JP5559911B1 - Information retrieval apparatus and program - Google Patents

Abstract

There is a demand for a technique for reducing the rate at which an information search system makes a response that does not match a question.
A query reception unit that receives a query input by a user, a first score acquisition unit that acquires a first score for the query for each of a plurality of response contents searched using a first search algorithm, For each of a plurality of response contents searched using a second search algorithm different from the first search algorithm, a second score acquisition unit that acquires a second score for the query, and a plurality of second contents acquired by the first score acquisition unit As a response content determination unit that determines the response content to the query based on the plurality of second scores acquired by the 1 score and the second score acquisition unit, and a response content output unit that outputs the response content determined by the response content determination unit Provide a program to make it function.
[Selection] Figure 2

Description

  The present invention relates to an information search apparatus and a program.

2. Description of the Related Art Conventionally, there has been known an answering device that outputs answer contents searched from a plurality of answer contents registered in advance for a question from a user. (For example, refer to Patent Document 1).
[Prior art documents]
[Patent Literature]
[Patent Document 1] Japanese Patent Application Laid-Open No. 2011-060218

  A technique for improving the possibility of responding to a query received from a user is desired.

  In the first aspect of the present invention, the computer acquires a first score for a query for each of a plurality of response contents searched using a query receiving unit that receives a query input by a user and a first search algorithm. A first score acquisition unit, a second score acquisition unit that acquires a second score for a query for each of a plurality of response contents searched using a second search algorithm different from the first search algorithm, a first score acquisition unit Based on the plurality of first scores acquired by the second score acquisition unit and the plurality of second scores acquired by the second score acquisition unit, the response content determination unit for determining the response content to the query and the response content determined by the response content determination unit are output. A program for functioning as a response content output unit is provided.

  The program includes a response content table reference unit that refers to a response content table in which a plurality of response contents are registered, a plurality of words, and a phrase score table in which scores associated with each of the plurality of words are registered. The phrase score table reference section for referring to the phrase, the phrase extraction section for extracting a plurality of phrases from the query received by the query reception section, and the plurality of response contents registered in the response contents table are extracted by the phrase extraction section. Of the plurality of phrases, the scores of the plurality of phrases included in the response content are acquired from the phrase score table and multiplied or added to further function as a first score calculation unit that calculates the first score of the response content. The first score acquisition unit may acquire the first score calculated by the first score calculation unit.

  In the response content table, a plurality of response contents and a plurality of words associated with each of the plurality of response contents may be registered, and the first score calculation unit includes a plurality of responses registered in the response content table. For each of the contents, out of a plurality of phrases extracted by the phrase extraction unit, a response content and scores of a plurality of phrases included in the associated plurality of phrases are acquired from the phrase score table and multiplied or added. Thus, the first score of the response content may be calculated. In the above program, the first score calculation unit may further calculate the first score by using an IDF (Inverse Document Frequency) method.

  For each of a plurality of response contents registered in the response content table, the computer calculates scores of a plurality of words included in the response content among a plurality of words extracted by the word / phrase extraction unit, using a TF (Term Frequency) method. And by calculating and adding at least one of the IDF method, the second score calculating unit may further function as a second score calculating unit that calculates the second score of the response content. You may acquire the calculated 2nd score. In the above program, the second score calculation unit applies a response content and a plurality of associated words out of a plurality of words extracted by the word extraction unit for each of a plurality of response contents registered in the response content table. The second score of the response content may be calculated by calculating and adding scores of a plurality of words included by at least one of the TF method and the IDF method. In the above program, the second score calculation unit depends on the weighting for the plurality of words extracted by the word extraction unit and whether the word is included in the response content or the plurality of words associated with the response content. A weight for the length of the response content, and a weight for the number of types of words included in the response content and the corresponding multiple words among the multiple words extracted by the word extraction unit. Further, the second score may be calculated based on the above.

  In the above program, the response content determination unit obtains each first score of the plurality of response contents by a first addition value obtained by adding the first scores of the plurality of response contents searched using the first search algorithm. You may normalize and you may normalize each 2nd score of several response content with the 2nd addition value which added each 2nd score of each response content searched using the 2nd search algorithm. In the above program, the response content determination unit responds to the query based on the plurality of first scores to which the reliability of the first search algorithm is applied and the plurality of second scores to which the reliability of the second search algorithm is applied. The content may be determined. In the above program, the response content determination unit may apply the number of response contents searched using the first search algorithm to the plurality of first scores, and to the plurality of second scores, You may apply the number of the some response content searched using the 2nd search algorithm. In the above program, the response content determination unit may increase the plurality of first scores as the number of the plurality of response contents searched using the first search algorithm increases, and search using the second search algorithm. The larger the number of response contents, the higher the plurality of second scores.

