JP5674689B2 - Knowledge amount estimation information generation device, knowledge amount estimation device, method, and program - Google Patents

Description

  The present invention relates to a knowledge amount estimation information generation device, a knowledge amount estimation device, a method, and a program, and in particular, a knowledge amount related to the content of an inquirer's dialogue from text data of an utterance string in the dialogue between the inquirer and the respondent. The present invention relates to a knowledge amount estimation information generation device, a knowledge amount estimation device, a method, and a program suitable for accurately estimating the amount of information.

  For example, Patent Documents 1 and 2 disclose conventional techniques for estimating the amount of human knowledge using a programmed computer. In these Patent Documents 1 and 2, a technique for estimating a background such as a knowledge amount and a knowledge depth for each field of a user who performs a document search is described.

  In document search, in order to obtain information that the user is searching for, a search is performed by specifying a keyword that is likely to be related to the information. In order to improve the accuracy of the search, it is known that acquisition of information for knowing the background of the user who performs the search is important. For this reason, in Patent Document 3, for example, the keyword history specified by the user or In addition, a technique for estimating a favorite field such as which field is interested from the history of a browsed electronic document is described.

  However, with the preference estimation technique described in Patent Literature 3, even if the field of information that the user frequently examines can be estimated, it is not possible to estimate how familiar the user is in that field.

  On the other hand, in Patent Documents 1 and 2, a query log that is a part of a user log of a search system is used, a degree of expertise of a query is calculated for each field from a query of the query log, and a background relating to a user's knowledge amount It is possible to estimate how familiar the user is in the field.

  Further, Non-Patent Document 1 discloses a technique similar to that of Patent Documents 1 and 2 described above. That is, paying attention to the expertise of the vocabulary used, the expertise of each word is calculated from its rarity, etc., and the average value of the expertise of the word used as the query from the history of the search query entered by a user Is described, and a technique for estimating the knowledge amount of the user is described.

  Non-Patent Document 2 describes a technique for estimating the amount of knowledge of a user in order to perform a collaborative dialogue with the user in a voice dialogue system for the user of the bus operation information guidance system. Here, a technique is described in which the ratio of bus stop attributes (bus stops used only by citizens / others) is determined from the history of bus stops searched by a certain user, and the amount of knowledge about the area of the user is estimated. .

JP 2011-170699 A JP 2011-221872 A JP 2000-148773 A

Daisuke Sato, Yoshihito Yasuda, Takayoshi Mochizuki, Tomoya Suzuki, Yumiko Matsuura, Ryoji Kataoka, “Knowledge Estimation by Search System User Field”, DEIM Forum 2010 I-3 (2010) Junichi Ueno, Kazunori Komatani, Tatsuya Kawahara, Hiroshi Okuno, “Generating Adaptive Responses Using User Models in Bus Operation Information Guidance Systems”, Spoken Language Information Processing 42-2 2002.7.12, vol.65, pp.5- 10 (2002)

  However, with the knowledge amount estimation techniques described in Patent Documents 1 and 2 and Non-Patent Document 1, it is not possible to estimate the speaker's knowledge amount for speakers in a dialogue between humans.

  In order to be able to estimate the amount of knowledge even for a conversation speaker, it is necessary to be able to handle the speech recognition result as an input for knowledge amount estimation. However, when the speech recognition technology is used in the technology for estimating the amount of knowledge of the user using the above-described vocabulary expertise, there is a problem that it is vulnerable to misrecognition included in the speech recognition result.

  As a specific problem, if some highly specialized words appear in the speech recognition result of a speaker's utterance, the knowledge amount of the speaker may be presumed to be erroneously high. is there.

  Examples of situations where knowledge amount estimation technology is used for such conversational speakers include contact center call analysis and voice dialogue systems. In either case, speech recognition is indispensable. Therefore, the knowledge amount estimation technique used in such a scene is required to be robust against misrecognition included in the speech recognition result.

  In the above-mentioned Non-Patent Document 2, when estimating the amount of knowledge of the user in the bus operation information guidance system, the voice information input by the user in a certain area by telephone or the like is recognized by the voice dialogue system, and A technique for estimating the amount of knowledge of the user is described.

  However, also in this technology, the voice information input by the user by telephone or the like is specific information such as a bus stop, a boarding place, and a getting-off place, and further, attributes of these specific information, that is, only citizens use it. Similar to the above-mentioned Patent Documents 1 and 2 and Non-Patent Document 1, such as bus stops and other bus stops, formal bus stop names, bus stop designations with the nearest facility names, etc. The amount of knowledge of the user in the area is estimated, and if some highly specialized words appear in error in the speech recognition result of the user's utterance, the amount of knowledge of the user is erroneously high. It may be estimated that.

  The problem to be solved by the present invention is that, when the technology for estimating the knowledge amount of the user is used for estimating the knowledge amount for the speaker of the dialogue by voice, focusing on the expertise of the conventional user's vocabulary There is a possibility that the knowledge amount of the speaker cannot be estimated correctly due to misrecognition included in the speech recognition result.

  The present invention has been made to solve the above-mentioned problems, and based on the result of dialogue, it is possible to estimate the knowledge amount of the inquirer with respect to the subject matter of the dialogue with high accuracy. An object is to provide an information generation device, a knowledge amount estimation device, a method, and a program.

  In order to achieve the above object, the knowledge amount estimation information generating apparatus according to claim 1 generates an utterance intention corresponding to a question and explanation to the other party from text data of an utterance string in a dialogue between an inquirer and a respondent. An intention feature amount indicating a feature amount related to the occurrence state of the speech intention in the text data, based on the speech intention corresponding to the speech determined by the speech intention determination unit; And extracting the total number of types of words used by the inquirer from the text data as a used vocabulary feature amount, and an intention feature amount and a used vocabulary feature amount extracted by the feature amount extraction unit. Assumed from the dialogue indicated by the text data with respect to the text data targeted for extraction of the intended feature quantity and the used vocabulary feature quantity Knowledge amount estimation information generating means for generating estimated information used for estimating the knowledge amount corresponding to the text data by using knowledge amount information given in advance as the learning data to indicate the knowledge amount of the inquirer And.

  According to the knowledge amount estimation information generating device according to claim 1, each utterance corresponds to a question and explanation to the other party from the text data of the utterance string in the dialogue between the inquirer and the respondent by the utterance intention determination means. An utterance indicating the utterance intention to be performed is determined, and a feature amount relating to the state of occurrence of the utterance intention in the text data is determined by the feature amount extraction unit based on the utterance intention to which the utterance determined by the utterance intention determination unit corresponds. The intention feature amount to be shown is extracted, and the total number of types of words used by the inquirer is extracted from the text data as the used vocabulary feature amount, and is extracted by the feature amount extraction unit by the knowledge amount estimation information generation unit. Intent text and used vocabulary feature quantity, and the text data that is the target of extraction of the intended feature quantity and used vocabulary feature quantity By using, as learning data, knowledge amount information given in advance as indicating the knowledge amount of the inquirer assumed from the dialogue indicated by the data, the estimation information used for estimating the knowledge amount corresponding to the text data is Generated.

  That is, in the present invention, the amount of knowledge of the inquirer, the question exchanged between the inquirer and the respondent in the dialogue, information indicating the intention of the speaker corresponding to the explanation (utterance intention), and the inquiry in the text data Vocabulary features, which is the total number of types of words used by the user, and this results in a problem associated with misrecognition in the technique for estimating the knowledge amount of the inquirer by focusing on the expertise of the conventional vocabulary To be able to solve.

  The learning data includes text data that transcribes the dialogue between the inquirer and the respondent who are the target of extraction of the intended feature amount and the used vocabulary feature amount, and voice of the dialogue between the inquirer and the respondent. Text data created by performing speech recognition processing from data is included. For example, the knowledge amount information is manually given in advance using at least one of text data and voice data of dialogue between the inquirer and the respondent.

