JP2011113426A - Dictionary generation device, dictionary generating program, and dictionary generation method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a dictionary generation device for generating dictionary information. <P>SOLUTION: A voice recognition part 11 generates a text resulting from the voice recognition of voice data. A morphological analysis part 12 extracts, in the order of utterance, words appearing in question solution information from a word group extracted by morphological analysis of a text to obtain a response history word group (G). A reference candidate generation part 13 selects each piece of question solution data in a question collection storage part 4 of which the word appearance order has a high matching level with respect to that of response history words G of a response failing in search in search history information, and removes words not included in the selected question solution data from the response history words G to obtain a reference candidate. A successful response detection part 14 removes words not included in browsed question solution data from response history words G of a response succeeding in search in search history information to obtain success instance words G. A reference-candidate collation part 15 specifies a pair maximizing word similarity between reference candidates and success instance words G. A dictionary information generation part 16 outputs the dictionary information of a search keyword group of the response history information of the specified pair. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a dictionary creation device, a dictionary creation program, and a dictionary creation method, and relates to a technique for creating a dictionary used in information retrieval processing.

  In a call center in which an operator provides solution information for an inquiry from a customer, a frequently asked question (FAQ) is prepared in order to shorten the response time for the inquiry. The answering operator enters a search keyword picked up from the customer's utterance into a problem database (FAQ database) that stores information on frequent problems and their solutions, performs a search process, and presents the solutions presented as search results. Refer to the information and provide customers with information such as the resolution procedure and countermeasures for the inquiry.

  Therefore, an appropriate search keyword needs to be used so that solution information for a query can be assigned (hit) in the search process.

  For example, in a call center that performs help desk operations that respond to customer inquiries, the operator obtains words related to the inquiries through dialogue with the customers and inputs them as search keywords. Search the database. Then, the operator refers to the solution information assigned in the search and answers the solution procedure and countermeasures for the inquiry.

  To enable efficient information retrieval even when the conversation with the customer is not good, the speech data is extracted from the voice data of the conversation between the operator and the customer, and the search keyword is extracted. Even if an error occurs in the speech recognition process, there is a process of automatically updating the search keyword combination by changing the search keyword combination or converting a part of the search keyword to the same meaning.

JP 2007-304793 A

  When searching for solution information based on the contents of the dialogue with the customer, a skilled operator uses the contents of the dialogue to identify words and phrases (hereinafter simply referred to as “words”) that are included in the data in the question collection database. That is, it is possible to associate a word suitable for a search as a search keyword. Therefore, a skilled operator can easily assign solution information to the inquiry from the question collection database, and can answer the customer's question in a short time.

  On the other hand, the beginner operator retrieves the contents of the dialog with the customer as they are and inputs them as search keywords to search the question collection database. For this reason, there is a problem that it takes a long time to answer because it is necessary to search again without assigning appropriate solution information. In addition, there was a problem of referring to inappropriate solution information and answering.

  Specifically, the following situation occurs. Assume that the question collection database stores information on the “warning sound” of the notification of occurrence of anomaly and information on how to solve it. Assume that a customer who inquires that the computer is producing unfamiliar sounds utters "The machine is beeping."

  The experienced operator infers from experience that “Peep” means “warning sound”, and searches for the word “warning sound” described in the question collection database as a search keyword. Therefore, the skilled operator can immediately assign solution information (information such as a solution procedure when a warning sound is generated) associated with the search keyword “warning sound”.

  On the other hand, the beginner operator takes “Peepy” from the customer's utterance and searches it as a search keyword as it is. If the word “Peep” does not exist in the question collection database, the solution information regarding “warning sound” cannot be immediately assigned.

  Even when words that are search keywords in the question collection database cannot be picked up from dialogue with the customer, it is possible to assist the operator's search operation by deriving words suitable for replaceable search keywords from the words included in the inquiry. For this purpose, dictionary information defining replaceable search keywords is effective, but it takes a lot of time to manually create dictionary information.

  An object of the present invention relates to a dictionary creation process for automatically creating dictionary information indicating a replaceable word group by using history information of each response including voice data in which an operator's voice indicating a response to an inquiry is recorded. The technology is to provide a dictionary creation device, a dictionary creation program, and a dictionary creation method.

  The outline of a typical dictionary creating apparatus disclosed in the present application will be briefly described as follows. That is, the dictionary creation device disclosed in the present application includes a question collection storage unit that stores question solution data including solution information for an inquiry, and an answer storage that stores voice data uttered by an operator in a dialog with a customer for each response to an inquiry. A search history including a voice storage unit, a search keyword used in a search process for each response, a search result indicating success or failure of the search process, and browsing data indicating question solution data browsed when the search is successful A search history storage unit for storing information, a speech recognition unit, a morpheme analysis unit, a reference candidate creation unit, a successful response detection unit, a reference candidate collation unit, and a dictionary information creation unit are provided.

  In the dictionary creation device, the speech recognition unit performs speech recognition processing on each speech data stored in the response speech storage unit, creates speech text data corresponding to the speech data, and stores the speech text data in the first storage unit To do. The morpheme analysis unit performs a morpheme analysis process on each speech text data in the first storage unit, and utters a word that appears in the question solution data in the question collection storage unit from the words obtained by the morpheme analysis process A response history word group is created in order and stored in the second storage unit.

  Then, the reference candidate creating unit takes out the response history word group of the response whose search result is unsuccessful from the second storage unit based on the search history information stored in the search history storage unit, and extracts the received response history The degree of coincidence between the appearance order of the words in the word group and the appearance order of the words described in each question solving data in the question collection storage unit is calculated, and the question solving data having a matching degree equal to or higher than a predetermined value is selected. . The reference candidate creation unit refers to a word that is not described in the question solution data selected based on the degree of coincidence with the response history word group from the response history word group of the response whose search result is failure. Candidates are created and stored in the third storage unit.

  The successful response detection unit detects all the responses whose search results are successful based on the search history information stored in the search history storage unit, and is a question which is browsing data of the response whose search results are successful Identify resolution data. Further, the successful response detection unit extracts a response history word group corresponding to each response whose search result is successful from the second storage unit, and is described in the question solution data specified from the extracted response history word group. Unsuccessful words are removed to create a success case word group, which is stored in the fourth storage unit.

  The reference candidate matching unit compares the reference candidate with each of the success case word groups, calculates the similarity of the words included in each, and calculates the similarity between the reference candidate having the maximum similarity and the success case word group. Identify combinations. The dictionary information creation unit extracts a search keyword from search history information of each of the specified combination of reference candidates and successful case word groups, and outputs the extracted search keyword group as dictionary information.

  According to the dictionary creation device disclosed in the present application, a response similar to a failed response is obtained by using a word contained in the operator's voice data in a response that fails to retrieve appropriate question solution data in response to a customer query. Therefore, it is possible to specify with higher accuracy, and it is possible to efficiently create dictionary information that associates search keywords with a successful response similar to a failed response.

  Therefore, it is possible to realize an efficient creation process of dictionary information for replacing a search keyword input by an operator with little experience in search processing with a more appropriate word.

It is a figure which shows the structural example of the dictionary creation apparatus disclosed. It is a figure which shows the example of the question solution data memorize | stored in the question collection memory | storage part in the disclosed dictionary creation apparatus. It is a figure which shows the example of the audio | voice text data in the dictionary creation apparatus disclosed.

