JP2019008367A - Question word weight calculation apparatus, question answer retrieval apparatus, question word weight calculation method, question answer retrieval method, program and record medium - Google Patents

Abstract

To reduce an influence of unnecessary words in an answer included in a question text of FAQ and perform FAQ retrieval with high accuracy even for unprotected FAQ.SOLUTION: A question word weight calculation apparatus 10 includes: a word weight calculation unit 14 for calculating a weight value indicating an importance in an answer of each word extracted from a question text using a plurality of words extracted from an answer text with respect to each pair of text pair groups by taking as input a text pair group composed of a pair of the question text and the answer text with respect to the question text; and an output unit 15 for outputting a question word weight table 16 for storing a weight value of each of the words calculated for the pair by the word weight calculation unit 14 for each of the pairs.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a question word weight calculation device, a question answer search device, a question word weight calculation method, a question answer search method, a program, and a storage medium, and particularly stores a pair of a question text and an answer text corresponding thereto. The present invention relates to a question word weight calculation device, a question answer search device, a question word weight calculation method, a question answer search method, a program, and a storage medium.

  In recent years, with the diversification of services, the number of services provided by companies and the like tends to increase. Along with this, companies that provide services are continuously increasing the amount of knowledge they need to learn about service specifications and handling methods. For example, it takes a considerable cost for a call center operator installed in a company or the like to acquire sufficient knowledge. On the other hand, responding to an operator who does not have sufficient knowledge may lead to a serious deterioration in service quality for companies and the like, such as making the customer uneasy and increasing the call time with the customer.

  Therefore, in order to provide high-quality services to customers while reducing the cost of acquiring the necessary knowledge, internal knowledge and know-how are prepared as frequently asked questions (hereinafter referred to as FAQ: Frequently Asked Questions). , Efforts to utilize this FAQ are being made. However, since it takes an enormous cost to continue to prepare FAQs for various inquiries, it is an issue how to efficiently maintain FAQs in response to new inquiries.

  For example, according to the conventional technique described in Patent Document 1, it is possible to construct a search system using only FAQ Q (question). Using this conventional technique, it is possible to find the Q that is closest to the customer's question sentence (input query).

  In the case of a prepared FAQ, it is considered effective to apply a conventional technique to construct a FAQ search system. In other words, when it is assumed that “Q in FAQ contains only necessary minimum information (words)”, it is an effective means to find candidates from the similarity between the input query and the question text. It is.

JP 2011-85991 A

  Incidentally, there are cases where a customer makes an inquiry to an operator using electronic mail (hereinafter simply referred to as mail). In this case, it is conceivable that the inquiry mail from the customer is Q (question), the operator's reply mail is A (answer), and the history of the pair of Q and A is used as it is as FAQ.

  However, in the case of a FAQ that is not maintained as inquiries by e-mail, the Q includes a lot of information (words) not directly related to the answer of A. For example, suppose that a customer issues a problem and issues an inquiry mail. Since customers often do not know what is the problem and what information is needed to solve the problem, they tend to include many explanations of situations that are not directly related to the problem. Moreover, the fixed sentence etc. in an email are also included as information not directly related to the solution of the problem.

  For this reason, in the prior art, when the FAQ is prepared, the similarity cannot be calculated correctly under the influence of a word that becomes noise, and it is difficult to realize sufficient accuracy.

  In addition, a clue word for answering A in Q is often lacking. In this case as well, it is difficult to calculate the degree of similarity correctly, which causes a decrease in accuracy.

  The present invention has been made in view of the above circumstances, reduces the influence of unnecessary words on answers contained in FAQ question text, and performs FAQ searches with high accuracy even for unprepared FAQs. An object of the present invention is to provide a question word weight calculation device, a question answer search device, a question word weight calculation method, a question answer search method, a program, and a storage medium.