  In the second aspect of the present invention, a query receiving unit that accepts a query input by a user and a first score for the query for each of a plurality of response contents searched using the first search algorithm. A score acquisition unit, a second score acquisition unit that acquires a second score for the query for each of a plurality of response contents searched using a second search algorithm different from the first search algorithm, and a first score acquisition unit Based on the plurality of first scores acquired by the first score and the plurality of second scores acquired by the second score acquisition unit, the response content determination unit that determines the response content to the query and the response content determined by the response content determination unit are output. An information search apparatus including a response content output unit is provided.

  It should be noted that the above summary of the invention does not enumerate all the necessary features of the present invention. In addition, a sub-combination of these feature groups can also be an invention.

An example of the communication environment of an information search system is shown roughly. 1 schematically shows a functional configuration of an information search system. An example of a content table is shown roughly. An example of a phrase score table is shown roughly. An example of a search result score is shown roughly. An example of an evaluation result score is shown roughly.

  Hereinafter, the present invention will be described through embodiments of the invention, but the following embodiments do not limit the invention according to the claims. In addition, not all the combinations of features described in the embodiments are essential for the solving means of the invention.

  FIG. 1 schematically shows an example of a communication environment of the information search system 100. The information search system 100 may be an example of a computer that executes various processes using a program according to the present embodiment. In the present embodiment, the information search system 100 searches the response content for the query received from the user 10 and outputs it to the user 10.

  FIG. 1 illustrates an example in which the information search system 100 receives a query and outputs a response content via the communication terminal 30. The information retrieval system 100 is not limited to accepting queries and outputting response contents. The query may be received directly from the user 10 and the response contents may be output. Further, the communication terminal 30 may function as the information search system 100.

  The user 10 inputs a query to the communication terminal 30 by text or voice, for example. The text input query is transmitted to the information search system 100 by the communication terminal 30. The query input by voice is converted into text by voice recognition technology and transmitted to the information search system 100. The voice recognition process may be executed by at least one of the communication terminal 30, the voice processing server 40 connected to the network 20, and the information search system 100.

  The information search system 100 searches the response content for the accepted query. Then, the information search system 100 outputs the identified response content to the user 10. For example, the information search system 100 causes the communication terminal 30 to display or output the response content. Further, the information search system 100 may directly display or output the sound to the user 10. The response content is converted into speech data by speech synthesis technology. The speech synthesis process may be executed by at least one of the information search system 100, the speech processing server 40, and the communication terminal 30.

  The information search system 100 outputs the response content to the query according to the processing flow described above. The information search system 100 according to the present embodiment is based on the results of searching using a plurality of types of search algorithms in order to reduce the proportion of responses that do not match the query when searching for the response content to the query. Executes processing to identify response contents.

  FIG. 2 schematically shows a functional configuration of the information search system 100. Here, a case will be described as an example where the information search system 100 specifies response contents based on search results obtained by two types of search algorithms, word condition search and full-text search. Note that the search algorithm used by the information search system 100 is not limited to this, and the search may be performed using another search algorithm. Moreover, you may identify the response content based on the search result by three or more types of search algorithms.

  The information search system 100 includes a scenario execution engine 200, a word condition search engine 300, a full-text search engine 400, and an evaluation engine 500. The scenario execution engine 200 includes a query reception unit 202, a morpheme analysis unit 204, a morpheme dictionary 205, an NG word filter 206, an NG word dictionary 207, a synonym expansion unit 208, a synonym dictionary 209, a score acquisition unit 210, a response content determination unit 216, And a response content output unit 218. The word condition search engine 300 includes a word condition search unit 302 and a phrase score table 310. The full text search engine 400 has a full text search unit 402. The evaluation engine 500 includes an evaluation unit 502.