It is a figure explaining the coincidence of the appearance order of each word which audio | voice text data contains. It is a figure which shows the example of removal of the word of a response history word group. It is a figure which shows the example of a customer's and operator's utterance contained in audio | voice data. It is a figure which shows the structural example in one Embodiment of the dictionary creation system containing a dictionary creation apparatus. It is a figure which shows the flow of a process of a dictionary creation system. It is a figure which shows the data structural example of the audio | voice data of a reception voice memory | storage part. It is a figure which shows the data structural example of the response history information of a response history memory | storage part. It is a figure which shows the data structural example of the search history information of a search history memory | storage part. It is a figure which shows the example of a data structure of the question solution data of a question collection memory | storage part. It is a figure which shows the data structural example of reception log | history audio | voice data. It is a figure which shows the data structural example of the response history audio | voice data in which the solution presentation area was provided. It is a figure which shows the example of the dictionary for speech recognition. It is a figure which shows the example of a response history word group. It is a figure which shows the example of a response history word group. It is a figure which shows the more detailed process flow of the process of step S5. It is a figure which shows the example of removal of the same word in a reception log | history word group. It is a figure which shows the example of removal of the same word in a reception log | history word group. It is a figure which shows the arrangement | sequence of the word used for calculation of the coincidence of the appearance order of a word with respect to question solution data of a response history word group. It is a figure which shows the example of removal based on the temporary candidate of the word group of a response history word group. It is a figure which shows the example of a successful response history word group. It is a figure which shows the example of dictionary information. It is a figure which shows the example of a data structure of the search history information to which the data item of the similar response was added. It is a figure for demonstrating calculation of the appearance distance of a word. It is a figure which shows the hardware structural example of the dictionary creation apparatus disclosed.

  First, a dictionary creation apparatus disclosed as one aspect of the present invention will be outlined.

  FIG. 1 is a diagram illustrating a configuration example of a disclosed dictionary creation device 1.

  The dictionary creation device 1 is a device for generating dictionary information indicating replaceable word groups, and includes a response speech storage unit 2, a search history storage unit 3, a question collection storage unit 4, a speech recognition unit 11, and a morpheme analysis unit 12. , A reference candidate creation unit 13, a successful response detection unit 14, a reference candidate collation unit 15, and a dictionary information creation unit 16.

  The answering voice storage unit 2 is a means for storing voice data in which an operator's speech is recorded for each answer to an inquiry. The voice data includes a solution presentation unit that is an utterance portion that presents solution information for an inquiry.

  For each response, the search history storage unit 3 searches the search keyword used in the search process, the search result indicating the success or failure of the search process, and the browsing data indicating the question solution data browsed when the search is successful. Is a means for storing search history information.

  The voice data of the reception voice storage unit 2 and the search history information of the search history storage unit 3 are associated with each other by information for identifying the history information of each reception.

  The question collection storage unit 4 is means for storing question solution information (question solution data) including an inquiry and solution information for the inquiry and solution information such as countermeasures.

  The voice recognition unit 11 performs voice recognition processing on each piece of voice data stored in the response voice storage unit 2, creates voice text data corresponding to the voice data, and stores the voice text data in the first storage unit 171.

  The morpheme analysis unit 12 performs a morpheme analysis process on each speech text data created by the speech recognition unit 11 and selects words appearing in the question solution data in the question collection storage unit 4 from the words obtained by the morpheme analysis process. A response history word group is created by extracting in the utterance order and stored in the second storage unit 172.

  Based on the search history information stored in the search history storage unit 3, the reference candidate creation unit 13 extracts the response history word group of the response whose search result is unsuccessful from the second storage unit 172. Furthermore, the reference candidate creation unit 13 calculates the degree of coincidence between the appearance order of the words in the extracted response history word group and the appearance order of the words described in the question solution data in the question collection storage unit 4. Selects question resolution data with a value equal to or greater than a predetermined value. Further, the reference candidate creation unit 13 removes and refers to words that are not described in the question solution data selected based on the degree of matching with the response history word group from the response history word group of the response whose search result is failure. Create a candidate. The reference candidates are stored in the third storage unit 173.

  In addition, the reference candidate creation unit 13 refers to the question solution data in the question collection storage unit 4 and calculates the question solution data from the words included in the response history word group before calculating the degree of coincidence of the word appearance order. Words not described in either can be removed.

  Further, the reference candidate creation unit 13 selects one of the corresponding words when there are a plurality of identical words appearing within a predetermined time in the response history word group before calculating the degree of coincidence of the word appearance order. Can only remove all other words.

  The successful response detection unit 14 is the browsing data of the response in which all the search results are detected based on the search history information stored in the search history storage unit 3 and the search result is successful. Identify question resolution data. Further, the successful response detection unit 14 extracts a response history word group corresponding to each response whose identified search results are successful from the second storage unit, and describes the extracted response history word group in the identified question solution data. Remove words that have not been created and create a success case word group. The success case word group is stored in the fourth storage unit 174.

  The reference candidate matching unit 15 compares the reference candidates in the third storage unit with each of the success case word groups in the fourth storage unit, calculates the similarity of the words included in each, and the similarity is maximized Identify combinations of reference candidates and success case word groups.

  The dictionary information creation unit 16 extracts search keywords from the search history information of the reference candidates and the success example word groups of the combinations specified by the reference candidate matching unit 15, and outputs the extracted search keyword group as dictionary information To do.

  The first storage unit 171 stores the voice text data generated by the voice recognition unit 11. The second storage unit 172 stores the response history word group generated by the morpheme analysis unit 12. The third storage unit 173 stores the reference candidates generated by the reference candidate creation unit 13. The fourth storage unit 174 stores the successful case word group generated by the successful response detection unit 14.

  The dictionary creation device 1 operates as follows.

  The speech recognition unit 11 of the dictionary creation device 1 performs speech recognition processing on the speech data stored in the response speech storage unit 2 to create speech text data corresponding to the speech data and store it in the first storage unit 171. . The question solution data in the question collection storage unit 4 is used as dictionary information for voice recognition processing.

  FIG. 2 is a diagram illustrating an example of the question solution information stored in the question collection storage unit 4. For example, it is assumed that words such as “sound panel”, “open”, “volume”, and “mute” are described in the solution part of the question solution data D1 of the question collection storage unit 4. Although not shown, it is assumed that “start button”, “click”, “list”, “setting”, and the like are described in the resolution unit D2.

  FIG. 3 is a diagram illustrating an example of speech text data.

  The speech text data shown in FIG. 3 is speech text data corresponding to speech data created by speech recognition processing and recorded from the utterances of a response operator whose search result is unsuccessful. In FIG. 3, the arrow from the top to the bottom indicates the passage of time, and only the portion related to the problem solving data in the unit segment of the speech recognition processing is illustrated. "Open the sound panel", "Please click the start button", "Please select a setting from the list", and "Is the volume muted?"

  The morpheme analysis unit 12 performs morpheme analysis processing on the speech text data created by the speech recognition unit 11, and extracts words that match a predetermined extraction condition (for example, an independent word) from the speech text data. Next, the morpheme analysis unit 12 extracts words appearing in the question solution data in the question collection storage unit 4 from the words extracted in the morpheme analysis process in order of utterances to create a response history word group, and the second storage unit Store to 172.

  Specifically, the morpheme analysis unit 12 extracts words from the four segments of the speech text data shown in FIG. 3 by morpheme analysis processing, and from the extracted word group to the question solution data resolution unit shown in FIG. Words such as “sound panel, open, start button, click, list, setting, volume, mute” that appear are extracted in the order of utterances to make a response history word group.

