction only, (2) noun and verb extraction only, or (3) noun phrase extraction only. Then, it progresses to step S2 and decomposes | disassembles a candidate query using the decomposition | disassembly method acquired by step S1 by the decomposition | disassembly candidate production | generation part 34, and the decomposition | disassembly candidate which consists of a word and the combination of a word is produced | generated. Next, in step S3, the feature quantities of the pre-decomposition query and the decomposition candidate are extracted, and in step S4, a score based on the similarity between the stored knowledge in the query decomposition knowledge database 30 is calculated for each pair of the pre-decomposition query and the post-decomposition query. To do.

  The feature amount extraction in step S3 and the score calculation in step S4 are performed by the score assigning unit 36. Finally, after the decomposition candidate presentation unit 38 sorts in order of score in step S5, the decomposition candidates are presented to the user.

  FIG. 9 is an explanatory diagram showing a specific processing operation of the candidate query processing unit 20 of FIG. 1 corresponding to FIG. In FIG. 9, “What is the weather in Kawasaki today?” Is input to the decomposition candidate generation unit 34 as the pre-decomposition query 76. For example, “Today”, “Kawasaki”, “Weather”, “Today's Kawasaki”, “Today's Kawasaki”, “Today + Kawasaki”, “Today” A decomposition candidate list 78 that generates decomposition candidates such as “+ Kawasaki + weather” is generated.

  Here, up to “Kawasaki weather” in the decomposition candidate list 78, the score obtained by the score calculation by the score assigning unit 36 is already given as indicated by the numerical value in parentheses.

  Next, the case where the score is calculated for the decomposition candidate 80 will be described as follows. First, the feature value 76-1 is extracted from the pre-decomposition query 76. Since this is performed in advance at the time of the first decomposition candidate “today”, it is used.

  On the other hand, the feature quantity 80-1 is extracted for “today + kawasaki”, which is the decomposition candidate 80 currently being processed. The extraction process of the feature amount 80-1 is the same as the process performed by the pre-decomposition query feature extraction unit 40 and the search success query feature extraction unit 42 provided in the query decomposition knowledge recording unit 26 shown in FIG.

  When the feature amount 76-1 of the pre-decomposition query 76 and the feature amount 80-1 of the decomposition candidate 80 currently being processed are obtained in this way, the feature amount 76-1 and the feature amount 80-1 are paired. The similarity with each of the stored knowledge 82-1 to 82-3 registered at this time in the query decomposition knowledge database 30 is calculated.

  In this example, the similarities sim1 and sim2 of the feature quantity 76-1 and the feature quantity 80-1 are respectively determined between the pair of the pre-decomposition query feature quantity 84 and the post-decomposition query feature quantity 86 included in the stored knowledge 82-1. A score between the saved knowledge 82-1 is obtained as a result of multiplication between the two. Such score calculation is performed between the remaining stored knowledge 82-2 and 82-3, the sum is obtained as a score, and the calculated score is given to the decomposition candidate 80.

The similarity calculation 88 in the score assigning unit 36 is performed by the following equation.

  When the score assignment unit 36 finishes assigning scores to each of the decomposition candidate lists 78, the decomposition candidate list 78 is sorted in the order of score, and then the decomposition candidate list 78 is displayed on the decomposition candidate display designating unit 22 and the user selects a decomposition candidate. You will be asked to specify.

  FIG. 10 is a flowchart showing another embodiment of the candidate query generation process in step S4 of FIG. 3. In this embodiment, the query selection knowledge database 32 of FIG. It is characterized by providing the above.

  10, steps S1, S2, and S3 of the candidate query generation process are the same as when the query decomposition knowledge database 30 of FIG. 8 is used, but in step S4, the query selection knowledge database for each decomposition candidate. The score based on the similarity to the stored knowledge of 32 is calculated and assigned, and in step S5, the score is sorted in the order of the score and the user is presented with the decomposition candidates to be selected.