  The query receiving unit 202 receives a query input by the user. The query receiving unit 202 receives a text or voice query from, for example, the communication terminal 30 and the voice processing server 40. When the query reception unit 202 receives a speech query, the query reception unit 202 may convert the query into text by executing speech recognition processing. Note that the query receiving unit 202 may directly receive a query via an input device such as a microphone and a keyboard.

  The morpheme analysis unit 204 performs morpheme analysis on the query using the morpheme dictionary 205 to extract a plurality of words. The phrase may be at least one of a word and a phrase. The morpheme dictionary 205 may include well-known morpheme data, and may include words registered by an administrator of the information search system 100 or the like. The morpheme dictionary 205 may be editable by an administrator of the information search system 100 or the like. The morpheme analysis unit 204 may extract a phrase in a unit that most matches the phrase registered in the morpheme dictionary 205.

  The NG word filter 206 removes NG words registered in the NG word dictionary 207 from the plurality of words extracted by the morphological analysis unit 204. The NG word dictionary 207 may include NG words registered by an administrator of the information search system 100 or the like. As the NG word, for example, a frequently used word that does not contribute to the identification of the response content is registered. The NG word dictionary 207 may be editable by an administrator of the information search system 100 or the like.

  The synonym expansion unit 208 uses the synonym dictionary 209 to expand the synonyms of the phrases extracted by the morphological analysis unit 204 and not removed by the NG word filter 206. The synonym dictionary 209 may include well-known synonym data, and may include synonyms registered by an administrator of the information search system 100 or the like. The synonym dictionary 209 may be editable by an administrator of the information search system 100 or the like.

  The score acquisition unit 210 includes a first score acquisition unit 212 and a second score acquisition unit 214. The first score acquisition unit 212 acquires a score for the query (sometimes referred to as a first score) for each of a plurality of response contents searched by the word condition search engine 300. The first score acquisition unit 212 transmits the plurality of phrases received from the synonym expansion unit 208 to the word condition search unit 302, and the first score of each of the plurality of response contents searched from the word condition search unit 302 is obtained. You may receive it.

  The word condition search unit 302 searches the response contents for the query by referring to the content table 280 and the phrase score table 310. In the content table 280, a plurality of response contents are registered. The plurality of response contents may be registered in advance by an administrator of the information search system 100 or the like. The content table 280 may be an example of a response content table.

  In the phrase score table 310, a plurality of phrases and scores associated with each of the plurality of phrases are registered. A plurality of words and their scores may be registered in advance by an administrator of the information search system 100 or the like.

  The word condition search unit 302 uses, for each of a plurality of response contents registered in the content table 280, out of a plurality of words received from the first score acquisition unit 212, scores of a plurality of words included in the response contents, The first score of the response content may be calculated by obtaining from the score table 310 and multiplying or adding. Then, the word condition search unit 302 may adopt a response content in which the calculated first score is greater than a predetermined threshold as a search result.

  The predetermined threshold may be 0, or may be a value arbitrarily determined by an administrator of the information search system 100 or the like. The search algorithm used by the word condition search unit 302 may be an example of a first search algorithm.

  The second score acquisition unit 214 acquires a score for the query (sometimes referred to as a second score) for each of a plurality of response contents searched by the full-text search engine 400. The second score acquisition unit 214 transmits the plurality of phrases received from the synonym expansion unit 208 to the full-text search unit 402, and receives the second scores of the plurality of retrieved response contents from the full-text search unit 402. It's okay.

  The full-text search unit 402 searches the response content with respect to the query by referring to the content table 280. Here, a case where the word condition search engine 300 and the full text search engine 400 refer to the same content table 280 will be described as an example, but the word condition search engine 300 and the full text search engine 400 are duplicated. The content table 280 may be included. Note that LUCENE, which is a full-text search engine managed by the Apache project, may be adopted as the full-text search engine 400.

  For each of a plurality of response contents registered in the content table 280, the full-text search unit 402 converts a plurality of words / phrases included in the response contents among the plurality of words / phrases acquired from the second score acquisition unit 214 into the TF method and the IDF. The second score of the response content may be calculated by calculating and adding according to at least one of the methods.

  Then, the full-text search unit 402 may adopt a response content in which the calculated second score is greater than a predetermined threshold as a search result. The predetermined threshold may be 0, or may be a value arbitrarily determined by an administrator of the information search system 100 or the like. The search algorithm used by the full text search unit 402 may be an example of a second search algorithm.