  Based on the search history information stored in the search history storage unit 3, the reference candidate creation unit 13 extracts the response history word group of the response whose search result is unsuccessful from the second storage unit 172, and extracts the received response. The degree of coincidence between the appearance order of the words in the history word group and the appearance order of the words described in the solution part of the question solution data in the question collection storage part 4 is calculated, and the question solution data having a predetermined degree of coincidence is calculated. Select a resolution.

  FIG. 4 is a diagram for explaining the coincidence of the appearance order of the words included in the speech text data.

  Of the words “sound panel, open, start button, click, list, settings, volume, mute” included in the speech text data (FIG. 3), the “sound panel, open, volume, mute” part is shown in FIG. The order of the words described in the solution part of the question solution data D1 and the “start button, click, list, setting” part are the same as the order of the words described in the solution part of the question solution data D2. Assume that. If the degree of coincidence is equal to or greater than a predetermined value, the reference candidate creation unit 13 selects a solution unit for the question solution data D1 and D2.

  Further, the reference candidate creation unit 13 uses the selected question solution data to remove the remaining word group (sound panel, open, volume, mute) from words in the response history word group that are not present in the solution unit D1. And the remaining word group “start button, click, list, setting” from which a word that does not exist in the resolution unit D2 is removed and stored in the third storage unit 173.

  Further, the reference candidate creation unit 13 refers to the question solution data in the question collection storage unit 4 before calculating the degree of coincidence of the appearance order of words, and converts the words included in the response history word group into question solution data. Missing words may be removed.

  In addition, the reference candidate creating unit 13 may calculate a plurality of the same words that appear in a predetermined number of segments (that is, within the elapsed time) of the voice text data before calculating the degree of coincidence of the appearance order of the words. You may remove all other words, leaving only one of the corresponding words in the response history word group.

  FIG. 5 is a diagram illustrating an example of removing words from the response history word group.

  FIG. 5A shows an example of a response history word group before the removal process. If there is no description “Please open the control panel” in the question resolution data resolution section of the question collection storage section 4, the reference candidate creation section 13 uses the words in the response history word group in FIG. Remove the corresponding word (control panel, open). As a result of this removal, the words in the response history word group become a word group as shown in FIG.

  Further, the reference candidate creation unit 13 selects one word when there is the same word (sound panel, open) that appears between a predetermined number of segments (for example, the number of segments = 5) in the word group of the response history word group. Remove only others, leaving only. As a result of this removal, the words in the response history word group become a word group as shown in FIG.

  Next, the successful response detection unit 14 refers to the search history information stored in the search history storage unit 3, extracts all the responses whose search results were successful, and uses them as browsing data for the search history information of each response. The solution part of the identified question solution data is specified. Then, the successful response detection unit 14 extracts the response history word group corresponding to the successful response from the second storage unit 172, and is described in the question answering data resolution unit specified from the extracted response history word group. The missing word is removed, and the removed word is stored in the fourth storage unit 174 as a success case word group.

  Subsequently, the reference candidate matching unit 15 calculates the similarity of each word for each combination of the reference candidates C1 and C2 in the third storage unit 173 and the successful case word group in the fourth storage unit 174. , The combination of the reference candidate with the highest similarity and the success case word group is specified.

  The dictionary information creation unit 16 extracts the search keywords from the search history information of the reference candidates of the combination identified by the reference candidate matching unit 15 and the successful response cases, and replaces the extracted word group (search keyword group). Output as dictionary information indicating possible words.

  For example, it is assumed that a combination of the reference candidate C1 and the success case word group Gx is specified. Further, it is assumed that the search keyword of the response search history information that is the basis of the reference candidate C1 is “many”, and the search keyword of the search history information of the success case word group Gx is “no sound”. The dictionary information creation unit 16 outputs a set of the search keywords “daman” and “no sound” as dictionary information indicating replaceable word groups.

  Based on the dictionary information generated by the dictionary creation device 1 in this way, a dictionary to be referred to when the operator in charge inputs a search keyword is prepared.

  For example, a dictionary based on dictionary information generated by the dictionary creation device 1 even when a novice operator receives a customer's utterance of “how much a computer is” and inputs “undice” as a search keyword as it is. Is referred to so that the replaceable word “no sound” can be found and used as a search keyword.

  FIG. 6 shows an example of customer and operator utterances included in the voice data.

  In FIG. 6, the arrow from the top to the bottom indicates the passage of time, the left side of the arrow indicating the time represents the customer's utterance, and the right side represents the operator's utterance. Although the voice data in FIG. 6A and the voice data in FIG. 6B are greatly different from each other in the total number of utterances and the utterance time, it can be seen that the contents of the inquiry and the solution information presented are almost the same. In general, even when solution information having the same content is presented in response to a customer inquiry, the utterance time of an operator who interacts with the customer varies. That is, even in the case of the same response, the size of the voice data varies depending on the response, so it is difficult to accurately determine the similarity of the response even if the word groups included in the voice data are directly compared. .

  The dictionary creation device 1 estimates the question solution data to be assigned in the search from the appearance order of the words included in the response voice data that has failed to be searched, and a group of words (reference candidates) that narrows down the words that are required for the search. Create Then, the analogy of the response is determined based on the similarity to the search keyword in the response that has been successfully searched.

  As a result, the dictionary creation device 1 detects a successful response similar to the content of the failed response using a general word similarity determination formula without being affected by the difference in size of the voice data itself. Therefore, it is possible to automatically create and output highly accurate dictionary information.

  Furthermore, in general, the operator repeats the same thing according to the degree of understanding of the customer during the reception, or paraphrases and explains in a different expression. As a result, even if the response is the same, not only will the operator's utterance time vary, but the appearance status of the words being uttered will also differ. The dictionary creation device 1 removes unnecessary words and duplicate words from words appearing in the utterance at the failed response, and then creates a word group necessary for the search. Therefore, the matching accuracy of the degree of coincidence with the question solution data can be increased.

  In addition, the dictionary information creation unit 16 uses the search history information of the reference candidates of the combination identified by the reference candidate matching unit 15 and the successful response examples, and the similarity of the response that is the source of the reference candidate (response for which the search has failed). As the response, similar response information in which the response that is the basis of the successful case word group (the response that has been successfully searched) is output is output.

  As a result, not only the search history information that has been successfully searched but also the search history information of the response that has failed to be searched can be used as usable information.

  Hereinafter, an embodiment of the dictionary creation device 1 will be described in more detail.

  FIG. 7 is a diagram illustrating a configuration example in an embodiment of a dictionary creation system including the dictionary creation device 1.

  The dictionary creation system shown in FIG. 7 includes a dictionary creation device 1, a response voice storage unit 2, a search history storage unit 3, a question collection storage unit 4, a response history management device 5, a discourse structure analysis device 6, a dictionary storage unit 7, A history storage unit 8 is provided.

  The dictionary creation device 1, the response voice storage unit 2, the search history storage unit 3 and the question collection storage unit 4 are the dictionary creation device 1, the response speech storage unit 2, the search history storage unit 3 and the question collection storage unit 4 shown in FIG. Respectively.

  In the dictionary creation system of the embodiment shown in FIG. 7, the reception voice storage unit 2 stores, for each reception, dialogue-recorded voice data obtained by recording voices uttered by the customer who made the inquiry and the operator who responded.