  FIG. 11 shows a specific operation of another embodiment by the candidate query processing unit 20 of FIG. 1 corresponding to FIG. In FIG. 11, as in the case of FIG. 9, the decomposition candidate generation unit 34 generates a decomposition candidate list 78 by applying the decomposition method set in the pre-decomposition query 76. In the decomposition candidate list 78, score assignment has been completed up to "Kawasaki weather" on the fourth line, and the score calculation of "Today + Kawasaki", which is the next decomposition candidate 80, is taken as an example.

  This score calculation is performed by the score assigning unit 36. The score assigning unit 36 extracts the feature value 80-1 from the target decomposition candidate 80, and calculates the similarity with each of the stored knowledge 90-1 to 90-3 registered in the query selection knowledge database 32. Calculate and calculate the sum as a score.

  For example, for the saved knowledge 90-1, the search success query feature value 92 is stored, and the similarity calculation 94 is performed between this and the feature value 80-1 of the decomposition candidate 80 to obtain the score sim3. Then, such similarity calculation is performed with the remaining stored knowledge 90-2 and 90-3, and the sum is used as a score. The similarity calculation 94 in this case is performed by the following equation.

  FIG. 12 is an explanatory diagram showing another specific example of the processing operation by the query decomposition knowledge recording unit 26 of FIG. In FIG. 12, in this specific example, the query 96 before decomposition is “the telephone number of the disaster medical center in Tachikawa”, and the search success query 98 is “Tachikawa + disaster medical center + phone number”. The pre-decomposition query 96 and the search success query 98 are input to the query decomposition knowledge recording unit 26 for processing.

  First, the pre-decomposition query feature extraction unit 40 extracts the feature quantity 100 from the pre-decomposition query 96. Subsequently, the search success query feature extraction unit 42 extracts the feature quantity 102 from the search success query 98. The database registration unit 44 registers the feature quantities 100 and 102 extracted in this way as a pair in the query decomposition knowledge database 30 as query decomposition knowledge.

  FIG. 13 is an explanatory diagram showing another specific example of the processing operation by the query selection knowledge recording unit 28 of FIG. In FIG. 13, “Tachikawa + disaster medical center + phone number” is targeted as the search success query 104, and the search success query feature extraction unit 46 processes this search success query 104 to extract the feature quantity 102. .

  This feature quantity 102 is the same as the feature quantity by the search success query feature extraction unit 46 of FIG. Subsequently, the database registration unit 48 registers the extracted feature quantity 102 in the query selection knowledge database 32.

  FIG. 14 is an explanatory diagram showing another specific example of the processing operation by the candidate query processing unit 20 of FIG. In FIG. 14, “disaster medical center telephone number in Tachikawa” is targeted as the pre-decomposition query 96, and the decomposition candidate list 105 is first generated by the decomposition candidate generation unit 34.

  Subsequently, the score assigning unit 36 extracts the feature quantity 96-1 from the pre-decomposition query 96, and also extracts the feature quantity 106-1 from “Tachikawa + phone number” which is the decomposition candidate 106 to be processed. Subsequently, each saved pair of saved knowledge 108-1, 108-2, 108-3 registered in the query decomposition knowledge database 30, for example, in the saved knowledge 108-1, is decomposed with the query feature quantity 110 before decomposition. Similarities sim1 and sim2 are calculated and multiplied by the similarity calculation 114 with the post-query feature quantity 111, and this is repeated for all stored knowledge.

  Specifically, the score is obtained by performing the similarity calculation shown in the equation (1) and given to the decomposition candidate 106. If the score calculation is completed for all the decomposition candidates, the decomposition candidate presenting unit 38 sorts them in the order of the scores and presents them to the decomposition candidate display designating unit 22 to receive the user's designation.

  FIG. 15 is an explanatory diagram showing another specific example of the processing operation according to another embodiment of the candidate query processing unit 20 of FIG. In FIG. 15, score calculation is performed using the query selection knowledge database 32, and the decomposition candidate list 105 is generated by the decomposition candidate generation unit 34 for the pre-decomposition query 96 as in FIG. 14.