  The response content determination unit 216 determines the response content to the query based on the plurality of first scores acquired by the first score acquisition unit 212 and the plurality of second scores acquired by the second score acquisition unit 214. The response content determination unit 216 evaluates the plurality of response contents searched by the word condition search unit 302 and the respective first scores, and the plurality of response contents searched by the full-text search unit 402 and the respective second scores. The part 502 may be evaluated. Here, the evaluation by the evaluation unit 502 may be to calculate final scores of a plurality of response contents based on the first score and the second score.

  The evaluation unit 502 calculates each score of a plurality of response contents for the query based on the plurality of first scores and the plurality of second scores. If the response content of one of the plurality of response content is included in only one of the search result by the word condition search unit 302 and the search result by the full-text search unit 402, the evaluation unit 502 adopts one of the scores To do. Further, the evaluation unit 502 corresponds to the first response content in the case where one response content among the plurality of response content is included in both the search result by the word condition search unit 302 and the search result by the full-text search unit 402. The score and the second score may be weighted and added.

  For example, the evaluation unit 502 applies the reliability of the first search algorithm to the first score and applies the reliability of the second search algorithm to the second score. Application of reliability is, for example, multiplying reliability. As a result, it is possible to provide a search result in which the search result by the search algorithm with lower reliability is also taken into consideration while placing importance on the search result by the search algorithm with higher reliability.

  The application of the reliability may be a calculation in which the score increases as the reliability increases, and may be, for example, multiplication by a coefficient having a magnitude corresponding to the magnitude of the reliability. The reliability of the first search algorithm and the reliability of the second search algorithm may be set in advance by an administrator of the information search system 100 or the like. For example, the reliability of the second search algorithm used by the full text search unit 402 may be set to be higher than the reliability of the first search algorithm used by the word condition search unit 302. The reliability of the first search algorithm and the reliability of the second search algorithm may be set so that the sum is 1.0.

  Further, the evaluation unit 502 may normalize the first score and the second score according to the rank of the search result. For example, the evaluation unit 502 normalizes each first score of the plurality of response contents by using a first addition value obtained by adding the first scores of the plurality of response contents searched by the word condition search unit 302. For example, the evaluation unit 502 normalizes each second score of the plurality of response contents by using a second addition value obtained by adding the second scores of the plurality of response contents searched by the full-text search unit 402. . Thereby, the difference in the score scale between the first search algorithm and the second search algorithm can be absorbed.

  Further, the evaluation unit 502 may apply the number of response contents searched by the word condition search unit 302 to the first score, and set the number of response contents searched by the full-text search unit 402 as the second score. May apply. Applying the number of retrieved response contents means, for example, multiplying the score by the number of retrieved response contents. Thereby, for example, a decrease in score value due to a large number of search results when the score is normalized can be compensated. The application of the number of retrieved response contents may be a calculation in which the score increases as the number of retrieved response contents increases. For example, the magnitude corresponding to the number of retrieved response contents is large. It is also possible to multiply by a coefficient having a certain length.

  The evaluation unit 502 may transmit the response content having the highest score among the plurality of response contents for which the scores have been calculated to the response content determination unit 216 as a search result. In addition, the evaluation unit 502 may transmit an arbitrary number of response contents from the plurality of response contents to the response content determination unit 216 in descending order of score. In addition, the evaluation unit 502 may transmit all response contents for which scores have been calculated to the response content determination unit 216.

  The response content output unit 218 outputs the response content for the query determined by the response content determination unit 216. The response content output unit 218 may output only the response content having the highest score among the plurality of response contents as a search result. Further, the response content output unit 218 may output an arbitrary number of response contents in descending order of score.

  As described above, the information search system 100 according to the present embodiment acquires a plurality of search results from the content table 280 by using a plurality of search algorithms, and obtains response contents based on the plurality of search results. Identify. This makes it possible to provide a search result that makes use of the advantages of a plurality of algorithms and compensates for the disadvantages. In particular, by configuring the reliability of the first search algorithm and the second search algorithm so as to be editable, tuning according to the scene, such as a search object advantageous for the first search algorithm, a search object advantageous for the second search algorithm, etc. It can be done easily.