  The search history storage unit 3 includes a search keyword used in the search process in dealing with the customer, information (search result) indicating whether or not the related problem solving data has been hit by the search, and the problem solving data that has been hit. The search history information including the identification information (browsing data) is stored.

  The response history management device 5 manages each information stored in the response voice storage unit 2, the response history storage unit 8, and the search history storage unit 3, and generates response history voice data of each response (response history ID). .

  The discourse structure analyzing apparatus 6 specifies an utterance period (solution presenting section) in which the operator presents solution information from the voice data included in the response history voice data created by the response history management apparatus 5.

  The dictionary storage unit 7 stores dictionary information output from the dictionary creation device 1.

  The response history storage unit 8 stores, for each response, response history information including identification information of the operator who has been processed, information indicating a response period, and information indicating success or failure of a search (search success / failure).

  Hereinafter, the processing operation of the dictionary creation system including the dictionary creation device 1 will be described.

  FIG. 8 is a diagram showing the flow of processing of the dictionary creation system.

  Step S1: The response history management device 5 sets the response voice storage unit 2, the response history storage unit 8, the response history storage unit 8 for each response (response history ID) of the response history information stored in the response history storage unit 8, for each response. Each record in the search history storage unit 3 is taken out and response history voice data is created.

  FIG. 9 is a diagram illustrating a data configuration example of voice data in the reception voice storage unit 2.

  The voice data in the reception voice storage unit 2 includes a response history ID, an operator ID, a recording start time, a recording end time, and data items of a data body.

  The service history ID is service identification information stored as a history. The operator ID is identification information of the operator who responds.

  The recording start time and the recording end time are information indicating the start and end times of the recording of the dialog recording voice data. The data body is two-channel binary data that is a voice data body.

  FIG. 10 is a diagram illustrating a data configuration example of the response history information in the response history storage unit 8.

  The response history information in the response history storage unit 8 includes data items such as response history ID, operator ID, response start time, response end time, and search success / failure.

  The response start time and response end time are information indicating the start and end times of the response. The success or failure of the search is information (flag) indicating whether the search of the question collection storage unit 4 has succeeded or failed.

  FIG. 11 is a diagram illustrating a data configuration example of search history information in the search history storage unit 3.

  The search history information in the search history storage unit 3 includes data items of response history ID, operator ID, search keyword, search result, and browsing data.

  The search keyword is a search keyword used in the search process. The search result is information indicating whether or not the question solution data has been allocated (hit) by the search using the search keyword. The browsing data is identification information of the question solution data browsed after being hit by the search.

  FIG. 12 is a diagram illustrating a data configuration example of the question solution data in the question collection storage unit 4.

  The question solution data in the question collection storage unit 4 includes data items of FAQID, question unit, and solution unit.

  FAQID is identification information of question solution data. The question part is a data item describing a question or a problem that is inquired by a customer. The solution section is a data item that describes solution information for the problem or failure described in the question section.

  FIG. 13 is a diagram illustrating a data configuration example of the response history voice data.

  The response history voice data includes data items of response history ID, operator ID, search keyword, search result, corresponding voice (data body), and corresponding voice section (recording start time: recording end time).

  The response history voice data is created based on the voice data collected using the response history ID as a key, the response history information, and the records extracted from the search history information. In the search keyword and search result, information of “search keyword” and “search result” of the search history information is stored, respectively. In the corresponding voice and the corresponding voice section, information of “data body” and “recording start time: recording end time” of the voice data is stored.

  Step S2: The discourse structure analysis device 6 acquires the response history voice data created by the response history management device 5, and performs the discourse structure analysis processing on the voice data of the corresponding voice (data body) for each response history voice data. Then, the voice data is divided into a question part (utterance section in which the customer is inquiring) and a solution presentation part (utterance section in which the operator is presenting the solution information), and the solution presentation part is specified.

  Specifically, the discourse structure analysis device 6 has a power value indicating the magnitude of the spoken voice for each unit section in each channel of voice data in which the voices of the customer and the operator are recorded in different channels. calculate. Next, the discourse structure analyzing apparatus 6 specifies a channel in which speech is performed for a longer period of time with a constant power in a predetermined section of the voice data and the speech is spoken in advance, and the speaker who is recorded on the channel is identified as “ It is determined as “leading-speaking utterer”, and the period in which the leading-speaking speaker continues speaking from the beginning of the utterance is specified. The utterance period of the other channel (operator side) following the question utterance part is determined as the “solution presentation part” of the operator.

  The details of the processing executed by the discourse structure analyzing device 6 are described in Japanese Patent Application No. 2008-999927 “Sound utterance part extraction processing program, method and apparatus for voice data, and customer inquiry tendency estimation processing program using the question utterance part of voice data,” As described in "Method and apparatus".

  The discourse structure analyzing apparatus 6 adds a data item “solution presentation section” to the response history voice data, and records information (start time: period) indicating the identified solution presentation section.

  FIG. 14 is a diagram showing a data configuration example of the response history voice data provided with the solution presentation section.

  In the response history audio data shown in FIG. 14, the response audio data of response history ID = Incident001 is the audio data of the section from time 15:30 to 16:00 recorded in the audio data record001, and time 15:32 It shows that the section of the utterance that starts for 5 seconds and continues for 5 seconds is the section of the solution presentation unit.

  Step S3: The speech recognition unit 11 receives the response history speech data from the discourse structure analysis device 6, extracts the speech data corresponding to the solution presentation section of the response history speech data, and a speech recognition processing dictionary for each predetermined utterance segment. Is used to create voice text data and store it in the first storage unit 171.

  FIG. 15 is a diagram illustrating an example of a speech recognition dictionary.

  The speech recognition dictionary is created by the morphological analysis unit 12 by extracting main independent words from the problem solving data solution unit of the question collection storage unit 4.

  Step S4: The morpheme analysis unit 12 performs morpheme analysis processing on all the speech text data stored in the first storage unit 171 and divides it into words, and from the divided words, a predetermined condition (for example, an independent word) A word that matches a word (for example, each noun, verb, adjective, adjective verb, adverb, etc.) is extracted. Further, the morpheme analysis unit 12 extracts words appearing in the solution part of the question solution data in the question collection storage unit 4 from the extracted word group in order of utterances to create a response history word group, and the second storage unit Store to 172.

  The response history word group is added to or associated with the corresponding record of the response history voice data.

  16 and 17 are diagrams showing examples of response history word groups.

  16A shows an example of response history word group # 1, FIG. 16B shows an example of response history word group # 2, and FIG. 17 shows an example of response history word group # 3.

  In the response history word groups # 1, # 2, and # 3, words (recognition result words) appearing in each utterance segment of the speech text data, which is the speech recognition processing result for the speech data corresponding to the response history ID, are in segment order, that is, Recorded in order of word appearance.

  Step S5: The reference candidate creation unit 13 creates a reference candidate from words in the reception history word group corresponding to the reception whose search result is failure.

  FIG. 18 is a diagram showing a more detailed processing flow of the processing in step S5.

  Step S51: The reference candidate creation unit 13 removes the same word that appears within a predetermined utterance segment interval (5 or less) for each response history word group.

  Specifically, the reference candidate creation unit 13 determines the relationship in which the same word appears within the specified segment distance (number of segments) for each of all the response history word groups stored in the second storage unit. Extraction is performed, and a removal flag (represented in the present embodiment by enclosing the word with #) is set for subsequent words in the extracted relationship.