  Thus, according to the knowledge amount estimation information generating apparatus according to claim 1, the utterance indicating the utterance intention corresponding to the question and explanation to the other party from the text data of the utterance string in the dialogue between the inquirer and the respondent. Based on the utterance intention to which the determined utterance corresponds, an intention feature amount indicating a feature amount related to the occurrence state of the utterance intention in the text data is extracted, and the type of the word used by the inquirer from the text data is extracted. The total number is extracted as the used vocabulary feature amount, and the extracted intention feature amount and the used vocabulary feature amount, and the dialogue assumed by the text data targeted for the extraction of the intended feature amount are assumed from the dialog Knowledge corresponding to the text data by using, as learning data, knowledge amount information given in advance to indicate the knowledge amount of the inquirer Since generates estimated information used for the estimation, by using the estimated information may be queried's knowledge of the estimated related matters were subject to dialogue accurately.

  Further, according to the present invention, as in the invention described in claim 2, the utterance intention determination means generates an utterance indicating the utterance intention corresponding to the conflict in addition to the question and explanation to the other party from the text data. It may be determined.

  Further, according to the present invention, as in the invention described in claim 3, the intention feature amount includes the number of questions of the inquirer, the number of questions of the respondent, the number of explanations of the inquirer, and the number of explanations of the respondent. May be included.

  Further, according to the present invention, as in the invention described in claim 4, the intention feature amount may further include at least one of the inquiry number of the inquirer and the inquiry number of the respondent.

  Further, according to the present invention, as in the invention described in claim 5, the feature amount extraction unit extracts an utterance including a questionable word in an utterance corresponding to the question as a questionable question sentence, and the intended feature amount Includes the number of questions of the inquirer, the number of questions of the respondent, the number of explanations of the inquirer, the number of explanations of the respondent, and the number of inquiries of the inquirer, the number of inquiries of the respondent, the inquirer It is also possible to include at least one of the number of interrogative question sentences and the number of interrogative question sentences of the respondent.

  The number of questions of the inquirer, the number of questions of the respondent, the number of times of explanation of the inquirer, the number of times of explanation of the respondent, the number of times of inquiry of the inquirer, the number of times of inquiry of the respondent, the questioning question of the inquirer Intention feature quantities such as the number of times and the number of interrogative question sentences of the respondent can be extracted relatively easily and with high precision, compared to the case where other feature quantities are applied. Thus, it is possible to generate the estimated information of the inquirer's knowledge amount with respect to the matter that is the object of the dialogue, with simpler and higher accuracy.

  On the other hand, in order to achieve the above object, the knowledge amount estimation device according to claim 6 is the estimation information generated by the knowledge amount estimation information generation device according to any one of claims 1 to 5. Predictive object that distinguishes utterances that indicate utterance intentions that correspond to questions and explanations to the other party from the text data of the utterance string in the dialogue between the inquirer and the respondent as the subject of estimation of the knowledge amount Based on the utterance intention to which the utterance determined by the utterance intention determination unit and the estimation target utterance intention determination unit corresponds, an extraction target intention feature amount indicating a feature amount related to the generation state of the utterance intention in the text data is extracted. And an estimation target feature amount extraction means for extracting the total number of types of words used by the inquirer from the text data as a used vocabulary feature amount, and the estimation target feature amount extraction The knowledge amount corresponding to the text data to be estimated is estimated using the estimation target intention feature value and the estimation target use vocabulary feature value extracted by the means and the estimation information stored in the storage means. Knowledge amount estimation means.

  According to the knowledge amount estimation device according to claim 6, the storage unit stores in advance the estimation information generated by the knowledge amount estimation information generation device according to any one of claims 1 to 5, The estimation target utterance intention discriminating means discriminates the utterance indicating the utterance intention corresponding to the question and explanation to the other party from the text data of the utterance string in the dialogue between the inquirer and the respondent whose knowledge is to be estimated. Based on the utterance intention corresponding to the utterance determined by the estimation target utterance intention determination unit by the target feature amount extraction unit, an estimation target intention feature amount indicating a feature amount related to the generation state of the utterance intention in the text data is obtained. A total number of types of words used by the inquirer is extracted from the text data as an estimation target use vocabulary feature amount, and an estimation target knowledge amount estimation means From the estimation target intention feature quantity and the estimation target used vocabulary feature quantity extracted by the estimation target feature quantity extraction unit, and the estimation information stored in the storage unit, the text data to be estimated The corresponding knowledge amount is estimated.

  Thus, according to the knowledge amount estimation device according to claim 6, the estimation information generated by the knowledge amount estimation information generation device according to any one of claims 1 to 5 is stored in advance, The utterance intention indicating the utterance intention corresponding to the question and explanation to the other party is determined from the text data of the utterance string in the dialogue between the inquirer and the respondent as the knowledge amount estimation target, and the utterance intention corresponding to the determined utterance Based on the above, the estimation target intention feature amount indicating the feature amount regarding the occurrence state of the utterance intention in the text data is extracted, and the total number of types of words used by the inquirer from the text data is used as the estimation target use vocabulary feature amount Using the extracted and extracted estimation target intention feature quantity and estimation target use vocabulary feature quantity and the estimation information stored in the storage means, the knowledge corresponding to the text data to be estimated is extracted. Since the estimated amount can be accurately query's knowledge of the estimated regarding matters subject to interaction.

  Further, according to the present invention, as in the invention described in claim 7, the estimation target utterance intention determination means indicates the utterance intention corresponding to the conflict in addition to the question and explanation to the other party from the text data. It is good also as what distinguishes utterance.

  Further, according to the present invention, as in the invention described in claim 8, the estimation target intention feature amount includes the number of questions of the inquirer, the number of questions of the respondent, the number of explanations of the inquirer, The number of times of explanation may be included.

  Further, according to the present invention, as in the invention described in claim 9, the estimation target intention feature amount may further include at least one of the number of inquiries of the inquirer and the number of inquiries of the respondent. .

  Further, according to the present invention, as in the invention described in claim 10, the estimation target feature amount extraction unit extracts an utterance including a questionable word in an utterance corresponding to the question as a questionable question sentence, and the intention The feature amount includes the number of questions of the inquirer, the number of questions of the respondent, the number of times of explanation of the inquirer, the number of explanations of the respondent, and the number of times of inquiry of the inquirer, the number of times of inquiry of the respondent, It may include at least one of the number of interrogator question sentences of the inquirer and the number of questionant question sentences of the respondent.

  In addition, the estimation target utterance intention determination unit and the estimation target feature amount extraction unit in the knowledge amount estimation device according to any one of claims 6 to 10 are described in any one of claims 1 to 5. The utterance intention determination unit and the feature amount extraction unit in the described knowledge amount estimation information generation device can be used in common.

  The number of questions of the inquirer, the number of questions of the respondent, the number of times of explanation of the inquirer, the number of times of explanation of the respondent, the number of times of inquiry of the inquirer, the number of times of inquiry of the respondent, the questioning question of the inquirer Intention feature quantities such as the number of times and the number of interrogative question sentences of the respondent can be extracted relatively easily and with high precision, compared to the case where other feature quantities are applied. Therefore, it is possible to estimate the amount of knowledge of the inquirer for the subject matter that is the object of the dialogue more simply and with high accuracy.

On the other hand, in order to achieve the above object, a knowledge amount estimation information generation method according to claim 11 is a knowledge amount estimation information generation apparatus including an utterance intention determination unit, a feature amount extraction unit, and a knowledge amount estimation information generation unit. A knowledge amount estimation information generating method, wherein the utterance intention discriminating means discriminates an utterance indicating intention of an utterance corresponding to a question to a partner and an explanation from text data of an utterance string in a dialogue between an inquirer and a respondent. An intention feature amount indicating a feature amount related to an occurrence state in the text data of the utterance determined in the utterance intention determination step by the feature amount extraction unit and the feature amount extraction means, and from the text data a feature amount extraction step of extracting a total number of word types that the querying party is used as the use lexical feature quantity, the amount of knowledge estimation information producing By means indicated, the feature quantity extraction intended feature amount extracted at step uses lexical feature amounts, and by the text data to target have been the text data extraction of the intended features and the use lexical feature quantity Knowledge that generates estimated information used to estimate the knowledge amount corresponding to the text data by using, as learning data, knowledge amount information given in advance as an indication of the knowledge amount of the inquirer assumed from the dialogue A quantity estimation information generation step.