  FIG. 3 schematically shows an example of the content table 280. The content table 280 may include a title, a body, and a registered phrase for each of a plurality of IDs. The title and body may be an example of response contents. For example, when the response content output unit 218 outputs the response content of ID: 0001, “the service subject to unlimited packet transmission is the usage fee (communication fee) of mail / web used in Japan” "It is the target of unlimited" The registered phrase may be a phrase registered for each ID by an administrator of the information search system 100 or the like. For example, the administrator of the information search system 100 registers a word / phrase related to the response content for each ID.

  FIG. 4 schematically shows an example of the phrase score table 310. In the phrase score table 310, a plurality of phrases and the scores of the plurality of phrases are registered. A plurality of words and their scores may be registered by an administrator of the information search system 100 or the like.

  The phrases registered in the phrase score table 310 may be registered in the morpheme dictionary 205. As a result, for example, when “unlimited packet” is included in the query, the morphological analysis unit 204 can extract “unlimited packet” without disassembling into “packet” and “unlimited”. Note that the morphological analysis unit 204 may extract “unlimited packet”, “packet”, and “unlimited” when the query includes “unlimited packet”. The phrase score table 310 may further include synonyms for each of a plurality of phrases. For example, in the example of FIG. 4, “charged amount” and “amount” are registered as synonyms for the phrase “money”.

  As described above, the word condition search unit 302 refers to the content table 280 and the phrase score table 310 to calculate first scores of a plurality of response contents with respect to the query. Here, the first score calculation of the response content of ID: 0001 when the query “Who can provide an unlimited packet service that can be used for a fixed amount” is received will be described with a specific example. Note that the word condition search unit 302 may or may not use the registered words / phrases in the content table 280 in calculating the first score, but here, a case where registered words / phrases are used will be described.

  First, from the query “objects for which an unlimited packet service can be used for a fixed amount”, the morphological analysis unit 204, the NG word filter 206, and the synonym expansion unit 208, for example, “fixed amount”, “unlimited packet”, “service”. , And “object” are extracted. The word condition search unit 302 identifies the phrase included in the response content and the registered phrase of ID: 0001 among the extracted phrases. Here, “fixed amount”, “all-you-can-packet”, “service”, and “target” are specified. Then, the word condition search unit 302 acquires a score corresponding to each word from the word score table 310 and multiplies or adds the scores. As a result, 180.0 obtained by multiplying 3.0, 10.0, 3.0, and 2.0 or 15.0 obtained by addition is calculated. As described above, the word condition search unit 302 calculates the first score by referring to the content table 280 and the phrase score table 310.

  Note that the word condition search unit 302 may further calculate the first score using the IDF method. That is, the word condition search unit 302 may calculate the first score in consideration of the number of response contents including the target word / phrase among all the response contents registered in the content table 280. For example, when the total number of response contents is 1000 and the number of response contents including “fixed amount” is 9, log (total number of response contents / number of response contents including “fixed amount” +1) = log ( 1000 / (9 + 1)) = 2 may be multiplied by a “flat rate” score of 3.0.

  The full-text search unit 402 may calculate the second score by at least one of the TF method and the IDF method. For example, when only the TF method is used, by calculating and adding the appearance frequencies of “fixed amount”, “unlimited packet”, “service”, and “target” in the response content of ID: 0001, A second score may be calculated.

  The full-text search unit 402 may further calculate the second score by applying various weights. For example, the full-text search unit 402 employs weighting for a plurality of phrases extracted from the query. For example, after extracting a plurality of phrases from the query, the plurality of phrases extracted from the query may be weighted by presenting the plurality of phrases to the user and accepting input of weights for the plurality of phrases. As a result, it is possible to provide a search result that emphasizes words that the user considers more important. When LUCENE is adopted as the full-text search engine 400, the weighting may be realized by a GetBoost function.

  Further, the full text search unit 402 may employ weighting for the search field. For example, the full-text search unit 402 may calculate the score by assigning different weights to the case where the phrase is included in the response content of the content table 280 and the case where it is included in the registered phrase. The full-text search unit 402 may calculate the score by weighting more heavily when it is included in the registered phrase than when it is included in the response content. When LUCENE is adopted as the full-text search engine 400, the weighting may be realized by a GetBoost function.

  The full-text search unit 402 may employ weighting for the length of the response content. For example, the full-text search unit 402 calculates the second score such that the shorter the response content length, the higher the score when a phrase is included. When LUCENE is adopted as the full-text search engine 400, the weighting may be realized by a LengthNorm function.