  19 and 20 are diagrams illustrating examples of removing the same word.

  19A shows an example of response history word group # 1, FIG. 19B shows an example of response history word group # 2, and FIG. 20 shows an example of response history word group # 3.

  In the response history word group # 1 in FIG. 19A, since there is “install” in the segment numbers = 17 and 19, it indicates that the removal flag (#) is set for the word that appears later. In the response history word group # 2 in FIG. 19B, since the segment numbers = 3, 5 and 19, 22 are “solid”, the removal flag is “solid” for the segment numbers = 5, 22 that appear later. Is set. In the response history word group # 3 in FIG. 20, since “No.” is present in segment Nos. 1 and 3 and “Install” is present in segment Nos. 10 and 13, a removal flag is set for each word that appears later. Yes.

  The reference candidate creation unit 13 removes the word for which the removal flag is set from the response history word group.

  Step S52: For the record (response) in which the search result of the response history voice data is “no hit”, the reference candidate creation unit 13 determines the appearance order of the words in the corresponding response history word group, the question collection storage unit 4 The degree of coincidence with the appearance order of the words in the solution part of each question solving data is calculated, and the question solving data having a coincidence degree higher than a predetermined threshold is selected to create a provisional candidate.

  The reference candidate creation unit 13 takes out one response history word group corresponding to the response whose search result of the response history voice data is “no hit”. The reference candidate creation unit 13 extracts the question solution data one by one from the question collection storage unit 4, and arranges the words arranged in the appearance order of the extracted response history word groups and the order described in the solution unit of the extracted question solution data. Calculate the degree of matching with a word.

  FIG. 21A is a diagram showing a sequence of words used for calculation of matching of appearance order of words with respect to question solution data # 30 in the word group of response history word group # 1. 21 (B) is a diagram showing a word sequence used for calculation of matching of appearance order of words with respect to the question solution data # 31 in the word group of the response history word group # 1.

  The degree of coincidence of the appearance order of words is calculated using, for example, an edit distance (Levenstein distance). The edit distance is a technique for determining how different two character strings are as the distance between characters by the number of edits, that is, the number of insertions, deletions, and substitutions in character units.

  As an example, the distance is calculated for the character string abcd, the character string bcd, and the character string bzd. In the case of the character string abcd and the character string bcd, one-time editing of “deleting the character a” is necessary to make the character string abcd the character string bcd. In the case of the character string abcd and the character string bzd, two edits of “delete character a” and “replace character c with character z” are required to make the character string abcd a character string bzd. Therefore, the pair of the character string abcd and the character string bcd has a shorter distance and a higher matching degree than the pair of the character string abcd and the character string bzd. Details of the edit distance processing will be described later.

  The reference candidate creation unit 13 receives the word continuation instead of the character continuation, obtains the distance between the two word groups of the response history word group and each of the question solving data, calculates the degree of coincidence, and obtains the question solving data. Find the degree of agreement with everything.

  Here, although one question solution data may be selected, it is desirable that there be a plurality. Therefore, the reference candidate creation unit 13 may select the top several of the calculated degrees of coincidence. In the present embodiment, the reference candidate creation unit 13 selects the top two question solution data with the calculated degree of coincidence. The selected question solution data is set as a provisional candidate.

  In the case of the word group shown in FIG. 21 of the response history word group # 1, the calculated distances to the question solution data # 30 and # 31 are 79.0 and 87.0, respectively, and the question solution data # 30 and # 30 If 31 is the top two, the reference candidate creation unit 13 sets question solution data # 30 and # 31 as provisional candidates.

  Step S53: The reference candidate creating unit 13 prepares the word groups of the response history word groups for the number of the selected temporary candidates, and selects the corresponding temporary candidates (question solution data # 30) from the copied word groups of the response history word groups. , # 31), a word that does not appear in the resolution unit is removed, and the word group after the removal is used as a reference candidate and stored in the third storage unit 173.

  FIG. 22 is a diagram showing an example of removal based on provisional candidates for the word group of the response history word group.

  FIG. 22A shows that the removal flag (#) is set for the word “install” that does not appear in the question solution data # 30, which is a provisional candidate, in the words of the response history word group # 1. In FIG. 22 (B), words in the response history word group # 1 are removed to the words “install”, “hot”, “exhaust”, “exhaust”, “mouth”, and “depot” that do not appear in the question solution data # 31 that is a temporary candidate. Indicates that the flag (#) is set.

  The reference candidate creation unit 13 removes the word for which the removal flag is set from the response history word group, and uses the response history word group after the removal as a reference candidate C1 (FIG. 22 (A)) and a reference candidate C2 (FIG. 22 ( B)).

  Step S6: The successful response detection unit 14 refers to the response history voice data, extracts all responses whose search result is “hit”, and specifies the problem solving data recorded in the browsing data of the response. Then, the successful response detection unit 14 extracts the successful response from the response history word group in the second storage unit, and removes words that do not appear in the identified problem solving data from each of the extracted response history word groups Then, a successful case word group is created from the removed word group and stored in the fourth storage unit 174.

  Specifically, for the response history word group # 2 corresponding to the response voice data for which the search result is “hit”, the successful response detection unit 14 determines the number of words in the response history word group # 2 as many as the number of question solution data. The words that do not appear in the solution section of each question solution data are removed from the prepared and copied word groups of the response history word group.

  FIG. 23 is a diagram illustrating an example of a success case word group.

  FIG. 23A shows that the removal flag is set for the word “firm” that does not appear in the solution of the question resolution data # 30 in the words of the response history word group # 2. FIG. 23B shows that the removal flag is set for the words “firm” and “install” that do not appear in the solution of the question resolution data # 31 in the words of the response history word group # 2.

  The successful response detection unit 14 removes the word set with the removal flag from the words of the response history word group # 2, and sets the word group after the removal as a successful case word group G2.

  Step S7: For each reference candidate, the reference candidate collation unit 15 collates the word group of the reference candidate with each word group of the success case word group, and calculates the similarity of the configured words. The reference candidate matching unit 15 identifies the combination (reference candidate and successful case word group) that has the highest calculated similarity.

  Specifically, the reference candidate matching unit 15 regards the word string of the word group of the reference candidate and the word group of the successful case word group as a document vector, and applies the following equation 1 to determine the similarity. Ask.

  The word similarity is calculated between the reference candidates C1 and C2 created from a certain response (response history ID = insident001) and the success case word groups G1 and G2, and as a result of the calculation, the reference candidate C1 and the success case word It is assumed that the similarity of the combination of the group G1 is the maximum.

  The reference candidate matching unit 15 outputs a combination of the reference candidate C1 and the success case word group G1.

  Step S8: The dictionary information creation unit 16 creates and outputs dictionary information and similar response information from the response history information of the reference candidate of the combination having the maximum similarity and the successful case word group.

  The dictionary information creation unit 16 takes out the response history voice data (response history ID = incident001, incident002) of the reference candidate C1 and the success case word group G1, and searches for the response history voice data search keywords “PC, falling”, “power supply, Is extracted (see FIG. 13). The dictionary information creating unit 16 outputs a set of the extracted search keywords “personal computer, falling” and “power, can be turned off” as dictionary information.

  FIG. 24 is a diagram illustrating an example of dictionary information. The dictionary information shown in FIG. 24 is accumulated in the dictionary storage unit 7.