  Therefore, according to the knowledge amount estimation information generation method described in claim 11, the method operates in the same manner as in the invention described in claim 1, so that the generated estimation information is used as in the invention described in claim 1. Thus, it is possible to accurately estimate the amount of knowledge of the inquirer with respect to the matter that is the subject of the dialogue.

On the other hand, in order to achieve the above object, a knowledge amount estimation method according to claim 12 is the knowledge amount in the knowledge amount estimation device including the estimation target utterance intention determination means, the estimation target feature amount extraction means, and the knowledge amount estimation means. a method for estimating a storing step of storing in advance in the storage device the estimated information generated by claim 11, wherein the knowledge estimation information generation method, by the estimated target speech intention determining means, the estimated target knowledgeable An estimation target utterance intention determination step for determining an utterance indicating the intention of the utterance corresponding to the question and explanation to the other party, from the text data of the utterance sequence in the dialogue between the inquirer and the respondent, and the estimation target feature quantity extraction means Accordingly, the estimation target speech intention determination the speech is determined in step, the estimation target intent feature quantity indicating a feature amount relating to occurrence of said text data While leaving the estimation target feature quantity extracting the total number of word types that the querying party from the text data is used as the estimated target using lexical feature quantity, by the knowledge amount estimating means, the estimation target feature quantity extraction The knowledge amount corresponding to the text data to be estimated is estimated using the estimation target intention feature amount and the estimation target use vocabulary feature amount extracted in the step and the estimation information stored in the storage device. A knowledge amount estimation step.

  Therefore, according to the knowledge amount estimation method described in claim 12, since it operates in the same manner as the invention described in claim 6, as in the invention described in claim 6, inquiries regarding matters subject to dialogue are obtained. The amount of knowledge of the person can be estimated with high accuracy.

  On the other hand, in order to achieve the above object, according to the program according to claim 13, the computer according to claim 1, or the knowledge amount estimation information generating apparatus according to claim 1 or claim 6. Since it can be operated in the same manner as the knowledge amount estimation device according to any one of items 10, an inquiry about an item that has been a target of dialogue, similarly to the knowledge amount estimation information generation device or the knowledge amount estimation device. The amount of knowledge of the person can be estimated with high accuracy.

  Regardless of the vocabulary used or the content of the conversation, by estimating the amount of knowledge of the inquirer regarding the subject matter of the conversation using the characteristic amount related to the state of the intention of utterance in the conversation between the inquirer and the respondent The amount of knowledge of the inquirer can be estimated with high accuracy. In addition, since the feature amount related to the occurrence state of the question that does not depend on the vocabulary used is used, the feature amount is not easily affected by a speech recognition error. There is no need to prepare learning data for each field of vocabulary or conversation content.

It is a block diagram (partial flowchart) which shows the functional structural example of the knowledge amount estimation apparatus which concerns on embodiment. It is a block diagram which shows the computer structural example of the knowledge amount estimation apparatus which concerns on embodiment. It is a flowchart which shows the flow of a process of the learning process program performed by the knowledge amount estimation apparatus which concerns on embodiment. It is a flowchart which shows the flow of a process of the estimation process program performed by the knowledge amount estimation apparatus which concerns on embodiment. It is a schematic diagram which shows an example of the question pattern information used with the knowledge amount estimation apparatus which concerns on embodiment. It is a schematic diagram which shows an example of the description pattern information used with the knowledge amount estimation apparatus which concerns on embodiment. It is a schematic diagram which shows an example of the conflict pattern information used with the knowledge amount estimation apparatus which concerns on embodiment. It is a schematic diagram which shows an example of the interrogative pattern information used with the knowledge amount estimation apparatus which concerns on embodiment. It is a figure with which it uses for description of operation | movement of the knowledge amount estimation apparatus which concerns on embodiment, and is a schematic diagram which shows an example of the text data which consists of an utterance row | line | column. It is a figure with which it uses for description of operation | movement of the knowledge amount estimation apparatus which concerns on embodiment, and is a schematic diagram which shows an example of criteria information. It is a figure with which it uses for description of operation | movement of the knowledge amount estimation apparatus which concerns on embodiment, and is a schematic diagram which shows an example of evaluation result information. It is a figure with which it uses for description of operation | movement of the knowledge amount estimation apparatus which concerns on embodiment, and is a schematic diagram which shows the other example of evaluation result information.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In this example, an explanation will be given of an example in which the amount of knowledge of a user is estimated by using a voice recognition result of an interaction between an operator and a user in a contact center (hereinafter referred to as a contact center call) as an input.

  The knowledge amount estimation apparatus 10 according to the present embodiment has the functional configuration shown in FIG. 1, and has the computer configuration shown in FIG. First, the configuration of the computer will be described with reference to FIG.

  As shown in FIG. 2, the knowledge amount estimation device 10 according to the present embodiment includes a CPU (Central Processing Unit) 22 that controls the operation of the entire knowledge amount estimation device 10, and when the CPU 22 executes various programs. RAM (Random Access Memory) 24 used as a work area, a ROM (Read Only Memory) 26 in which various control programs and various parameters are stored in advance, and a hard disk 28 (see FIG. The external interface 30 (referred to as “External I / F” in the figure) is responsible for exchanging various types of information among the “HDD” 28), the keyboard 14, mouse 16, display 18, and externally connected devices. These are connected to each other by a system bus BUS. Note that the external interface 30 is connected to a communication device 50 such as a headset for inputting / outputting voice data exchanged through dialogue, for example.

  The CPU 22 accesses the RAM 24, the ROM 26 and the hard disk 28, acquires various information via the keyboard 14 and the mouse 16, displays various information on the display 18, and exchanges via the call device 50 connected to the external interface 30. Input / output of dialogue information consisting of voice data can be performed.

  Each process in the knowledge amount estimation device 10 according to the present invention shown in FIG. 1 is executed by the CPU 22 reading a program for controlling the processing as the knowledge amount estimation device according to the present invention stored in the hard disk 28 into the RAM 24 and executing it. Part functions are executed.

  As shown in FIG. 1, in the knowledge amount estimation device 10 according to the present embodiment, an utterance sequence extraction unit 1, an utterance intention determination unit 2, a feature amount extraction unit 3, an estimated information generation unit 4, and a knowledge amount estimation unit 5 Are provided as functions realized by programmed computer processing.

  The processing unit in the knowledge amount estimation apparatus 10 according to the present embodiment configured as described above is roughly divided into a “learning unit” and an “estimation unit”.

  The learning unit includes an utterance string extraction unit 1, an utterance intention determination unit 2, a feature amount extraction unit 3, and an estimation information generation unit 4, and is prepared as an inquirer (hereinafter also referred to as a user) and a reply. When dialogue information of a speech recognition result (learning call) 7 for dialogue with a person (hereinafter also referred to as an operator) is input, first, the utterance string extraction unit 1 extracts text data composed of the utterance string.

  Next, in the utterance intention discriminating section 2, the text data of the utterance string in the dialogue between the inquirer and the respondent extracted by the utterance string extracting section 1 is described in detail for each utterance "question" and "explanation". ”,“ Utterance ”, and an utterance indicating the intention of the utterance (hereinafter also referred to as the utterance intention) are determined. The speech intention determined here is output in association with the speech to be determined.

  Further, the feature amount extraction unit 3 calculates “the number of different words in the inquirer's vocabulary”, which will be described in detail later, and each of the “question”, “explanation”, and “consideration” determined by the utterance intention determination unit 2. The number of appearances of the utterance intention is calculated. Furthermore, among “question” utterances, utterances including question words are extracted as “question word question sentences”, and the number of appearances is calculated. The calculated “number of different vocabulary words of the inquirer”, the number of occurrences of the “question”, “explanation”, and “interaction” utterance intentions determined by the utterance intention determination unit 2, and the appearance of “question word question sentence” The number of times is an intention feature amount indicating a feature amount related to the state of occurrence of the utterance intention in the text data. Such a plurality of feature amounts are called feature amount vectors. The elements of the feature quantity vector output by the feature quantity extraction unit 3 are the number of questions, the number of explanations, the number of questions, the number of interrogative questions, and the number of words used in different vocabularies.