  In order to achieve the above object, a question word weight calculation device according to a first aspect of the present invention is directed to a text pair group consisting of a pair of a question text and an answer text for the question text, and each pair of the text pair group. For each of the pairs, a word weight calculation unit that calculates a weight value indicating importance in an answer of each word extracted from the question text using a plurality of words extracted from the answer text And an output unit for outputting a question word weight table storing the weight value of each word calculated for the pair by the weight calculation unit.

  Further, the question word weight calculation device according to the second invention is the first invention, wherein the text pair group and the external document are input, and each pair of the text pair group is input from the external document and from the answer text. In-document text extraction unit for extracting related description locations including extracted words as in-document text, and for each pair, the pair questions from the in-document text extracted by the in-document text extraction unit An expanded word extraction unit that extracts words not included in the text as expanded words, and adds the word weight values extracted as expanded words to the question word weight table as the word weights for the pair. It is a waste.

  On the other hand, in order to achieve the above object, the question answer search device according to the third invention uses a text pair group consisting of a pair of a question text and an answer text for the question text as input, A question word weight table storing weight values indicating importance in answers of each word extracted from the question text, calculated using a plurality of words extracted from the answer text, and an input query Based on each word included and the question word weight table, for each pair, the sum of the weight values of each word that matches the word included in the input query among the weight values of each word of the pair. A FAQ search unit that calculates a search score and outputs the pair corresponding to the input query as a search result according to the search score; Is Dressings.

  On the other hand, in order to achieve the above object, in the question word weight calculation method according to the fourth invention, the word weight calculation unit receives as input a text pair group consisting of a pair of a question text and an answer text to the question text. Calculating, for each pair of the text pair group, a weight value indicating a degree of importance in an answer of each word extracted from the question text using a plurality of words extracted from the answer text; and an output unit Outputting a question word weight table storing the weight value of each word calculated for the pair by the word weight calculator for each pair.

  The question word weight calculation method according to a fifth aspect of the present invention is the method for calculating a question word weight according to the fourth aspect, wherein the in-document text extraction unit inputs the text pair group and the external document and inputs the external pair for each pair of the text pair group. A step of extracting as a text in a document a related description part including a word extracted from the answer text from the document, and an extended word extracting unit for each pair, the text in the document is extracted for the pair A word not included in the question text of the pair is extracted as an extended word from the text in the document, and a weight value of the word extracted as the extended word is added to the question word weight table as a word weight for the pair. And a step.

  On the other hand, in order to achieve the above object, in the question answer search method according to the sixth aspect of the present invention, the FAQ search unit includes each word included in the input query, a question word weight table, a question text, the question Each text extracted from the question text is calculated using a plurality of words extracted from the answer text for each pair of the text pair group, taking as input a text pair group consisting of a pair of answer text to the text Based on a question word weight table storing weight values indicating importance in word responses, for each pair, each word that matches the word included in the input query among the weight values of each word of the pair Is calculated as a search score, and the pair corresponding to the input query is output as a search result according to the search score. It is intended to include a step.

  On the other hand, in order to achieve the above object, a program according to a seventh aspect of the invention provides a computer, each part of the question word weight calculation device according to the first or second aspect, or the question answer according to the third aspect. This is for causing each unit included in the search device to function.

  Furthermore, in order to achieve the above object, a storage medium according to an eighth aspect of the invention provides a computer, each part included in the question word weight calculation device according to the first or second aspect, or the question according to the third aspect. A program for functioning as each unit included in the answer search device is stored.

  As described above, according to the question word weight calculation device, the question answer search device, the question word weight calculation method, the question answer search method, the program, and the storage medium according to the present invention, the answers included in the FAQ question text Thus, it is possible to reduce the influence of unnecessary words and to perform FAQ searches with high accuracy even for FAQs that are not maintained.