  Subsequently, the score assigning unit 36 extracts the feature quantity 106-1 from the decomposition candidate 106 that is currently processed, and the query selection knowledge of the stored knowledge 116-1 to 116-3 stored in the query selection knowledge database 32, for example, The similarity calculation 120 is performed with the search success query feature value 118 in the stored knowledge 116-1, and this is performed for all the stored knowledge 116-1 to 116-3 to execute the above (1) to obtain the sum. To obtain a score and assign it to the decomposition candidate 106.

  When the scores of all the decomposed character strings have been calculated, after sorting in the order of score by the decomposition candidate presenting section 38, they are displayed on the decomposition candidate display designating section 22 and selected by the user.

  Further, the present invention provides a program for information retrieval, and this program has the processing contents shown in the flowcharts of FIGS. 3, 4, 6, 8, and 10.

  The present invention also provides a recording medium storing a computer-readable search program. Here, portable storage media such as CD-ROM, floppy disk (R), DVD disk, magneto-optical disk, and IC card, and storage devices such as hard disk drives provided inside and outside the computer system, as well as via lines It includes a database holding the program, other computer systems, the database, and a transmission medium on a line.

  Further, the function as the information retrieval apparatus of the present invention shown in FIG. 1 may be incorporated into a personal computer that is connected to a network in a stand-alone manner and used personally, or installed in an internal network in a company. A search process may be executed in response to a search request using a query from a client connected to the server and returned to the search result.

  The present invention includes appropriate modifications that do not impair the objects and advantages thereof, and is not limited by the numerical values shown in the above embodiments.

Here, the features of the present invention are enumerated as follows.

(Appendix)
(Appendix 1)
On the computer,
A query input step for entering a query specified by the user;
A search step for searching using the query;
If the search fails in the search step, a candidate query generation step of presenting a plurality of candidate queries generated by decomposing the query to the user and causing the search step to search using the query selected by the user;
And execute
Further, in the candidate query generation step, any one of noun extraction, noun and verb extraction, and noun phrase extraction is set in advance as a decomposition method of the pre-decomposition query, and decomposition is performed based on the set decomposition condition. A search program that generates a combination of a phrase and a phrase extracted from a previous query as a candidate query. (1)

(Appendix 2)
In the search program according to attachment 1, the candidate query generation step executes a candidate query generation process by determining that the search has failed when no search result is obtained by the search step and no hit is found. A search program characterized by

(Appendix 3)
In the search program according to appendix 1, each feature amount is further extracted from the computer into a set of the pre-decomposition query input in the query input step and the post-decomposition query selected by the user in the candidate query generation step. And execute the step of registering in the query decomposition knowledge database,
The candidate query generation step extracts each feature quantity for a set of pre-decomposition query and post-decomposition query to be processed, and each feature quantity of the pre-decomposition query and post-decomposition query registered in the knowledge database A search program characterized in that a score is calculated from the similarity to and given to the post-decomposition query to be processed, and the post-decomposition query sorted in the order of the score is presented to the user as a candidate query and selected. (2)

(Appendix 4)
In the search program according to appendix 1, the computer is further caused to execute a step of extracting a feature quantity of the decomposed query that has been successfully searched in the search step and registering it in a query selection knowledge database,
The candidate query generation step extracts a feature amount of a post-decomposition query to be processed, calculates a score from a similarity with a post-decomposition query registered in the query selection knowledge database, and performs the processing A search program characterized in that a post-decomposition query that is given to a target post-decomposition query and sorted in the order of the score is presented to a user as a candidate query and selected. (3)

(Appendix 5)
In the search program described in Appendix 1 or 2,
The feature amount of the query before decomposition includes one or a plurality of combinations of a constituent element subjected to morphological analysis, the number of words, and a content word having a specific meaning,
The feature value of the post-decomposition query is a noun only extraction, noun and verb only extraction or noun phrase extraction decomposition method, extracted word position on the query before decomposition, extracted word position, number of extracted words , A ratio of the number of elements after decomposition to the number of words of a query before decomposition, or a combination of one or more meaning labels indicating the meaning of extracted words.