  Then, in the estimation information generation unit 4, the intention feature amount 4a extracted by the feature amount extraction unit 3 (described as “feature amount 4a” in the figure) and the text data that is the target of extraction of the intention feature amount 4a Use knowledge amount information 4b (denoted as “knowledge amount label 4b” in the figure) given in advance as an indication of the knowledge amount of the inquirer assumed from the dialogue indicated by the text data as learning data Thus, estimation information used to estimate the knowledge amount corresponding to the text data is generated by the model generation function 4c (described as “model generation 4c for associating the knowledge amount with the feature amount” in the drawing).

  Thus, the learning unit of this example corresponds to the knowledge amount estimation information generation device according to the present invention. Then, the learning unit performs such processing on a relatively large number of previously prepared dialogs, thereby generating estimation information as model information 5a (denoted as “model 5a” in the figure), and a hard disk It memorize | stores in memory | storage devices, such as 28.

  On the other hand, the estimation unit includes a knowledge amount estimation unit 5 together with an utterance sequence extraction unit 1, an utterance intention determination unit 2, and a feature amount extraction unit 3 shared with the learning unit, and in the learning unit (knowledge amount estimation information generation device). Using the model information 5a that is generated and stored in a storage device such as the hard disk 28, the amount of knowledge regarding the matter of the inquirer in the individual dialogue to be estimated is estimated.

  That is, when a speech recognition result (estimation target) 6 of a call i, which is an individual conversation to be estimated, is input, first, the utterance string extraction unit 1 extracts text data including the utterance string. Note that the speech recognition result (estimation target) 6 of the call i may not be used if the recognition accuracy is too low.

  Next, in the utterance intention discriminating unit 2, the utterance string text data in the dialogue between the inquirer to be estimated and the respondent extracted by the utterance string extracting unit 1 is set for each utterance. An utterance (estimation target utterance intention) indicating the intention of the utterance corresponding to “question”, “explanation”, and “conflict” of the respondent) is determined. The speech intention determined here is output in association with the speech to be determined.

  Further, the feature amount extraction unit 3 extracts the total number of types of words used by the inquirer from the text data as the estimation target used vocabulary feature amount, and also determines the estimation target utterance intention determined by the utterance intention determination unit 2. The number of appearances of each “question”, “explanation”, and “consideration” included is calculated. Furthermore, among “question” utterances, utterances including question words are extracted as “question word question sentences”, and the number of appearances is calculated. The calculated “number of different words in the inquirer's vocabulary”, the number of appearances of the estimated utterance intentions of “question”, “explanation”, and “consideration” determined by the utterance intention determination unit 2, “question question sentence” Is the estimation target intention feature value indicating the feature value related to the utterance intention occurrence state in the text data. The elements of the feature quantity vector output by the feature quantity extraction unit 3 are the number of questions, the number of explanations, the number of questions, the number of interrogative questions, and the number of words used in different vocabularies.

  Then, in the knowledge amount estimation unit 5, the estimation target intended feature amount and the estimation extracted by the feature amount extraction unit 3 by the collation function 5 c (in the drawing, described as “estimate the knowledge amount by collating with the model 5 c”). Knowledge corresponding to text data to be estimated using the target vocabulary feature amount and the estimation information generated by the estimation information generation unit 4 of the learning unit and stored as model information 5a in the storage device such as the hard disk 28 Estimate the amount.

  The knowledge amount corresponding to the text data estimated in this way is described as the knowledge amount 8 of the speaker (user) of the call i inputted as the estimation target (in the figure, “output: knowledge amount label 8 of the call i”). ) Is output.

  In this example, the learning unit and the estimation unit use the same utterance sequence extraction unit 1, utterance intention determination unit 2, and feature quantity extraction unit 3, but the learning unit and the estimation unit are identical. It is good also as a structure which uses the utterance sequence extraction part 1, the utterance intention discrimination | determination part 2, and the feature-value extraction part 3 which are different in each.

  Further, in the knowledge amount estimation apparatus 10 according to the present embodiment, voice recognition processing is performed on voice data input / output by the call device 50 shown in FIG. Then, the speech recognition result (estimation target) 6 and the speech recognition result (learning call) 7 of the call i are generated.

  Then, the utterance string extraction unit 1 extracts only the portion corresponding to the utterance from the speech recognition result (learning call) 7 which is a text file. Usually, the speech recognition result often includes time information, information indicating recognition accuracy, and the like, so this processing is necessary.

  If the data input to the utterance string extraction unit 1 is not a text file indicating a speech recognition result, but text data including an utterance string is input, processing in the utterance string extraction unit 1 is not necessary. .

  Further, in this example, the utterance intention determination unit 2 determines whether the utterances are “question”, “explanation”, and “compatibility” for all utterances of the operator and the user in the dialogue.

  In addition, the feature amount extraction unit 3 calculates the number of appearances of each of “question”, “explanation”, and “consideration” extracted by the utterance intention determination unit 2, and further includes a question word in the “question” utterance. The utterance is extracted as “question word question sentence”, the number of appearances is calculated, and in the case of the learning unit, it is extracted as the intention feature amount (in the case of the estimation unit, the estimation target intention feature amount), and the text data The total number of types of words used by the inquirer is extracted as the used vocabulary feature amount and the estimation target used vocabulary feature amount. The elements of the feature quantity vector output by the feature quantity extraction unit 3 are the number of questions, the number of explanations, the number of questions, the number of interrogative questions, and the number of words used in different vocabularies.

  In addition, modeling of the association between the intention feature quantity and the knowledge quantity in the estimation information generation unit 4 and automatic estimation of the knowledge quantity of the speaker on the inquiry (user) side in the knowledge quantity estimation unit 5 are, for example, publicly known For example, “Support Vector Machine (SVM)” which is one of identification techniques of “Supervised learning” is used.

  Hereinafter, using FIG. 3 and FIG. 4, the utterance string extraction unit 1, the utterance intention determination unit 2, the feature amount extraction unit 3, the estimated information generation unit 4, and the knowledge of the knowledge amount estimation device 10 according to the present embodiment. The details of the processing performed by the quantity estimation unit 5 will be described.

  FIG. 3 shows the processing contents in the learning unit of the knowledge amount estimation apparatus 10 according to the present embodiment. First, it is determined whether or not a predetermined amount of speech data has been input (step S101). When input, the voice data is reproduced (step S102), voice recognition processing is performed to transcribe the voice data into text data (step S103), and the speech string extraction unit 1 transcribes the voice recognition result. The text data of the speech string corresponding to the speech is extracted from the text data (step S104).

  For example, if the speech recognition result is an xml file, and the part corresponding to the utterance is enclosed in <TEXT> and </ TEXT>, <TEXT> and </ TEXT If the part surrounded by> is specified, the utterance can be extracted.

  Next, in the utterance intention determination unit 2, each of the text data composed of the utterance sequence extracted by the utterance sequence extraction unit 1 is exchanged between the user and the operator exchanged in the dialogue indicated by the input voice data. A determination process is performed to determine whether the utterance intention of the utterance is “question”, “explanation”, or “consideration” (step S105).

  To determine whether each utterance exchanged in this dialogue is “question”, “explanation”, or “commitment”, for example, “Kenji Imamura, Atsuko Izumi, Genichiro Kikui, Satoshi Sato,“ Predicate Functional Expression Semantic label tagger, using the well-known techniques described in the 17th Annual Conference of the Language Processing Society of Japan (March 2011), etc. It may be performed by using a pattern matching technique with a language expression representing each utterance intention of “question”, “explanation”, and “consideration” prepared in advance.

  In this example, the latter pattern matching technique is adopted, and the question pattern information shown in FIG. 5, which will be described in detail later, the explanation pattern information shown in FIG. 6, the conflict pattern information shown in FIG. 7, and the question information shown in FIG. To determine the utterance intention.