It is a block diagram which shows an example of a functional structure of the question word weight calculation apparatus which concerns on 1st Embodiment. It is a block diagram which shows an example of a functional structure of the question answer search apparatus which concerns on 1st Embodiment. It is a flowchart which shows an example of the flow of the process by the question word weight calculation program which concerns on 1st Embodiment. It is a flowchart which shows an example of the flow of the process by the question answer search program which concerns on 1st Embodiment. It is a block diagram which shows an example of a functional structure of the question word weight calculation apparatus which concerns on 2nd Embodiment. It is a flowchart which shows an example of the flow of a process by the question word weight addition program which concerns on 2nd Embodiment.

  Hereinafter, an example of an embodiment for carrying out the present invention will be described in detail with reference to the drawings.

[First Embodiment]
In this embodiment, each word constituting the question text is weighted by “whether or not it is a word necessary for an answer” using the answer text, or is excluded from the search target. The FAQ search with reduced influence is realized.

FIG. 1 is a block diagram illustrating an example of a functional configuration of a question word weight calculation apparatus 10A according to the first embodiment.
As shown in FIG. 1, a question word weight calculation device 10A according to the present embodiment includes a FAQ database (hereinafter, FAQDB) 12, a word weight calculation unit 14, an output unit 15, and a question word weight table 16. Prepare.

  The FAQ DB 12 stores a text pair group consisting of a pair of a question text included in an inquiry mail from a customer and an answer text included in an operator's answer mail for the inquiry mail.

  The word weight calculation unit 14 receives input of each pair constituting the text pair group from the FAQ DB 12. The word weight calculation unit 14 performs processing on each FAQ pair. That is, the word weight calculation unit 14 performs a morphological analysis on each of the question text and the answer text constituting one pair, and extracts a plurality of words from each of the question text and the answer text. The word weight calculation unit 14 uses a plurality of words extracted from the answer text to calculate a weight value indicating the importance in the answer of each word extracted from the question text.

  Specifically, a plurality of words Wa_i (i = 1,..., M) constituting the answer text are used to weight each word Wq_j (j = 1,..., N) constituting the question text. I do.

  For example, the word weight calculation unit 14 is configured such that each word Wq_j (j = 1,..., N) constituting the question text is a plurality of words Wa_i (i = 1,..., M) constituting the answer text. The weight value of each word Wq_j is calculated based on whether it is included. Specifically, when Wq_j is included in Wa_i, the weight value S (Wq_j) of Wq_j is set to “1”. When Wq_j is not included in Wa_i, the weight value S (Wq_j of Wq_j is set. ) Is set to “0”. When S (Wq_j) = 0, the word is excluded from the search target in the subsequent process.

  On the other hand, a word weight value may be calculated as a continuous value using a tf (term frequency) -idf (inverse document frequency) value that is often used in information retrieval. The tf-idf value is a kind of weight value related to a word in the text, tf indicates the frequency (number of times) of the word appearing in the text, and idf indicates the inverse text frequency. When the idf value of a word w is idf (w), idf (w) is calculated by the following formula. Note that df (w) is the number of texts in which the word w appears, and N is the total number of texts.

  idf (w) = log (N / df (w))

  If the tf-idf value for the word w of the text d is TFIDF (d, w), TFIDF (d, w) is calculated by the following equation. Note that tf (d, w) is the frequency (number of times) the word w appears in the text d.

  TFIDF (d, w) = tf (d, w) × idf (w)

  When calculating the weight value of a word as a continuous value using the tf-idf value described above, the word weight calculation unit 14 first uses all the answer texts of FAQ (one text is regarded as one document), Calculate the tf-idf value of the word. In the present embodiment, the calculation result of the tf-idf value regarding the answer text is expressed as follows.

  TFIDF (Wa_i),, i = 1,..., M

  Similarly, the tf-idf value of each word is calculated using all the question texts of the FAQ. In the present embodiment, the calculation result of the tf-idf value regarding the question text is expressed as follows.