  In addition, the full-text search unit 402 may employ a weight corresponding to the number of types of phrases included in the response content and the registered phrase among the plurality of phrases extracted from the query by the morpheme analyzer 204. For example, the full-text search unit 402 calculates the second score so that the second score is higher when more types of words / phrases are included. When LUCENE is adopted as the full-text search engine 400, the weighting may be realized by a coord function.

  FIG. 5 schematically shows an example of the search result score and the search result score after normalization. The word condition search result score 350 in FIG. 5 includes four response contents and respective scores. The score of ID: 0001 is 24.0, the score of ID: 0002 is 10.0, the score of ID: 0003 is 4.0, the score of ID: 0004 is 2.0, and the total score is 40.0 .

  The evaluation unit 502 calculates a post-word condition normalization score 352 by normalizing each score with the total score 40.0. For ID: 0001, 24.0 / 40.0 × 100 = 60.0 is calculated. Similarly, ID: 0002, ID: 0003, and ID: 0004 are 25.0, 10.0, 5 respectively. .0 is calculated.

  The full-text search result score 450 includes six response contents and respective scores. ID: 0003 score is 12.0, ID: 0005 score is 10.0, ID: 0006 score is 9.0, ID: 0004 score is 6.0, ID: 0007 score is 2.0, The score of ID: 0008 is 1.0 and the total score is 40.0.

  The evaluation unit 502 calculates the post-full-text search normalized score 452 by normalizing each score with the total score 40.0. For ID: 0003, 12.0 / 40.0 × 100 = 30.0 is calculated. Similarly, ID: 0005, ID: 0006, ID: 0004, ID: 0007, and ID: 0008 are each 25. 0.0, 22.5, 15.0, 5.0, 2.5 are calculated.

  FIG. 6 schematically shows an example of the evaluation result score. FIG. 6 shows an example of the evaluation result score 550 generated by the evaluation unit 502 based on the search result score shown in FIG. The evaluation result score 550 includes a score for each ID of the response content. In the example shown in FIG. 6, a score calculation formula is shown for explanation. Since ID: 0003 is included in both the word condition normalized score 352 and the full text search normalized score 452, the evaluation unit 502 calculates the score in the word condition normalized score 352 and the full text search normalized score. The score of ID: 0003 is calculated from both the score at 452.

  In this example, the evaluation unit 502 calculates the score by multiplying the normalized score by the reliability of the search algorithm and the number of search results. In this example, the reliability of the search by the word condition search engine 300 is 0.3, and the reliability of the search by the full-text search engine 400 is 0.7.

  For ID: 0003, the evaluation unit 502 multiplies the score 10.0 in the word condition normalized score 352 by the reliability of the search algorithm 0.3 and the number of search results 4 to give a score of 12.0. calculate. In addition, the evaluation unit 502 calculates a score of 126.0 by multiplying the score 30.0 in the full-text search normalized score 452 by the reliability 0.7 of the search algorithm and the number of search results 6. Then, the evaluation unit 502 calculates the score of ID: 0003 as 138.0 by adding 12.0 and 126.0.

  Since ID: 0005 is only included in the post-full-text search normalized score 452, the evaluation unit 502 compares the score of 25.0 in the full-text search normalized score 452 with a reliability of the search algorithm of 0.7. And the score of 105.0 is calculated by multiplying the number of search results 6. Similarly for other response contents, 94.5 for ID: 0006, 72.0 for ID: 0001, 69.0 for ID: 0004, 30.0 for ID: 0002, 21 for ID: 0007 1.0, ID: 0008 is calculated as 10.5.

  As described above, the search by the word condition search unit 302 and the full-text search unit 402 are calculated by calculating the score of the response content to the query based on both the search by the word condition search unit 302 and the search by the full-text search unit 402. By taking advantage of the search by, you can provide results that compensate for the shortcomings.

  In the above description, each unit of the information search system 100 may be realized by hardware or may be realized by software. Further, it may be realized by a combination of hardware and software. For example, the scenario execution engine 200, the word condition search engine 300, the full-text search engine 400, and the evaluation engine 500 may be realized by different hardware and software, respectively.