(Appendix 6)
A query input part for inputting a query specified by the user;
A search unit for searching using the query;
A candidate query generation unit that presents to the user a plurality of candidate queries generated by decomposing the query when the search fails in the search unit, and causes the search unit to search using the query selected by the user;
With
Further, the candidate query generation unit preliminarily sets any one of extraction of nouns, extraction of only nouns and verbs, and extraction of noun phrases as a decomposition method of the query before decomposition, and decomposes based on the set decomposition method. An information search apparatus that generates a combination of a phrase and a phrase extracted from a previous query as a candidate query. (4)

(Appendix 7)
In the information search device according to attachment 6, the candidate query generation unit performs a candidate query generation process by determining that the search has failed when no search result is obtained by the search unit and no hit is found. An information retrieval apparatus characterized by that.

(Appendix 8)
The information search device according to appendix 6, wherein each feature is extracted and registered for a set of the pre-decomposition query input by the query input unit and the post-decomposition query selected by the user by the candidate query generation unit. Query decomposition knowledge database
The candidate query generation unit extracts each feature amount of a set of a pre-decomposition query and a post-decomposition query to be processed, and each of the set of the pre-decomposition query and the post-decomposition query registered in the query decomposition knowledge database An information search apparatus characterized in that a score is calculated from a similarity to a feature amount, is given to the post-decomposition query to be processed, and the post-decomposition query sorted in the order of the score is presented to the user as a candidate query and selected. .

(Appendix 9)
The information search device according to appendix 6, further comprising a query selection knowledge database in which feature values of the decomposed query that has been successfully searched by the search unit are extracted and registered,
The candidate query generation unit extracts a feature amount of a decomposed query to be processed, calculates a score from a similarity to a feature amount of the decomposed query registered in the query selection knowledge database, and performs the processing An information search apparatus characterized in that a post-decomposition query given to a target post-decomposition query and sorted in the order of scores is presented to a user as a candidate query and selected.

(Appendix 10)
In the information search device according to appendix 6 or 7,
The feature amount of the query before decomposition includes one or a plurality of combinations of a constituent element subjected to morphological analysis, the number of words, and a content word having a specific meaning,
The feature value of the post-decomposition query is a noun only extraction, noun and verb only extraction or noun phrase extraction decomposition method, extracted word position on the query before decomposition, extracted word position, number of extracted words An information search apparatus comprising one or more combinations of the number of elements indicating the ratio of the number of elements after decomposition with respect to the number of words in the query before decomposition, or the meaning label indicating the meaning of the extracted word.

(Appendix 11)
A query input step for entering a query specified by the user;
A search step for searching using the query;
If the search fails in the search step, a candidate query generation step of presenting a plurality of candidate queries generated by decomposing the query to the user and causing the search step to search using the query selected by the user;
With
Further, in the candidate query generation step, any one of noun extraction, noun and verb extraction, and noun phrase extraction is set in advance as a decomposition method of the pre-decomposition query, and decomposition is performed based on the set decomposition condition. An information search method, wherein a phrase extracted from a previous query and a combination of phrases are generated as a candidate query. An information search method characterized by that. (5)

(Appendix 12)
In the information search method according to attachment 11, in the candidate query generation step, if no search result is obtained by the search step and no hit is found, it is determined that the search has failed and a candidate query generation process is executed. An information search method characterized by that.

(Appendix 13)
The information search method according to claim 11, further comprising: extracting a feature amount of each of the pair of the pre-decomposition query input in the query input step and the post-decomposition query selected by the user in the candidate query generation step, and executing the query Comprising the step of registering in the decomposition knowledge database;
In the candidate query generation step, each feature amount is extracted for a set of pre-decomposition query and post-decomposition query to be processed, and each set of pre-decomposition query and post-decomposition query registered in the query decomposition knowledge database An information search method characterized in that a score is calculated from a similarity with a feature quantity, given to the post-decomposition query to be processed, and the post-decomposition query sorted in the order of the score is presented to the user as a candidate query and selected. .