  TFIDF (Wq_j),, j = 1,..., N

  Finally, the word weight calculation unit 14 multiplies the tf-idf values of the answer text and the question text by using the following equation (1), thereby weighting a plurality of words Wq_j constituting the question text. A value S (Wq_j) is calculated.

if Wq_j eq Wa_i then
S (Wq_j) = TFIDF (Wa_i) × TFIDF (Wq_j) (1)

  According to the formula (1), the weight value of the word in the question text that is not included in the answer text or is determined to be less important in the answer text becomes a low value. The calculation of the word weight value is not limited to the tf-idf value described above, and other word weight calculation methods such as Okapi BM25 may be used.

  The output unit 15 outputs a question word weight table 16 in which the word weight calculation unit 14 stores the weight value of each word calculated for each pair of FAQs using the above equation (1).

  Here, the question word weight calculation device 10A according to the present embodiment is a computer including a CPU (Central Processing Unit), a RAM (Random Access Memory), a ROM (Read Only Memory), an HDD (Hard Disk Drive), and the like. Composed. The ROM stores a question word weight calculation program according to the present embodiment. The question word weight calculation program may be stored in the HDD.

  The above question word weight calculation program may be installed in advance in the question word weight calculation device 10A, for example. This question word weight calculation program may be realized by being stored in a nonvolatile storage medium or distributed via a network and appropriately installed in the question word weight calculation device 10A. Examples of the nonvolatile storage medium include a CD-ROM (Compact Disc Read Only Memory), a magneto-optical disk, a DVD-ROM (Digital Versatile Disc Read Only Memory), a flash memory, and a memory card.

  The CPU functions as the word weight calculation unit 14 and the output unit 15 by reading and executing the question word weight calculation program stored in the ROM. The HDD functions as each of the above FAQDB 12 and the question word weight table 16.

  Next, the configuration of the question answer search device 20 that performs the FAQ search using the above FAQDB 12 and the question word weight table 16 will be described. In the present embodiment, the case where the question answer search device 20 is configured separately from the question word weight calculation device 10A will be described. However, the question answer search device 20 and the question word weight calculation device 10A are integrally configured. May be.

FIG. 2 is a block diagram illustrating an example of a functional configuration of the question / answer search apparatus 20 according to the first embodiment.
As shown in FIG. 2, the question answer search device 20 according to this embodiment includes a FAQ DB 12, a question word weight table 16, and a FAQ search unit 22. The FAQDB 12 and the question word weight table 16 may be provided outside the question answer search device 20.

  The FAQ search unit 22 performs morphological analysis on the input query (question sentence) 24 and extracts each word from the input query 24. The FAQ search unit 22 matches each word included in the input query 24 and the word included in the input query 24 among the weight values of each word of the pair, based on each word included in the input query 24 and the question word weight table 16. The sum of the weight values of each word to be calculated is calculated as a search score.

  Specifically, for each word Wi_k (k = 1,..., X) constituting the input query 24, a word weight value S (q, W) in a certain question text (here, q) is used. Then, the search score SS (q) of q is calculated by the following equation (2).

if S (q, Wi_k) is defined then
SS (q) = ΣS (q, W) (2)

  According to the above equation (2), the search score SS (q) has a higher value as q is more consistent with the input query 24 and more important words.

  Finally, the FAQ search unit 22 extracts q, for example, in descending order of the search score SS (q), and outputs a pair including the extracted q as the search result 26 corresponding to the input query 24.

  Note that the question answer search device 20 according to the present embodiment is configured as a computer including a CPU, a RAM, a ROM, an HDD, and the like, similar to the question word weight calculation device 10A. The ROM stores a question answer search program according to the present embodiment. The question answer search program may be stored in the HDD.

  The above question answer search program may be installed in advance in the question answer search device 20, for example. This question / answer search program may be realized by being stored in a non-volatile storage medium or distributed via a network and appropriately installed in the question / answer search device 20. Examples of non-volatile storage media include CD-ROM, magneto-optical disk, DVD-ROM, flash memory, memory card, and the like.

  The CPU functions as the FAQ search unit 22 by reading and executing the question answer search program stored in the ROM.