  For example, as illustrated as the question pattern information in FIG. 5, the question utterance is determined using a pattern such as “? The question pattern information shown in FIG. 5, the explanation pattern information shown in FIG. 6, the conflict pattern information shown in FIG. 7, and the question information shown in FIG. 8 are stored in advance in the hard disk 28 in FIG. It shall be issued and used.

  Next, the feature quantity extraction unit 3 obtains the number of occurrences of “question”, “explanation”, and “compatibility” in each utterance sequence of the user and the operator from each intention utterance sequence determined by the utterance intention determination unit 2. Furthermore, when the interrogative shown in FIG. 8 appears during the utterance of “question” as the utterance intention, the utterance is determined as “interrogative question sentence”, the number of appearances is obtained, and the intention feature amount is obtained. The total number of types of words used by the inquirer is extracted from the text data as the used vocabulary feature amount. The intention feature quantity and the used vocabulary feature quantity are set as feature quantity vectors having the intention feature quantity and the used vocabulary feature quantity as elements (step S106). The elements of the feature quantity vector output by the feature quantity extraction unit 3 are the number of questions, the number of explanations, the number of questions, the number of interrogative questions, and the number of words used in different vocabularies. Such a plurality of feature amounts are called feature amount vectors.

  Then, in the estimated information generation unit 4, knowledge amount information (knowledge amount label 4b) as a correct answer label is input to the dialogue indicated by the sound data in accordance with the sound data reproduced in the process of step S102. (Step S107), the feature quantity vector of the intention feature quantity and used vocabulary feature quantity extracted by the feature quantity extraction unit 3 and the input knowledge quantity information (knowledge quantity label 4b) are learned as a one-to-one pair. The model information 5a (estimation used for estimating the knowledge amount from the feature amount is obtained by using the learning data as input and learning the model using the “Support Vector Machine (SVM)” which is a known pattern identification method. Information) is generated (step S108), and then this processing is terminated.

  “Support Vector Machine (SVM)” is a binary discrimination method. When the knowledge amount is estimated in two stages using the “Support Vector Machine (SVM)”, a model for determining whether the knowledge amount is “small” or not is created. When creating a model, for example, learning data corresponding to “small knowledge amount” is set as positive learning data, and other learning data is negative learning data (learning data corresponding to “high knowledge amount”). Do.

  When the knowledge amount is estimated in three stages using "Support Vector Machine (SVM)", three models are created for each stage by learning. Model 1 that discriminates whether “knowledge amount is large” or not, Model 2 that discriminates whether “knowledge amount is medium” or other, and “knowledge amount is small” Or model 3 for discriminating whether it is other than that. For example, in the model 1, learning is performed with learning data corresponding to “a large amount of knowledge” as positive learning data, and other learning data as negative learning data. In the model 2, learning is performed with learning data corresponding to “medium amount of knowledge” as positive learning data and other learning data as negative learning data. In the model 3, learning is performed with learning data corresponding to “small knowledge amount” as positive learning data and other learning data as negative learning data.

  In this example, model learning is performed using “Support Vector Machine (SVM)” for pattern identification. However, the present invention is not limited to this. For example, a neural network using back propagation learning is used. It is also possible to do learning. Alternatively, a method of creating a feature amount vs. knowledge amount table and determining the knowledge amount based on the table may be used. The generated model information 5a is stored in a storage device such as the hard disk 28 in FIG.

  In this example, “question”, “explanation”, “conformity”, and “question word question sentence” are explained as utterance intentions, but only “question” and “explanation” are implemented as utterance intentions. May be. In that case, the processing in the utterance intention determination unit 2 and the feature amount extraction unit 3 of the learning unit and the processing in the utterance intention determination unit 2 and the feature amount extraction unit 3 of the estimation unit are different from the above example.

  In the learning unit, the utterance intention determination unit 2 corresponds to “question” and “explanation” for each utterance for the text data of the utterance sequence in the dialogue between the inquirer and the respondent extracted by the utterance sequence extraction unit 1. The utterance indicating the utterance intention to be determined is determined, and the utterance intention corresponding to the determined utterance is output in association with the utterance to be determined. Further, the feature quantity extraction unit 3 calculates “the number of different words of the inquirer's vocabulary” and calculates the number of appearances of each utterance intention of “question” and “explanation” determined by the utterance intention determination unit 2. . The calculated “number of different words of the inquirer's vocabulary”, the number of appearances of the “question” and “explanation” of the utterance intention determined by the utterance intention determination unit 2, and the feature amount relating to the occurrence state of the utterance intention in the text data Is an intention feature amount. The elements of the feature quantity vector output by the feature quantity extraction unit 3 are the number of questions, the number of explanations, and the number of words different in the used vocabulary.

  In the estimator, the utterance intention discriminator 2 uses the utterer (inquirer) for each utterance for the text data of the utterance string in the dialogue between the inquirer to be estimated and the respondent extracted by the utterance string extractor 1. The utterance indicating the intention of the utterance corresponding to the “question” and “explanation” of the respondent) is determined. The determined utterance is output in association with the utterance intention corresponding to the utterance. Further, the feature amount extraction unit 3 extracts the total number of types of words used by the inquirer from the text data as the estimation target used vocabulary feature amount, and also determines the estimation target utterance intention determined by the utterance intention determination unit 2. The number of appearances of the utterance intention (estimation target utterance intention) of each of the “question” and “explanation” included is calculated. The number of occurrences of the estimated utterance intentions of the calculated “number of different words in the vocabulary used by the inquirer”, the “question” and “explanation” determined by the utterance intention determination unit 2, and the occurrence state of the utterance intention in the text data It is assumed that the estimation target intention feature amount indicates the feature amount. The elements of the feature quantity vector output by the feature quantity extraction unit 3 are the number of questions, the number of explanations, and the number of words different in the used vocabulary.

  Hereinafter, the above-mentioned “question”, “explanation”, “conflict” and “question question” used in the knowledge amount estimation apparatus 10 according to the present embodiment will be described with reference to FIGS. .

  First, the question pattern information shown in FIG. 5 will be described. As the utterance string that represents the question, “Is it?”, “Is it?”, “Is it?”, “Is it?”, “Is it?”, “Is it?” “Kane.”, “Kashirane.”, “Kana.”, And “Koke.” Are illustrated. It should be noted that “ne” matches the pattern even if it does not exist.

  Next, the details of the explanation pattern information shown in FIG. 6 will be described. Among the types of explanation in FIG. 6, “description sentence” and “identification sentence” are, for example, “Shigeru Sakahara,“ Role, Ga Ha, Unagi sentence ”, Cognitive Science Development Vol. 3, Japan Cognitive Science Society, pp. .29-66 Kodansha (May 20, 1990) ".

  That is, the “description sentence” is a sentence that describes the attribute of a certain thing, and the “identification sentence” is a sentence that assigns a value to a certain role. For example, “Japan is a country made up of the Japanese archipelago and surrounding islands” is a descriptive statement that describes the attributes of Japan, and “Japan's prime minister is Mr. Noda.” This is an identification sentence to which the value “Mr.” is assigned.

  For the extraction of these descriptive sentences and identification sentences, the particle “ha” determined by the syntax analysis may be used. However, in this example, “ Use a pattern such as

  The next “explanation of reason” is an utterance for adding a reason why the utterance content (proposition) is established with respect to the preceding utterance. Here, the utterance ends with “So” or “From” is defined as “Explanation of Reason”.

  Finally, “presentation of information / conditions that contribute to the decision-making of the listener” refers to the information that is presented so that the listener can use it to do something. This is defined as, for example, “Hiroyuki Shirakawa,“ Sentence Finished with “Ked” ”Bulletin of Hiroshima University Japanese Language Course No. 6, pp. 9-17 (1996)”. This is a function, which is added here as part of the description.

  In this example, utterances that match the above three types of definitions are extracted as explanatory utterances.

  Next, the conflict pattern information shown in FIG. 7 will be described.