  For example, a computer may function as a part of the information search system 100 by executing a program on the information search system 100. The program may be stored in a computer-readable medium, or may be stored in a storage device connected to a network. In an information processing apparatus having a general configuration including a data processing apparatus having a CPU, ROM, RAM, communication interface, etc., an input apparatus, an output apparatus, and a storage apparatus, the operation of each part of the information search system 100 is defined. The information search system 100 may be realized by starting the software or program.

  A program that is installed in a computer and causes the computer to function as a part of the information search system 100 according to the present embodiment includes a module that defines the operation of each unit of the information search system 100. These programs or modules work on the CPU or the like to cause the computer to function as each part of the information search system 100. Information processing described in these programs functions as a specific means in which software and the various hardware resources described above cooperate with each other by being read by a computer. A specific measurement device according to the purpose of use can be constructed by realizing calculation or processing of information according to the purpose of use of the computer in the present embodiment by these specific means. The information search system 100 may be an example of an information processing device.

  As mentioned above, although this invention was demonstrated using embodiment, the technical scope of this invention is not limited to the range as described in the said embodiment. It will be apparent to those skilled in the art that various modifications or improvements can be added to the above-described embodiment. It is apparent from the scope of the claims that the embodiments added with such changes or improvements can be included in the technical scope of the present invention.

  The order of execution of each process such as operations, procedures, steps, and stages in the apparatus, system, program, and method shown in the claims, the description, and the drawings is particularly “before” or “prior to”. It should be noted that it can be realized in any order unless the output of the previous process is used in the subsequent process. Regarding the operation flow in the claims, the description, and the drawings, even if it is described using “first,” “next,” etc. for convenience, it means that it is essential to carry out in this order. It is not a thing.

10 users, 20 networks, 30 communication terminals, 40 voice processing servers, 100 information retrieval systems, 200 scenario execution engines, 202 query reception units, 204 morpheme analysis units, 205 morpheme dictionaries, 206 NG word filters, 207 NG word dictionaries, 208 Synonym expansion unit, 209 Synonym dictionary, 210 score acquisition unit, 212 first score acquisition unit, 214 second score acquisition unit, 216 response content determination unit, 218 response content output unit, 280 content table, 300 word condition search engine, 302 Word condition search unit, 310 Phrase score table, 350 Word condition search result score, 352 Word condition normalized score, 400 Full-text search engine, 402 Full-text search unit, 450 Full-text search result score, 452 Full-text search normalized score , 500 evaluation engine, 502 evaluation unit

Claims (7)