  Next, the operation of the question word weight calculation device 10A according to the first embodiment will be described with reference to FIG. FIG. 3 is a flowchart showing an example of the flow of processing by the question word weight calculation program according to the first embodiment.

  First, in step 100 of FIG. 3, the word weight calculation unit 14 receives an input of a question text and answer text pair in the FAQ from the FAQ DB 12.

  In step 102, the word weight calculation unit 14 performs a morphological analysis on each of the question text and answer text of the pair that has received the input, and extracts a plurality of words from each of the question text and answer text.

  In step 104, the word weight calculation unit 14 calculates a weight value of each word in the question text based on whether or not each word in the question text is included in a plurality of words in the answer text. Alternatively, the word weight calculation unit 14 calculates the weight value S (Wq_j) of each word in the question text using the above equation (1).

  In step 106, the output unit 15 outputs the question word weight table 16 in which the weight value of each word calculated for the pair is stored.

  In step 108, the output unit 15 determines whether or not the weight calculation processing has been completed for all pairs stored in the FAQDB 12. When it is determined that the weight calculation process has been completed for all pairs (in the case of an affirmative determination), the above series of processes ends. On the other hand, when it is determined that the weight calculation processing has not been completed for all pairs (in the case of negative determination), the processing returns to step 100 and is repeated.

  Next, with reference to FIG. 4, the operation of the question / answer search device 20 according to the first embodiment will be described. FIG. 4 is a flowchart showing an example of a process flow by the question answer search program according to the first embodiment.

  First, in step 110 of FIG. 4, the FAQ search unit 22 receives an input query 24.

  In step 112, the FAQ search unit 22 performs morphological analysis on the input query 24 and extracts a plurality of words from the input query 24.

  In step 114, the FAQ search unit 22 calculates a search score SS (q) between the input query 24 and the question text (q) using the above equation (2).

  In step 116, the FAQ search unit 22 determines whether or not the search score calculation processing has been completed for all the question texts stored in the FAQ DB 12. When it is determined that the search score calculation processing has been completed for all the question texts (in the case of an affirmative determination), the above series of processing ends. On the other hand, when it is determined that the search score calculation process has not been completed for all the question texts (in the case of negative determination), the process returns to step 114 and is repeated.

  In step 118, the FAQ search unit 22 extracts q in descending order of the search score, outputs a pair including the extracted q as the search result 26 corresponding to the input query 24, and ends the series of processes.

  According to the present embodiment, it is possible to reduce the influence of words unnecessary for answers included in the question text of the FAQ. For this reason, FAQ search can be performed with high accuracy even for FAQs that are not maintained.

[Second Embodiment]
In this embodiment, from the external document such as a manual related to the service, the description part highly relevant to the answer text is extracted as the text in the document, and the words not included in the question text are expanded from the extracted text in the document Extract as a word. Thereby, the FAQ search considering the word to be included in the question text as an extended word is realized.

FIG. 5 is a block diagram illustrating an example of a functional configuration of the question word weight calculation device 10B according to the second embodiment.
The apparatus configuration described in the second embodiment includes the same parts as the apparatus configuration described in the first embodiment, and the same parts are denoted by the same reference numerals and description thereof is omitted. Only the differences will be described.

  As shown in FIG. 5, the question word weight calculation device 10B according to the present embodiment includes a FAQ DB 12, a word weight calculation unit 14, an output unit 15, a question word weight table 16, an external document DB 30, and a document. An internal text extraction unit 32 and an extended word extraction unit 34 are provided.

  The external document DB 30 stores external documents such as manuals and instruction manuals related to services provided to customers.