  The role played by the listener in the dialogue is, for example, as described in `` Senko K. Maynard, Japanese Communication: Language and Thought in Context, University of Hawai'i Press, Honolulu (1997) ''. Is to encourage the continuation of the utterance and at the same time indicate that the other person's utterance is understood.

  Therefore, it can be considered that there is some relationship between the number of times of hitting and the amount of knowledge. In this example, as shown in FIG. 7, an utterance in which “Yes”, “Yes”, and “Ye” appear at the end of the utterance is used as a combined utterance, and the number of appearances is used as a feature amount.

  Next, the interrogative information shown in FIG. 8 will be described.

  Among the utterances that take the form of questions, for example, “Masato Tokuhisa, Hiroaki Maeda, Jinichi Murakami, Satoru Ikehara,“ Construction of a text dialogue corpus with dialogue action tags representing mental states ”IEICE technical report , Thought and Language, TL 2007-45, pp. 25-30, (2007) ", there are question sentences in which hypotheses are stated and others that are not.

  The former is considered to fulfill the function of “confirmation” for the purpose of verifying its own assumptions and hypotheses, such as “Is optical line expensive?”, And the latter is “What is optical line?” It is thought that it fulfills the function of requesting purely unknown information.

  In estimating the amount of knowledge of a speaker, it is considered important to extract the latter question that requires unknown information. Therefore, the number of “question word questions” spoken by the operator and the user is also determined. Extracted and used as a feature value.

  In this example, the interrogative question sentence includes the interrogative shown in FIG. 8 among the utterances whose utterance intention label is “question”.

  In the process of step S105 in FIG. 3, the speech intention determination unit 2 performs the above question pattern information shown in FIG. 5, the explanation pattern information shown in FIG. 6, the conflict pattern information shown in FIG. 7, and the question information shown in FIG. Is used to determine the utterance corresponding to the intention by pattern matching.

  Next, processing contents in the estimation unit of the knowledge amount estimation apparatus 10 according to the present embodiment will be described with reference to FIG.

  In FIG. 4, the processing content in the estimation part of the knowledge amount estimation apparatus 10 according to the present embodiment is shown, and dialogue (call) between the user and the operator at the contact center to be estimated (hereinafter referred to as “ When speech data of “estimation target dialogue” is input (step S201), speech recognition processing is performed on the speech data (step S203), and then the utterance string extraction unit is performed as in steps S104 to S106 of FIG. 1, each process of the speech intention determination unit 2 and the feature amount extraction unit 3 (steps S204, S205, and S206) is performed, and thereafter, in the knowledge amount estimation unit 5, the “Support Vector Machine ( SVM) ”is used to calculate the estimation target intention feature amount and the estimation target use vocabulary feature amount (feature vector) obtained by the processing in step S206, as described above. The user who is performing the estimation target dialogue by identifying which knowledge amount the estimation target dialogue corresponds to by comparing with the model information 5a generated by the process (FIG. 3) The knowledge amount of (speaker) is estimated (step S209).

  As for the knowledge amount estimation unit 5, similarly to the estimated information generation unit 4, the knowledge amount is estimated using “Support Vector Machine (SVM)”. Any estimation method may be used. When using “Support Vector Machine (SVM)” to estimate the amount of knowledge using, for example, the above models 1 to 3, discrimination is performed using each of the models 1 to 3, and the positive data of the most suitable model is used. Let the amount of knowledge corresponding to be the speaker's amount of knowledge.

  Next, four types of intention feature amounts extracted by the feature amount extraction unit 3, that is, “1. The number of questions for each of the user and the operator”, “2. The number of question sentences for the user and the operator”, “ 3. “User and Operator Description Frequency” and “4. User and Operator Conflict Count”, and feature quantity vectors composed of vocabulary feature quantities used extracted from text data will be described in detail.

  First, “1. The number of questions for each of the user and the operator” will be described below.

  Of the roles played by questions in dialogue (call) at a contact center or the like, the most basic one is to request information from the other party. Specifically, a question that a user issues is for the operator to provide knowledge that the user lacks, and a question that the operator issues is information about the user's requirements and situation. It is for having you offer.

  In other words, it can be said that the user's number of questions reflects the amount of knowledge that the user lacks, and the operator's number of questions reflects the effort required to grasp the user's requirements and situation. Here, when the operator tries to grasp the user's requirement / situation, it is considered that more labor is required in the dialogue with the user having a small amount of knowledge, so if the number of questions of the user and the operator is used, The amount of knowledge of the user can be estimated.

  In this example, the values of the user's question count and the operator's question count are used as feature values. For example, in the case of the dialogue content illustrated in the text data 500 including the utterance string in FIG. 9, the utterance including the underlined part (“Is it?”, “Is it?”) Is “Question”, In the example of FIG. 9, the number of questions is two times for both the operator and the user.

  Accordingly, here, the value used as the feature amount is “2” for both the number of questions by the operator and the number of questions by the user. 9 is stored in a storage device such as the RAM 24 and the hard disk 28.

  Next, “2. The number of questionable question sentences of the user and the operator” will be described below.

  As explained in FIG. 8, it is considered important to extract questions that require unknown information in estimating the amount of knowledge of the speaker. The number of “sentences” is also extracted and used as a feature quantity.

  In the case of the dialogue content exemplified by the text data 500 including the utterance string of FIG. 9, the question lines shown in FIG. 8 are included, and the portions indicated by the underline of the wavy lines (“how”, “what”) are the question words. In the example of FIG. 9, the number of questionable question sentences of the operator is “1”, the number of questionable question sentences of the user is “2”, and the value used as the feature amount is the questionable question text of the operator Is “1”, and the number of questionable question sentences of the user is “2”.

  Next, “3. Number of times of explanation by user and operator” will be described below.

  In this example, the description is defined as an act of providing information to the other party. Specifically, the three types of utterances shown in FIG. 6 ("description sentence / identification sentence", "explanation of reason", and "presentation of information / conditions that contribute to the listener's decision making") are regarded as explanations.

In the example of FIG. 9, the parts indicated by double underscores (“Masu”, “Issuga”, “Bur” ”,“ Iss ”) correspond to the description, and the values used as the feature values are specifically determined by the operator. The number of explanations is “1”, and the number of explanations by the user is “3”.
Next, “4. The number of conflicts between the user and the operator” will be described below.

  In this example, an utterance in which “Yes”, “Yes”, and “Ye” appear at the end of the utterance shown in FIG. 7 is used as a mutual utterance, and the number of times is used as a feature amount.

  In the example of FIG. 9, a portion (“Yes”) indicated by a broken line corresponds to the conflict, and the value used as the feature amount is specifically “1” for the number of conflicts by the operator and “0” for the number of conflicts by the user. It becomes.

  Next, the vocabulary feature amount used will be described below. Here, the used vocabulary feature amount will be described as “the number of words different in the user's used vocabulary”.

  The number of words is different from the number of words used in a call. The fact that a user needed to use more types of words than other users to finish a call on the same message indicates that the user's communication was inefficient.

  Since it is considered that a user with a small amount of knowledge is not good at accurately communicating his / her state, in this embodiment, in order to characterize a user with a small amount of knowledge, the number of words used in the user's vocabulary differs. Will be used.

  In this embodiment, in order to count the number of different words, first, morphological analysis is performed using an utterance string as input, the utterance string is divided into words, and parts of speech are given to each word. Then, word types are counted only for nouns and unknown words.

  In this embodiment, the morpheme analysis is described in, for example, “Yuji Matsumoto,“ Morphological Analysis System “Chaya” ”, Information Processing Vol. 41, No. 11, November 2000, pp. 1208-1214 (2000)”. The system “tea bowl” is used, but any morphological analysis tool may be used. If the speech recognition result is output at word breaks and part of speech is given, the output word breaks and part of speech may be used as they are.

  For example, in the case of the dialog content illustrated by the text data 500 including the utterance string in FIG. 9, the nouns and unknown words used by the user role are “PC”, “Internet”, and “line” shown in italics. There are three types. Therefore, the value used as the used vocabulary feature amount is “3”.