Computer
A response content table reference unit for referring to a response content table in which a plurality of words associated with a plurality of response content and a plurality of response content are registered;
A phrase score table reference unit that refers to a phrase score table in which a plurality of phrases and scores associated with each of the plurality of phrases are registered;
A query reception unit that receives a query entered by the user,
A phrase extraction unit that extracts a plurality of phrases from the query received by the query reception unit;
Using the first search algorithm, for each of the plurality of response contents registered in the response content table, out of a plurality of words extracted by the word / phrase extraction unit, scores of a plurality of words / phrases included in the response content are calculated. , A first score calculation unit that calculates a first score for the query of the response content by obtaining from the phrase score table and multiplying or adding the result,
A first score acquisition unit that acquires the first score calculated by the first score calculation unit for each of a plurality of response contents searched using the first search algorithm;
Using a second search algorithm different from the first search algorithm, for each of the plurality of response contents registered in the response content table, among the plurality of phrases extracted by the phrase extraction unit, the response contents and By calculating and adding scores of a plurality of phrases included in a plurality of phrases associated with the response content by at least one of a TF (Term Frequency) method and an IDF method, the response content with respect to the query 2nd score calculation part which calculates a 2nd score, Comprising: The number of the kind of phrase contained in the some phrase matched with the response content and the said response content among the some phrase extracted by the said phrase extraction part A second score calculation unit for calculating the second score based on weighting for
A second score acquisition unit that acquires the second score calculated by the second score calculation unit for each of a plurality of response contents searched using the second search algorithm;
A response content determination unit that determines response content to the query based on the plurality of first scores acquired by the first score acquisition unit and the plurality of second scores acquired by the second score acquisition unit; A program for functioning as a response content output unit that outputs the response content determined by the response content determination unit.   The first score calculation unit is associated with the response content and the response content among the plurality of words extracted by the word extraction unit for each of the plurality of response contents registered in the response content table. The program according to claim 1, wherein the first score of the response content is calculated by obtaining scores of a plurality of phrases included in a plurality of phrases from the phrase score table and multiplying or adding the scores. Said first score calculating unit, IDF (Inverse Document Frequency) further calculates the first score with a method, a program according to claim 1 or 2. The response content determination unit
Normalizing each first score of the plurality of response contents by a first addition value obtained by adding the first scores of the plurality of response contents searched using the first search algorithm;
The second score of each of the plurality of response contents is normalized by a second addition value obtained by adding the second scores of the plurality of response contents searched using the second search algorithm. 4. The program according to any one of items 1 to 3 . The response content determination unit
Applying the number of response contents searched using the first search algorithm to the plurality of first scores;
The program according to any one of claims 1 to 4 , wherein the number of the plurality of response contents searched by using the second search algorithm is applied to the plurality of second scores. Computer
A query reception unit that receives a query entered by the user,
A first score acquisition unit that acquires a first score for the query for each of a plurality of response contents searched using the first search algorithm;
A second score acquisition unit that acquires a second score for the query for each of a plurality of response contents searched using a second search algorithm different from the first search algorithm;
A response content determination unit that determines response content to the query based on the plurality of first scores acquired by the first score acquisition unit and the plurality of second scores acquired by the second score acquisition unit; The first score of each of the plurality of response contents is normalized and normalized by a first addition value obtained by adding the first scores of the plurality of response contents searched using the first search algorithm. Each of the plurality of response contents searched using the second search algorithm is multiplied by the number of the plurality of response contents searched using the first search algorithm to the plurality of first scores. The second score of each of the plurality of response contents is normalized by a second addition value obtained by adding the second score, and the second search algorithm is obtained for the normalized second scores. Multiplying the number of retrieved plurality of response content using a rhythm, response content determination unit, and a program for functioning as a response content output unit which outputs the response content to the response content determination unit has determined. A response content table reference unit that refers to a response content table in which a plurality of response contents and a plurality of words associated with each of the plurality of response contents are registered;
A phrase score table reference unit that refers to a phrase score table in which a plurality of phrases and scores associated with each of the plurality of phrases are registered;
A query receiving unit for receiving a query input by a user;
A phrase extraction unit that extracts a plurality of phrases from the query received by the query reception unit;
Using the first search algorithm, for each of the plurality of response contents registered in the response content table, out of a plurality of words extracted by the word / phrase extraction unit, scores of a plurality of words / phrases included in the response content are calculated. A first score calculation unit that calculates a first score for the query of the response content by acquiring from the phrase score table and multiplying or adding;
A first score acquisition unit that acquires the first score calculated by the first score calculation unit for each of a plurality of response contents searched using the first search algorithm;
Using a second search algorithm different from the first search algorithm, for each of the plurality of response contents registered in the response content table, among the plurality of phrases extracted by the phrase extraction unit, the response contents and By calculating and adding scores of a plurality of phrases included in a plurality of phrases associated with the response content by at least one of a TF (Term Frequency) method and an IDF method, the response content with respect to the query A second score calculation unit for calculating a second score, wherein a plurality of words extracted by the word extraction unit are included in the response content or a plurality of words associated with the response content Depends on the weight and the response content among the plurality of words extracted by the word extraction unit and the response content A second score calculating unit that calculates the second score based on weighting for the number of types of phrases included in the plurality of phrases,
A second score acquisition unit that acquires the second score calculated by the second score calculation unit for each of a plurality of response contents searched using the second search algorithm;
A response content determination unit that determines response content to the query based on the plurality of first scores acquired by the first score acquisition unit and the plurality of second scores acquired by the second score acquisition unit; The first score of each of the plurality of response contents is normalized and normalized by a first addition value obtained by adding the first scores of the plurality of response contents searched using the first search algorithm. Each of the plurality of response contents searched using the second search algorithm is multiplied by the number of the plurality of response contents searched using the first search algorithm to the plurality of first scores. The second score of each of the plurality of response contents is normalized by a second addition value obtained by adding the second score, and the second search algorithm is obtained for the normalized second scores. A plurality of first scores obtained by multiplying the number of the plurality of response contents searched using a rhythm and applying a reliability set in advance for the first search algorithm, and the second search algorithm A response content determination unit that determines response content to the query based on a plurality of second scores to which a predetermined reliability is applied;
An information search device comprising: a response content output unit that outputs the response content determined by the response content determination unit.

Images