  Furthermore, the dictionary information creation unit 16 was able to hit the question solution data as an answer similar to the original answer of the reference candidate C1, based on the combination information of the reference candidate C1 and the success case word group G1. Similar response information indicating the response that is the basis of the successful case word group G1 is created and output.

  Based on the similar response information, the response history management device 5 adds the data item “similar response history” to the search history information in the search history storage unit 3 and associates it with the corresponding information indicated in the similar response information. Record the response history ID of a successful response.

  FIG. 25 is a diagram illustrating a data configuration example of search history information to which a data item of a similar response history is added. In FIG. 25, the response history ID (incident002) of the response that was able to hit the question solution data is recorded in “similar response history” of the record of response history ID = incident001.

  FIG. 26 is a diagram for explaining the calculation of the appearance distance of words.

  A character string A (abcd) and a character string B (bzd) to be compared are used. As shown in FIG. 26A, as the distance table, a two-dimensional table is prepared in which the column of the table is the number of elements of the character string + 1, and the row of the table is the number of elements of the character string B + 1.

  Then, as shown in FIG. 26B, the first row of the distance table is sequentially filled with the number of elements from the beginning of the character string A (0, 1, 2, 3, 4). Next, the first column of the distance table is sequentially filled with the number of elements from the beginning of the character string B (0, 1, 2, 3).

  Next, as shown in FIG. 26C, the distance is set for the element in the smallest row and the smallest column among the distances not yet calculated in the distance table column. Here, the operation with the lowest cost among the operations of addition, deletion, and replacement is set as the distance to that. Considering the distance between the character a and the character b, the distance is 2 for character insertion, the distance is 2 for character deletion, and the distance is 1 for character replacement. Therefore, the minimum distance is set to 1. As shown in FIG. 26D, distances are set for all the columns of the distance table. Then, as shown in FIG. 26 (E), the value (2) of the maximum element (maximum row and maximum column) of the distance table is set as the distance between the character string A and the character string B.

  FIG. 27 shows a hardware configuration example of the dictionary creation device 1.

  As shown in FIG. 27, the dictionary creation device 1 includes a CPU 101, a main storage unit (memory) 103, an input / output interface 105, an external storage device 110, an input device (such as a keyboard) 120, and an output device (such as a display) 130. It can be implemented by the computer 100.

  The dictionary creating apparatus 1 can be implemented by a program that can be executed by the computer 100. In this case, a program describing the processing contents of the functions that the dictionary creating apparatus 1 should have is provided. When the computer 100 executes the provided program, the processing function of the dictionary creating apparatus 1 described above is realized on the computer 100.

  That is, the speech recognition unit 11, the morpheme analysis unit 12, the reference candidate creation unit 13, the successful response detection unit 14, the reference candidate collation unit 15, the dictionary information creation unit 16 and the like of the dictionary creation device 1 can be configured by a program. The answering voice storage unit 2, the search history storage unit 3, the question collection storage unit 4, and the first to fourth storage units 171 to 174 can be configured by the external storage device 110.

  The computer 100 can also read a program directly from a portable recording medium and execute processing according to the program. In addition, each time the program is transferred from the server computer, the computer 100 can sequentially execute processing according to the received program.

  Further, this program can be recorded on a recording medium readable by the computer 100.

As shown in the above embodiment, when the dictionary creation device 1 is applied to the creation processing of paraphrase dictionary information that supports the search processing of the call center operator, the following effects are obtained. That is,
(1) According to the dictionary creation device 1, when an operator uses the accumulation of question solution data to present a question or a problem solution that a customer has inquired, a word input in the search process and a question It is easy to create a dictionary that shows the correspondence with the word used in the search that hits the solution data. Therefore, words input by an inexperienced operator during search can be replaced with more appropriate words using a dictionary, so that the hit rate of question solution data can be improved.

  (2) According to the dictionary creation device 1, since the replacement dictionary is automatically created using not only the history information of the successful response but also the history information of the failed response, the dictionary information can be stored efficiently. It can be carried out. Of the responses that occur in the call center, the number of cases that failed to be searched is small, and the response history information of such cases can be used effectively.

  (3) According to the dictionary creation device 1, it is possible to output information that associates the response history information of a failed response with the response history information of a similar response that has been successfully searched. Therefore, the usefulness of the response history information is further improved, and the utilization range of the response history information can be expanded.

  The features in the embodiments of the present invention are listed as follows.

(Supplementary note 1) A dictionary creation device for generating dictionary information indicating replaceable word groups,
A question collection storage unit for storing question solution data including solution information for the inquiry;
For each response to an inquiry, a response voice storage unit that stores voice data uttered by an operator in a dialog with a customer;
For each response, search history information including a search keyword used in the search process, a search result indicating success or failure of the search process, and browsing data indicating the question solution data browsed when the search is successful. A search history storage unit for storing;
A first storage unit for storing voice text data;
A second storage unit for storing a response history word group;
A third storage unit for storing reference candidates;
A fourth storage unit for storing success case word groups;
A speech recognition unit that performs speech recognition processing on each of the speech data stored in the reception speech storage unit, creates speech text data corresponding to the speech data, and stores the speech text data in the first storage unit;
A morpheme analysis process is performed on each speech text data in the first storage unit, and words appearing in the question solution data in the question collection storage unit are extracted from the words obtained by the morpheme analysis process in the utterance order. A morpheme analyzer that creates a history word group and stores it in the second storage unit;
Based on the search history information stored in the search history storage unit, a process of extracting a response history word group of a response whose search result is unsuccessful from the second storage unit;
The degree of coincidence between the appearance order of words in the extracted response history word group and the appearance order of words described in each question solution data of the question collection storage unit is calculated, and the degree of coincidence is equal to or greater than a predetermined value. The process of selecting certain question resolution data;
A reference candidate is created by removing words that are not described in the question solution data selected based on the degree of matching with the response history word group from the response history word group of the response whose search result is failure, A reference candidate creation unit that performs processing to be stored in the third storage unit;
A process of detecting all the responses whose search results are successful based on the search history information stored in the search history storage unit, and identifying the question solution data as the browsing data of the responses whose search results are successful When,
A response history word group corresponding to each response whose search result is successful is extracted from the second storage unit, and words not described in the identified question solution data are removed from the extracted response history word group A successful response detection unit that creates a success case word group and stores it in the fourth storage unit;
One of the reference candidates in the third storage unit is compared with each successful case word group in the fourth storage unit, the similarity of the words included in each is calculated, and the reference candidate having the maximum similarity A reference candidate matching unit that identifies combinations with successful case word groups;
A dictionary information creating unit that extracts a search keyword from the response history information of each of the specified combination of reference candidates and successful case word groups, and outputs the extracted search keyword group as dictionary information. Dictionary creation device.

(Additional remark 2) The said reference candidate preparation part refers to the question solution data of the said question collection memory | storage part, before calculating the coincidence degree of the appearance order of the said word from the word contained in the said response history word group, The dictionary creation device according to appendix 1, wherein words that are not described in any of the question solution data are removed.

(Supplementary Note 3) The reference candidate creation unit corresponds to the case where a plurality of identical words appearing within a predetermined time period exist in the response history word group before calculating the degree of coincidence of the appearance order of the words. The dictionary creation apparatus according to either Supplementary Note 1 or Supplementary Note 2, wherein only one word is retained and all other words are removed.