(Appendix 14)
The information search method according to appendix 11, further comprising a step of extracting a feature amount of the decomposed query that has been successfully searched in the search step and registering it in a query selection knowledge database.
The candidate query generation step extracts a feature amount of a post-decomposition query to be processed, calculates a score from a similarity with a post-decomposition query registered in the query selection knowledge database, and performs the processing An information search method, characterized in that a post-decomposition query given to a target post-decomposition query and sorted in the score order is presented to a user as a candidate query and selected.

(Appendix 15)
In the information retrieval method according to appendix 11 or 12,
The feature amount of the query before decomposition includes one or a plurality of combinations of a constituent element subjected to morphological analysis, the number of words, and a content word having a specific meaning,
The feature value of the post-decomposition query is a noun only extraction, noun and verb only extraction or noun phrase extraction decomposition method, extracted word position on the query before decomposition, extracted word position, number of extracted words And a ratio of the number of elements after decomposition to the number of words in a query before decomposition, or a combination of one or more meaning labels indicating the meaning of extracted words.

The block diagram of the function structure which showed embodiment of the information search device by this invention Block diagram of the hardware environment of a computer on which the search program of this embodiment is executed Flowchart of information retrieval processing according to the present invention The flowchart which showed the detail of the query decomposition | disassembly knowledge recording process of step S8 of FIG. Explanatory drawing which showed operation | movement of the query decomposition | disassembly knowledge recording process part of FIG. 1 corresponding to FIG. The flowchart which showed the detail of the query selection knowledge recording process of step S10 of FIG. Explanatory drawing which showed operation | movement of the query selection knowledge recording part of FIG. 1 corresponding to FIG. The flowchart which showed the detail of the candidate query production | generation process of step S4 of FIG. Explanatory drawing which showed operation | movement of the candidate query process part of FIG. 1 corresponding to FIG. The flowchart which showed other embodiment of the candidate query production | generation process of step S4 of FIG. Explanatory drawing which showed operation | movement of the candidate query process part of FIG. 1 corresponding to FIG. Explanatory drawing which showed the other specific example of operation | movement by the query decomposition knowledge recording process part of FIG. Explanatory drawing which showed the other specific example of operation | movement by the query selection knowledge recording part of FIG. Explanatory drawing which showed the other specific example of operation | movement by the candidate query process part of FIG. Explanatory drawing which showed the other specific example of operation | movement by other embodiment of the candidate query process part of FIG.

Explanation of symbols

10: Query input unit 12: Search control unit 14: Search engine 16: Internet 18: Search result display unit 20: Candidate query processing unit 22: Decomposition candidate display designation unit 24: Knowledge recording unit 26: Query decomposition knowledge recording unit 28: Query selection knowledge recording unit 30: Query decomposition knowledge database 32: Query selection knowledge database 34: Decomposition candidate generation unit 36: Score assignment unit 38: Decomposition candidate presentation unit 40: Predecomposition query feature extraction unit 42, 46: Search success query feature Extraction units 44 and 48: database registration unit 50: CPU
52: Bus 54: RAM
55: ROM
56: Hard disk drive 58: Device interface 60: Keyboard 62: Mouse 64: Display 66: Network adapter 68, 76, 96: Pre-disassembly queries 70, 98, 104: Search success queries 72, 74, 76-1, 80-1 , 96-1, 100, 102, 106-1: feature amount 78, 105: decomposition candidate list 80, 106: post-decomposition query (candidate query)
82-1 to 82-3, 90-1 to 90-3, 108-1 to 108-3, 116-1 to 116-3: stored knowledge 84, 110: pre-decomposition query feature value 86, 112: post-decomposition query Feature amount 88, 94, 114: Similarity calculation 92: Search success query feature amount

Claims (4)