  Each feature quantity extracted as described above is output as a feature quantity vector. That is, the feature quantity vector used in this embodiment is “1: operator role question count, 2: user role question count, 3: operator role description count, 4: user role description count, 5: operator role count. 6: Number of questions of user role, 7: Number of questions of questioning question of operator role, 8: Number of questions of questioning question of user role, 9: Number of words differing in user role .

  Hereinafter, an evaluation experiment example for the knowledge amount estimation apparatus 10 according to the present embodiment will be described. Here, the amount of knowledge was estimated for each user role in a call at a contact center, and an evaluation experiment was performed.

  Here, as the input to the knowledge amount estimation apparatus 10 according to the present embodiment, the speech recognition result of both the operator and the user of the call at each contact center is used, and the speech sequence obtained as the speech recognition result is used. The above-described various intention feature amounts and used vocabulary feature amounts are extracted using text data.

  In addition, as a correct label (knowledge amount information) used for evaluation of the estimation result of the knowledge amount for each call, each user in each call is classified as “small knowledge amount” or “large knowledge amount” in human subjective evaluation. Whether it is applicable or not is determined according to the criterion shown as criterion information in FIG. 10, and knowledge amount information (knowledge amount label) is assigned to each call. 10 is stored in advance in the hard disk 28, and is read from the hard disk 28 and stored in the RAM 24 for use.

  In this evaluation experiment example, a pair of operator and user voice recognition results and knowledge amount labels are prepared for 180 calls. Among them, the 94 call has the “small knowledge” label, and the 86 call has the “large knowledge” label.

  Learning the model that associates the knowledge amount label given to each call with the intention feature amount and vocabulary feature amount extracted from each call, and the evaluation of the estimation of the knowledge amount learns 9 out of 10 divided data This was performed by 10-fold cross-validation in which a data set of 10 patterns was prepared by using the remaining one as evaluation data.

  In this example, the intention feature used to estimate the knowledge amount is (i) the number of questions for the operator role, (ii) the number of questions for the user role, (iii) the number of times the operator role is explained, (iv) The number of explanations, (v) the number of competing operator roles, (vi) the number of competing user roles, (vii) the number of questionable question sentences for operator roles, (viii) the number of questionable question sentences for user roles, and (ix ) Nine types of words with different user roles. In addition, learning of the correspondence between the intention feature amount and used vocabulary feature amount extracted from each call and the knowledge amount correct label, and the knowledge amount estimation based on the estimation target intention feature amount and the estimation target used vocabulary feature amount, Vector Machine (SVM) ”was used.

  On the other hand, in the accuracy evaluation of this example, “precision” that is an accuracy index is used as an evaluation index. This index indicates how many of the estimated results coincide with the correct knowledge amount label.

  In this way, there is a reason to place importance on the precision. That is, the most characteristic feature of the call at the contact center, which is the target of knowledge amount estimation in this example, is that the number of data is enormous. Considering that it is a major purpose of using the present invention to accurately narrow down the objects to be analyzed from the enormous amount of call data, it can be said that the relevance rate has an important meaning.

  In this evaluation example, it was regarded as a problem of estimating “small amount of knowledge”, and the precision was calculated using the following formula.

Relevance rate = number of “small amount of knowledge” that can be correctly estimated ÷ total number estimated as “small amount of knowledge” In this experimental example, in addition to this relevance rate, “recall rate”, which is an index of completeness Was also evaluated and used as a reference. The recall was calculated using the following formula.

Recall rate = number of “small amount of knowledge” that can be correctly estimated ÷ number of “small amount of knowledge” in all data In the 10-fold cross-validation of this example, 10 learning data / evaluation data from the data for all 180 calls The pair is made and learning and accuracy evaluation are performed. In this experimental example, the evaluation is performed by the micro average of the relevance ratios calculated from the estimation results of the 10 pairs of data.

  In the evaluation of this example, the technique using the expertise of the user's vocabulary shown as Non-Patent Documents 1 and 2 above, and the word feature (bag of words) indicating how many times the user has used As a baseline, a comparison with this example will be made.

  According to Non-Patent Document 1, when specializing a user's vocabulary to be featured, in this embodiment, an average value of the rarity of words used by the user is used. “Inverse Document Frequency (IDF)” is used as an unusual word.

  When the total number of calls is N and the number of calls including the word t is df (t), the IDF value of the word t is calculated using the following equation (1).

idf (t) = log (N / df (t)) ... Formula (1)
In this embodiment, only nouns and unknown words are used to calculate the average IDF of the user's vocabulary.

  As a result of such evaluation, it was confirmed that when the feature amount obtained by the present embodiment was used, it was possible to extract a call of a user with a small knowledge amount with a precision of 0.8 and a recall of 0.43.

  As shown in FIG. 11, when recall rates are uniform, this embodiment can extract a call of a user with a small knowledge amount with the highest relevance rate. In other words, when calls of small users with a small amount of knowledge in the target data can be extracted as much as possible, the percentage of calls (correct answers) of small users with a small amount of knowledge included in the extracted calls is the highest according to this embodiment. It is.

  Furthermore, as shown in FIG. 12, even when the matching ratios are uniform, this embodiment can extract a call of a user with a small knowledge amount with the highest recall rate. In other words, when it is possible to extract the calls of small-knowledge users as accurately as possible, it is according to this embodiment that the calls of small-knowledge users can be extracted most comprehensively. From the above, it can be said that the feature amount obtained in this embodiment is effective in estimating the speaker's knowledge amount.

  As described above, in the knowledge amount estimation apparatus 10 according to the present embodiment, the feature amount of the utterance intention and the feature amount of the used vocabulary that appear in the dialogue between the operator and the user in the contact center or the like are used. The amount of knowledge of the speaker (user) is estimated.

  Specifically, as a function implemented by programmed computer processing, at least an utterance intention determination unit 2, a feature amount extraction unit 3, an estimation information generation unit 4, and a knowledge amount estimation unit 5 are provided, and the utterance intention determination unit 2 Determines the utterance indicating the intention of the utterance corresponding to the question, the explanation, and the companion from the text data of the utterance string in the dialogue between the inquirer and the respondent, and the feature amount extraction unit 3 determines the utterance intention The intention feature amount indicating the feature amount related to the occurrence state of the utterance determined in the section 2 in the text data is extracted, and the total number of word types used by the inquirer is extracted from the text data as the used vocabulary feature amount. The estimated information generation unit 4 includes the intention feature amount and the used vocabulary feature amount extracted by the feature amount extraction unit 3, and the text that is the target of extraction of the intended feature amount and the used vocabulary feature amount. The amount of knowledge corresponding to the text data can be estimated by using, as learning data, knowledge amount information given in advance to indicate the knowledge amount of the inquirer assumed from the dialogue indicated by the text data. Generate estimated information to use.

  The estimation information generated by the estimation information generation unit 4 in this way is stored in a storage device such as the hard disk 28.

  The utterance intention discriminating unit 2 also utters an utterance indicating the intention of the utterance corresponding to the question, explanation, conflict, etc. to the other party from the text data of the utterance string in the dialogue between the inquirer and the respondent whose knowledge is to be estimated. The feature amount extraction unit 3 extracts an estimation target intention feature amount indicating a feature amount related to the occurrence state of the utterance of the estimation target determined by the utterance intention determination unit 2 in the text data, and from the text data The total number of types of words used by the inquirer is extracted as the estimated target vocabulary feature quantity, and further, the utterances that contain interrogative words in the utterance corresponding to the question to the other party are extracted as interrogative question sentences. The feature quantity regarding the occurrence state of the lyrics in the text data is also extracted as the estimation target intention feature quantity, and the knowledge quantity estimation unit 5 extracts the estimation target intention feature quantity extracted by the feature quantity extraction unit 3. And fine estimation target using lexical feature quantity, by using the estimated information stored in advance in the storage device, and estimates the amount of knowledge corresponding to the text data to be estimated.

  Here, the intention feature amount includes the number of questions of the inquirer, the number of questions of the respondent, the number of explanations of the inquirer, the number of explanations of the respondent, the number of inquiries of the inquirer, the number of inquiries of the respondent, the questioner of the inquirer This includes the number of questions and the number of questions of the respondent's question, but the number of questions and the number of questions may be used as necessary.