  The in-document text extraction unit 32 accepts input of each pair constituting the text pair group from the FAQ DB 12 and accepts input of the external document from the external document DB 30. The in-document text extraction unit 32 performs processing for each FAQ pair. That is, the in-document text extraction unit 32 extracts, as in-document text, a related description location including a word extracted from the answer text from the external document for each pair of the question text and the answer text. As an extraction method, for example, a word is extracted for each answer text, and the number of extracted words included in each sentence block in the external document is counted. The unit of a sentence block may be a sentence, a paragraph, or a chapter. Then, sentence chunks are rearranged in descending order of the number of words contained, and related description locations are extracted as in-document text based on a predetermined extraction criterion. As an extraction criterion, for example, a threshold based on the number of words contained may be used, or the top N may be used.

  Note that when counting the number of words contained, the tf-idf value for each answer text calculated by the word weight calculator 14 described above may be used. In this case, the sum of tf-idf values is used instead of the count value. Thereby, the text in a document which considered the importance of the word can be extracted.

  For each pair of FAQs, the extended word extraction unit 34 extracts, as an extended word, a word that is not included in the question text of the pair from the text in the document extracted for the pair by the in-document text extraction unit 32. The extended word extraction unit 34 adds the word weight value S extracted as the extended word to the question word weight table 16 as the word weight value for the pair. At this time, the weight value of each word as an extended word may be determined uniformly by Z, or the tf-idf value for the external document is calculated for each sentence block defined above, and the tf-idf value is calculated. Based on this, the weight value of each word as an extended word may be determined.

  Note that the question word weight calculation device 10B according to the present embodiment is configured as a computer including a CPU, a RAM, a ROM, an HDD, and the like, similar to the question word weight calculation device 10A described above. The ROM stores a question word weight calculation program and a question word weight addition program according to the present embodiment. The CPU reads and executes the question word weight calculation program and the question word weight addition program stored in the ROM, thereby executing the word weight calculation unit 14, the output unit 15, the in-document text extraction unit 32, and the extended word. It functions as the extraction unit 34.

  Next, the operation of the question word weight calculation device 10B according to the second embodiment will be described with reference to FIGS. FIG. 6 is a flowchart showing an example of the flow of processing by the question word weight addition program according to the second embodiment.

First, the question word weight calculation device 10B executes processing by the question word weight calculation program shown in FIG.
Next, the question word weight calculation device 10B executes processing by the question word weight addition program shown in FIG.

  First, in step 120 of FIG. 6, the in-document text extraction unit 32 receives an input of a question text and answer text pair in the FAQ from the FAQ DB 12.

  In step 122, the in-document text extraction unit 32 performs morphological analysis on the paired answer texts that have received the input, and extracts words from the answer texts.

  In step 124, the in-document text extraction unit 32 extracts a related description portion including a word extracted from the answer text from the external document stored in the external document DB 30 as in-document text.

  In step 126, the extended word extraction unit 34 extracts words that are not included in the question text as extended words from the in-document text extracted by the in-document text extraction unit 32.

  In step 128, the extended word extraction unit 34 determines the weight value of the word extracted as the extended word.

  In step 130, the extended word extraction unit 34 adds the word weight value determined above to the question word weight table 16 as the word weight value for the pair.

  In step 132, the extended word extraction unit 34 determines whether or not the weight addition processing has been completed for all pairs stored in the FAQ DB 12. When it is determined that the weight addition processing has been completed for all pairs (in the case of an affirmative determination), the above series of processing ends. On the other hand, when it is determined that the weight addition processing has not been completed for all pairs (in the case of negative determination), the processing returns to step 120 and is repeated.

  In the present embodiment as well, the question answer search device 20 can be applied in the same way as in the first embodiment when executing the FAQ search for the input query 24.

  According to the present embodiment, the word related to the answer text and not included in the question text is added as an extended word from the external document while reducing the influence of the word unnecessary for the answer included in the question text of the FAQ. Can do. For this reason, it is possible to perform FAQ search with higher accuracy even for FAQs that are not maintained.

  As described above, the question word weight calculation device and the question answer search device have been described as examples. The embodiment may be in the form of a program for causing a computer to function as each unit included in the question word weight calculation device or the question answer search device. The embodiment may be in the form of a computer-readable storage medium storing this program.