(Additional remark 4) The said dictionary information preparation part is based on the combination which the said reference candidate collation part specified, The response used as the origin of this combined success example word group to the response which becomes the reference candidate of this combination The dictionary creation apparatus according to any one of the supplementary notes 1 to 3, wherein the similar response information for associating is generated and output.

(Supplementary note 5) A dictionary creation program for causing a computer to execute processing for generating dictionary information indicating replaceable word groups,
A question collection storage unit for storing question solution data including solution information for the inquiry;
For each response to an inquiry, a response voice storage unit that stores voice data uttered by the operator;
For each response, search history information including a search keyword used in the search process, a search result indicating success or failure of the search process, and browsing data indicating the question solution data browsed when the search is successful. A search history storage unit for storing;
A first storage unit for storing voice text data;
A second storage unit for storing a response history word group;
A third storage unit for storing reference candidates;
In a computer comprising a fourth storage unit for storing success case word groups,
Processing for performing voice recognition processing on each of the voice data stored in the reception voice storage unit, creating voice text data corresponding to the voice data, and storing the voice text data in the first storage unit;
A morpheme analysis process is performed on each speech text data in the first storage unit, and words appearing in the question solution data in the question collection storage unit are extracted from the words obtained by the morpheme analysis process in the utterance order. A process of creating a history word group and storing it in the second storage unit;
Based on the search history information stored in the search history storage unit, a process of extracting a response history word group of a response whose search result is unsuccessful from the second storage unit;
The degree of coincidence between the appearance order of words in the extracted response history word group and the appearance order of words described in each question solution data of the question collection storage unit is calculated, and the degree of coincidence is equal to or greater than a predetermined value. The process of selecting certain question resolution data;
A reference candidate is created by removing words that are not described in the question solution data selected based on the degree of matching with the response history word group from the response history word group of the response whose search result is failure, Processing to be stored in the third storage unit;
A process of detecting all the responses whose search results are successful based on the search history information stored in the search history storage unit, and identifying the question solution data as the browsing data of the responses whose search results are successful When,
A response history word group corresponding to each response whose search result is successful is extracted from the second storage unit, and words not described in the identified question solution data are removed from the extracted response history word group A process of creating a success case word group and storing it in the fourth storage unit;
One of the reference candidates in the third storage unit is compared with each successful case word group in the fourth storage unit, the similarity of the words included in each is calculated, and the reference candidate having the maximum similarity A process for identifying combinations with success case word groups;
A process of extracting a search keyword from the response history information of each of the identified combination reference candidate and success case word group, and outputting the extracted search keyword group as dictionary information,
Dictionary creation program to be executed.

(Additional remark 6) Before calculating the degree of coincidence of the appearance order of the words to the computer, the question resolution data is referred to from the words included in the response history word group by referring to the question resolution data in the question collection storage unit. The process of removing words that are not described in any of
The dictionary creation program according to appendix 5 for execution.

(Supplementary note 7) Before the computer calculates the degree of coincidence of the appearance order of the words, if there are a plurality of identical words appearing within a predetermined time in the response history word group, Process to keep only one and remove all other words,
The dictionary creation program according to any one of Supplementary Note 5 and Supplementary Note 6 for execution.

(Additional remark 8) The similarity which associates the response which becomes the origin of this combined success example word group with the response which becomes the reference candidate of this combination based on the combination which the said reference candidate collation part specified to the said computer Processing to create and output the response information
The dictionary creation program according to any one of the supplementary notes 5 to 7 for execution.

(Supplementary note 9) A dictionary creation method in which a computer generates dictionary information indicating replaceable word groups,
A question collection storage unit for storing question solution data including solution information for the inquiry;
For each response to an inquiry, a response voice storage unit that stores voice data uttered by the operator;
For each response, search history information including a search keyword used in the search process, a search result indicating success or failure of the search process, and browsing data indicating the question solution data browsed when the search is successful. A search history storage unit for storing;
A first storage unit for storing voice text data;
A second storage unit for storing a response history word group;
A third storage unit for storing reference candidates;
A computer comprising a fourth storage unit for storing success case word groups;
A process of performing voice recognition processing on each of the voice data stored in the reception voice storage unit, creating voice text data corresponding to the voice data, and storing the voice text data in the first storage unit;
A morpheme analysis process is performed on each speech text data in the first storage unit, and words appearing in the question solution data in the question collection storage unit are extracted from the words obtained by the morpheme analysis process in the utterance order. A process of creating a history word group and storing it in the second storage unit;
Based on the search history information stored in the search history storage unit, a process of retrieving a response history word group of a response whose search result is unsuccessful from the second storage unit;
The degree of coincidence between the appearance order of words in the extracted response history word group and the appearance order of words described in each question solution data of the question collection storage unit is calculated, and the degree of coincidence is equal to or greater than a predetermined value. A process of selecting certain question resolution data;
A reference candidate is created by removing words that are not described in the question solution data selected based on the degree of matching with the response history word group from the response history word group of the response whose search result is failure, A process of storing in the third storage unit;
A process of detecting all the responses whose search results are successful based on the search history information stored in the search history storage unit, and identifying the question solution data as the browsing data of the responses whose search results are successful Process,
A response history word group corresponding to each response whose search result is successful is extracted from the second storage unit, and words not described in the identified question solution data are removed from the extracted response history word group A process of creating a success case word group and storing it in the fourth storage unit;
One of the reference candidates in the third storage unit is compared with each successful case word group in the fourth storage unit, the similarity of the words included in each is calculated, and the reference candidate having the maximum similarity A process of identifying combinations with successful case word groups,
A dictionary comprising: a search keyword extracted from the response history information of each of the identified combination reference candidate and successful case word group, and the extracted search keyword group is output as dictionary information. How to make.

(Additional remark 10) In the process of selecting the question solution data whose degree of coincidence is a predetermined value or more, before calculating the degree of coincidence, referring to the question solution data in the question collection storage unit, the response history The dictionary creation method according to appendix 9, wherein words that are not described in any of the question solution data are removed from words included in a word group.

(Additional remark 11) In the process of selecting the question solution data in which the degree of coincidence is a predetermined value or more, before calculating the degree of coincidence, the same word appearing in the response history word group within a predetermined time The dictionary creation method according to Supplementary Note 9 or Supplementary Note 10, wherein when there are a plurality of words, only one of the corresponding words is retained and all other words are removed.

(Supplementary Note 12) In the process of outputting the dictionary information, based on the combination specified by the reference candidate matching unit, the response of the combined success case word group is used as the source of the reference candidate of the combination. 13. The dictionary creation method according to any one of the supplementary note 9 to the supplementary note 11, wherein similar response information that associates the response to be generated is generated and output.