On the computer,
A query input step for entering a query specified by the user;
A search step for searching using the query;
If the search fails in the search step, a candidate query generation step of presenting a plurality of candidate queries generated by decomposing the query to the user and causing the search step to search using the query selected by the user;
Query decomposition knowledge database registration step of extracting each feature quantity and registering it in the query decomposition knowledge database for a set of the pre-decomposition query input in the query input step and the post-decomposition query selected by the user in the candidate query generation step; ,
And execute
Further, in the candidate query generation step, any one of noun extraction, noun and verb extraction, and noun phrase extraction is set in advance as a decomposition method of the pre-decomposition query, and decomposition is performed based on the set decomposition condition. extracting each feature quantity for the set of decomposition after the query and degradation before a query that is a product to Rutotomoni processed combinations of words and phrases extracted from the previous query as a candidate query, registered in the query decomposition knowledge database The score is calculated from the similarity between each feature quantity of the combination of the pre-decomposition query and the post-decomposition query and is given to the post-decomposition query to be processed, and the post-decomposition query sorted in the order of the scores is presented to the user as a candidate query The search program characterized by making it select .

The search program according to claim 1, further comprising: a query selection knowledge database registration step for extracting, in the computer, a feature amount of a post-decomposition query that has been successfully searched in the search step and registering it in a query selection knowledge database ;
And execute
The candidate query generation step extracts a feature amount of a post-decomposition query to be processed, calculates a score from a similarity with a post-decomposition query registered in the query selection knowledge database, and performs the processing A search program characterized in that a post-decomposition query that is given to a target post-decomposition query and sorted in the order of the score is presented to a user as a candidate query and selected.
A query input part for inputting a query specified by the user;
A search unit for searching using the query;
A candidate query generation unit that presents to the user a plurality of candidate queries generated by decomposing the query when the search fails in the search unit, and causes the search unit to search using the query selected by the user;
A query decomposition knowledge database registration unit for extracting each feature quantity and registering it in a query decomposition knowledge database for a set of a pre-decomposition query input in the query input step and a post-decomposition query selected by the user in the candidate query generation step; ,
With
Further, the candidate query generation unit preliminarily sets any one of extraction of nouns, extraction of only nouns and verbs, and extraction of noun phrases as a decomposition method of the query before decomposition, and decomposes based on the set decomposition method. A phrase extracted from the previous query and a combination of phrases are generated as candidate queries, and each feature amount is extracted for the combination of the pre-decomposition query and the post-decomposition query to be processed, and is registered in the query decomposition knowledge database. A score is calculated from the similarity between each feature quantity of the pair of the pre-decomposition query and the post-decomposition query and is given to the post-decomposition query to be processed, and the post-decomposition query sorted in the order of the scores is presented to the user as a candidate query. An information search device characterized by being selected by the user.
Computer
A query input step for entering a query specified by the user;
A search step for searching using the query;
If the search fails in the search step, a candidate query generation step of presenting a plurality of candidate queries generated by decomposing the query to the user and causing the search step to search using the query selected by the user;
Query decomposition knowledge database registration step of extracting each feature quantity and registering it in the query decomposition knowledge database for a set of the pre-decomposition query input in the query input step and the post-decomposition query selected by the user in the candidate query generation step; ,
With
Further, in the candidate query generation step, any one of noun extraction, noun and verb extraction, and noun phrase extraction is set in advance as a decomposition method of the pre-decomposition query, and decomposition is performed based on the set decomposition condition. A phrase extracted from the previous query and a combination of phrases are generated as candidate queries, and each feature amount is extracted for the combination of the pre-decomposition query and the post-decomposition query to be processed, and is registered in the query decomposition knowledge database. A score is calculated from the similarity between each feature quantity of the pair of the pre-decomposition query and the post-decomposition query and is given to the post-decomposition query to be processed, and the post-decomposition query sorted in the order of the scores is presented to the user as a candidate query. And an information retrieval method characterized by the fact that it is selected .

Images