  In addition, the configurations of the question word weight calculation device and the question answer search device described in the above embodiment are merely examples, and may be changed according to the situation without departing from the gist.

  Further, the processing flow of the program described in the above embodiment is an example, and unnecessary steps may be deleted, new steps may be added, or the processing order may be changed within a range not departing from the gist. Good.

  Moreover, although the said embodiment demonstrated the case where the process which concerns on embodiment was implement | achieved by a software structure using a computer by running a program, it is not restricted to this. The embodiment may be realized by, for example, a hardware configuration or a combination of a hardware configuration and a software configuration.

10A, 10B Question word weight calculation device 12 FAQDB
14 Word Weight Calculation Unit 15 Output Unit 16 Question Word Weight Table 20 Question Answer Search Device 22 FAQ Search Unit 24 Input Query 26 Search Result 30 External Document DB
32 Text extractor in document 34 Extended word extractor

Claims (8)

Extracted from the question text using a plurality of words extracted from the answer text for each pair of the text pair group, taking as input a text pair group consisting of a question text and an answer text to the question text A word weight calculation unit for calculating a weight value indicating the importance of each word answer;
For each of the pairs, an output unit that outputs a question word weight table storing the weight value of each word calculated for the pair by the word weight calculation unit;
Question word weight calculation device including An in-document text extraction unit that extracts, as the in-document text, a related description location including a word extracted from the answer text from the external document for each pair of the text pair group, using the text pair group and the external document as input. When,
For each of the pairs, from the in-document text extracted for the pair by the in-document text extraction unit, a word not included in the question text of the pair is extracted as an extended word, and the word weight value extracted as the extended word An expanded word extraction unit for adding to the question word weight table as a word weight for the pair;
The question word weight calculation device according to claim 1, further comprising: The question, which is calculated using a plurality of words extracted from the answer text for each pair of the text pair group, taking as input a text pair group consisting of a question text and an answer text to the question text A question word weight table storing weight values indicating the importance of answers in each word extracted from the text;
Based on each word included in the input query and the question word weight table, for each pair, the weight value of each word that matches the word included in the input query among the weight values of each word of the pair A FAQ search unit that calculates a total of the search query and outputs the pair corresponding to the input query as a search result according to the search score;
Question answering search device including The word weight calculation unit receives a text pair group consisting of a pair of a question text and an answer text for the question text, and uses a plurality of words extracted from the answer text for each pair of the text pair group. Calculating a weight value indicating the importance of each word answer extracted from the question text;
An output unit that outputs, for each pair, a question word weight table that stores a weight value of each word calculated for the pair by the word weight calculation unit;
Question word weight calculation method including The in-document text extraction unit receives the text pair group and the external document as input, and for each pair of the text pair group, the related description location including the word extracted from the answer text from the external document as the in-document text Extracting, and
An extended word extraction unit extracts, as an extended word, a word that is not included in the question text of the pair from the in-document text extracted for the pair by the in-document text extraction unit for each pair. Adding the extracted word weight value to the question word weight table as a word weight for the pair;
The question word weight calculation method according to claim 4, further comprising:   The FAQ search unit inputs each text included in the input query, a question word weight table, and a text pair group consisting of a pair of a question text and an answer text to the question text. Based on a question word weight table storing weight values indicating importance in answers of each word extracted from the question text, calculated using a plurality of words extracted from the answer text for the pair, For each pair, among the weight values of each word of the pair, the sum of the weight values of each word that matches the word included in the input query is calculated as a search score, and the input query is determined according to the search score. A question answer search method including a step of outputting the corresponding pair as a search result.   The program for functioning a computer as each part with which the question word weight calculation apparatus of Claim 1 or 2 is provided, or each part with which the question answer search apparatus is provided with Claim 3.   The storage medium which memorize | stored the program for functioning a computer as each part with which the question word weight calculation apparatus of Claim 1 or 2 is provided, or each part with which the question answer search apparatus is provided with Claim 3.