DESCRIPTION OF SYMBOLS 1 Dictionary creation apparatus 11 Speech recognition part 12 Morphological analysis part 13 Reference candidate creation part 14 Successful response detection part 15 Reference candidate collation part 16 Dictionary information creation part 171 1st memory | storage part 172 2nd memory | storage part 173 3rd memory | storage part 174 4th Storage unit 2 Response voice storage unit 3 Search history storage unit 4 Question collection storage unit 5 Response history management device 6 Discourse structure analysis device 7 Dictionary storage unit 8 Response history storage unit

Claims (6)

A dictionary creation device for generating dictionary information indicating replaceable word groups,
A question collection storage unit for storing question solution data including solution information for the inquiry;
For each response to an inquiry, a response voice storage unit that stores voice data uttered by the operator;
For each response, search history information including a search keyword used in the search process, a search result indicating success or failure of the search process, and browsing data indicating the question solution data browsed when the search is successful. A search history storage unit for storing;
A first storage unit for storing voice text data;
A second storage unit for storing a response history word group;
A third storage unit for storing reference candidates;
A fourth storage unit for storing success case word groups;
A speech recognition unit that performs speech recognition processing on each of the speech data stored in the reception speech storage unit, creates speech text data corresponding to the speech data, and stores the speech text data in the first storage unit;
A morpheme analysis process is performed on each speech text data in the first storage unit, and words appearing in the question solution data in the question collection storage unit are extracted from the words obtained by the morpheme analysis process in the utterance order. A morpheme analyzer that creates a history word group and stores it in the second storage unit;
Based on the search history information stored in the search history storage unit, a process of extracting a response history word group of a response whose search result is unsuccessful from the second storage unit;
The degree of coincidence between the appearance order of words in the extracted response history word group and the appearance order of words described in each question solution data of the question collection storage unit is calculated, and the degree of coincidence is equal to or greater than a predetermined value. The process of selecting certain question resolution data;
A reference candidate is created by removing words that are not described in the question solution data selected based on the degree of matching with the response history word group from the response history word group of the response whose search result is failure, A reference candidate creation unit that performs processing to be stored in the third storage unit;
A process of detecting all the responses whose search results are successful based on the search history information stored in the search history storage unit, and identifying the question solution data as the browsing data of the responses whose search results are successful When,
A response history word group corresponding to each response whose search result is successful is extracted from the second storage unit, and words not described in the identified question solution data are removed from the extracted response history word group A successful response detection unit that creates a success case word group and stores it in the fourth storage unit;
One of the reference candidates in the third storage unit is compared with each successful case word group in the fourth storage unit, the similarity of the words included in each is calculated, and the reference candidate having the maximum similarity A reference candidate matching unit that identifies combinations with successful case word groups;
A dictionary information creating unit that extracts a search keyword from the response history information of each of the specified combination of reference candidates and successful case word groups, and outputs the extracted search keyword group as dictionary information. Dictionary creation device. The reference candidate creation unit refers to the question solution data in the question collection storage unit and calculates the question solution data from the words included in the response history word group before calculating the degree of coincidence of the appearance order of the words. The dictionary creation apparatus according to claim 1, wherein words that are not described in any of the words are removed. The reference candidate creation unit may calculate one of the corresponding words when the response history word group includes a plurality of identical words that appear within a predetermined time before calculating the degree of coincidence of the appearance order of the words. 3. The dictionary creation device according to claim 1, wherein only a word is stored and all other words are removed. The dictionary information creation unit, based on the combination specified by the reference candidate collation unit, associates a response that is a source of the reference candidate of the combination with a response that is a source of the combined successful case word group. The dictionary creation device according to any one of claims 1 to 3, wherein information is created and output. A dictionary creation program for causing a computer to execute a process for generating dictionary information indicating replaceable word groups,
A question collection storage unit for storing question solution data including solution information for the inquiry;
For each response to an inquiry, a response voice storage unit that stores voice data uttered by the operator;
For each response, search history information including a search keyword used in the search process, a search result indicating success or failure of the search process, and browsing data indicating the question solution data browsed when the search is successful. A search history storage unit for storing;
A first storage unit for storing voice text data;
A second storage unit for storing a response history word group;
A third storage unit for storing reference candidates;
In a computer comprising a fourth storage unit for storing success case word groups,
Processing for performing voice recognition processing on each of the voice data stored in the reception voice storage unit, creating voice text data corresponding to the voice data, and storing the voice text data in the first storage unit;
A morpheme analysis process is performed on each speech text data in the first storage unit, and words appearing in the question solution data in the question collection storage unit are extracted from the words obtained by the morpheme analysis process in the utterance order. A process of creating a history word group and storing it in the second storage unit;
Based on the search history information stored in the search history storage unit, a process of extracting a response history word group of a response whose search result is unsuccessful from the second storage unit;
The degree of coincidence between the appearance order of words in the extracted response history word group and the appearance order of words described in each question solution data of the question collection storage unit is calculated, and the degree of coincidence is equal to or greater than a predetermined value. The process of selecting certain question resolution data;
A reference candidate is created by removing words that are not described in the question solution data selected based on the degree of matching with the response history word group from the response history word group of the response whose search result is failure, Processing to be stored in the third storage unit;
A process of detecting all the responses whose search results are successful based on the search history information stored in the search history storage unit, and identifying the question solution data as the browsing data of the responses whose search results are successful When,
A response history word group corresponding to each response whose search result is successful is extracted from the second storage unit, and words not described in the identified question solution data are removed from the extracted response history word group A process of creating a success case word group and storing it in the fourth storage unit;
One of the reference candidates in the third storage unit is compared with each successful case word group in the fourth storage unit, the similarity of the words included in each is calculated, and the reference candidate having the maximum similarity A process for identifying combinations with success case word groups;
A process of extracting a search keyword from the response history information of each of the identified combination reference candidate and success case word group, and outputting the extracted search keyword group as dictionary information,
Dictionary creation program to be executed. A dictionary creation method in which a computer generates dictionary information indicating replaceable word groups,
A question collection storage unit for storing question solution data including solution information for the inquiry;
For each response to an inquiry, a response voice storage unit that stores voice data uttered by the operator;
For each response, search history information including a search keyword used in the search process, a search result indicating success or failure of the search process, and browsing data indicating the question solution data browsed when the search is successful. A search history storage unit for storing;
A first storage unit for storing voice text data;
A second storage unit for storing a response history word group;
A third storage unit for storing reference candidates;
A computer comprising a fourth storage unit for storing success case word groups;
A process of performing voice recognition processing on each of the voice data stored in the reception voice storage unit, creating voice text data corresponding to the voice data, and storing the voice text data in the first storage unit;
A morpheme analysis process is performed on each speech text data in the first storage unit, and words appearing in the question solution data in the question collection storage unit are extracted from the words obtained by the morpheme analysis process in the utterance order. A process of creating a history word group and storing it in the second storage unit;
Based on the search history information stored in the search history storage unit, a process of retrieving a response history word group of a response whose search result is unsuccessful from the second storage unit;
The degree of coincidence between the appearance order of words in the extracted response history word group and the appearance order of words described in each question solution data of the question collection storage unit is calculated, and the degree of coincidence is equal to or greater than a predetermined value. A process of selecting certain question resolution data;
A reference candidate is created by removing words that are not described in the question solution data selected based on the degree of matching with the response history word group from the response history word group of the response whose search result is failure, A process of storing in the third storage unit;
A process of detecting all the responses whose search results are successful based on the search history information stored in the search history storage unit, and identifying the question solution data as the browsing data of the responses whose search results are successful Process,
A response history word group corresponding to each response whose search result is successful is extracted from the second storage unit, and words not described in the identified question solution data are removed from the extracted response history word group A process of creating a success case word group and storing it in the fourth storage unit;
One of the reference candidates in the third storage unit is compared with each successful case word group in the fourth storage unit, the similarity of the words included in each is calculated, and the reference candidate having the maximum similarity A process of identifying combinations with successful case word groups,
A dictionary comprising: a search keyword extracted from the response history information of each of the identified combination reference candidate and successful case word group, and the extracted search keyword group is output as dictionary information. How to make.

Images