  In this way, in this example, the amount of knowledge of the speaker is estimated from the dialogue using the features using the utterances corresponding to the questions, explanations, companions, and interrogative questions in the dialogue, It is possible to avoid problems in the case where the speech recognition result is handled as an input for estimating the knowledge amount in the prior art.

  In other words, when speech recognition results using conventional vocabulary expertise are treated as input for estimation of the amount of knowledge, when some highly specialized words appear erroneously in the speech recognition results of a speaker's utterance May be presumed that the speaker's knowledge amount is erroneously high, but in this example, the feature amount using utterances corresponding to questions, explanations, companions, and interrogative questions in the dialogue By utilizing this, the knowledge amount of the speaker is estimated from the dialogue, and such problems can be avoided.

  Thereby, by using the knowledge amount estimation apparatus 10 according to the present embodiment, it is possible to accurately estimate the knowledge amount of the speaker (inquirer) for the item that is the subject of the dialogue based on the dialogue result. it can.

  In addition, this invention is not limited to the example mentioned above, A various deformation | transformation and application are possible within the range which does not deviate from the summary of this invention.

  For example, in this example, the dialogue (call) between the user and the operator in the contact center has been described as an example of the dialogue to be processed. However, the present invention is not limited to this, for example, the bus operation information guidance system of Non-Patent Document 2 above. The present invention is applicable to a system that estimates the amount of knowledge of a user in a dialog using other speech recognition, such as a dialog with a responder to an inquiry from the user.

  The dialogue may be any system (respondent) versus human (inquirer), human (respondent) versus human (inquirer), or any conversation between two parties.

  The text data of the dialog may be anything as long as it is a text-converted dialog, such as a speech recognition result of the dialog, a transcription, or a text chat.

  In the computer configuration shown in FIG. 2, a program for realizing the function of each processing unit according to the present invention is recorded on a computer-readable recording medium, and the program recorded on the recording medium is stored in a computer system. By reading and executing, the processing by each configuration may be executed, or the program may be read by using a communication function not shown.

  The computer-readable recording medium refers to a portable medium such as a flexible disk, a magneto-optical disk, a ROM, and a CD-ROM, and a storage device such as a hard disk built in the computer system.

  The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line.

  The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, what is called a difference file (difference program) may be sufficient.

  As described above, the embodiment for carrying out the present invention has been described in detail with reference to the drawings. However, the specific configuration is not limited to this embodiment, and the scope of the present invention is not deviated. Design etc. are also included.

DESCRIPTION OF SYMBOLS 1 Speech sequence extraction part 2 Speech intention discrimination | determination part 3 Feature-value extraction part 4 Estimated information generation part 4a Intention feature-value 4b Knowledge-amount information (knowledge-amount label)
4c Model generation function 5 Knowledge amount estimation unit 5a Model information 5b Feature amount 5c of call i Collation function 6 Speech recognition result of call i (estimation target)
7 Speech recognition results (call for learning)
8 Knowledge amount of speaker (user) of call i 10 Knowledge amount estimation device 14 Keyboard 16 Mouse 18 Display 22 CPU
24 RAM
26 ROM
28 Hard Disk 30 External I / F (Interface)
50 Communication equipment

Claims (13)

An utterance intention discriminating means for discriminating an utterance indicating an utterance intention corresponding to a question and explanation to the other party from text data of an utterance sequence in the dialogue between the inquirer and the respondent;
Based on the utterance intention to which the utterance determined by the utterance intention determination means corresponds, an intention feature amount indicating a feature amount regarding the occurrence state of the utterance intention in the text data is extracted, and the inquirer is extracted from the text data. Feature quantity extraction means for extracting the total number of types of words used by as vocabulary feature quantities used;
From the dialogue indicated by the text data for the intended feature quantity and the used vocabulary feature quantity extracted by the feature quantity extracting means, and the text data as the extraction target of the intended feature quantity and the used vocabulary feature quantity Knowledge amount estimation information for generating estimation information used for estimation of the knowledge amount corresponding to the text data by using knowledge amount information given in advance as the learning data indicating the knowledge amount of the inquirer assumed Generating means;
A knowledge amount estimation information generating apparatus comprising: The knowledge amount estimation information generating apparatus according to claim 1, wherein the utterance intention determination unit determines, from the text data, an utterance indicating an utterance intention corresponding to a conflict in addition to the question and explanation to the other party. The knowledge amount estimation information generation according to claim 1, wherein the intention feature amount includes the number of questions of the inquirer, the number of questions of the respondent, the number of times of explanation of the inquirer, and the number of times of explanation of the respondent. apparatus. The knowledge amount estimation information generation device according to claim 3, wherein the intention feature amount further includes at least one of the inquiry number of the inquirer and the inquiry number of the respondent. The feature amount extraction means extracts an utterance including a questionable word in an utterance corresponding to the question as a questionable question sentence, and the intention feature amount includes the number of questions of the inquirer, the number of questions of the respondent, Including the number of times the inquirer explained, the number of times the answerer explained, the number of inquiries from the inquirer, the number of inquiries from the respondent, the number of questions from the inquirer, and the question from the respondent The knowledge amount estimation information generation device according to claim 2, comprising at least one of the number of times. Storage means for storing in advance the estimation information generated by the knowledge amount estimation information generation device according to any one of claims 1 to 5,
An estimation target utterance intention discriminating means for discriminating an utterance indicating a utterance intention corresponding to a question to a partner, an explanation, from text data of an utterance sequence in a dialogue between an inquirer and a respondent as an estimation target of a knowledge amount;
Together with the utterance is determined by the estimated target speech intention determining means on the basis of the utterance intention appropriate, to extract the estimated target intended feature quantity indicating a feature amount relating to occurrence of the utterance intention of the text data, the text An estimation target feature amount extraction means for extracting a total number of types of words used by the inquirer from data as a used vocabulary feature amount;
Corresponding to the text data to be estimated using the estimation target intention feature quantity and the estimation target used vocabulary feature quantity extracted by the estimation target feature quantity extraction unit and the estimation information stored in the storage unit Knowledge amount estimating means for estimating the knowledge amount;
A knowledge amount estimation device. The knowledge amount estimation apparatus according to claim 6, wherein the estimation target utterance intention determination unit determines, from the text data, an utterance indicating an utterance intention corresponding to a conflict in addition to the question and explanation to the other party. The knowledge amount estimation according to claim 6 or 7, wherein the estimation target intention feature amount includes the number of questions of the inquirer, the number of questions of the respondent, the number of times of explanation of the inquirer, and the number of times of explanation of the respondent. apparatus. The knowledge amount estimation device according to claim 8, wherein the estimation target intention feature amount further includes at least one of the inquiry number of the inquirer and the inquiry number of the respondent. The estimation target feature amount extraction unit extracts an utterance including a questionable word in an utterance corresponding to the question as a question word question sentence, and the estimation target intention feature amount includes the number of questions of the inquirer, the answerer's question number, Including the number of questions, the number of explanations of the inquirer, the number of explanations of the respondent, the number of questions of the inquirer, the number of questions of the answerer, the number of questions of the questioner of the inquirer, and the number of questions of the respondent The knowledge amount estimation device according to claim 7, comprising at least one of the number of interrogative question sentences. A knowledge amount estimation information generation method in a knowledge amount estimation information generation device including an utterance intention determination unit, a feature amount extraction unit, and a knowledge amount estimation information generation unit,
An utterance intention determination step of determining an utterance indicating intention of an utterance corresponding to a question to a partner, an explanation, from text data of an utterance string in a dialogue between an inquirer and an answerer by the utterance intention determination means ,
The feature amount extraction means extracts an intention feature amount indicating a feature amount related to the occurrence state of the utterance determined in the utterance intention determination step in the text data, and a word used by the inquirer from the text data. A feature amount extraction step of extracting the total number of types as vocabulary feature amounts used;
The intention feature amount and the used vocabulary feature amount extracted in the feature amount extraction step by the knowledge amount estimation information generating means , and the text data that is the target of extraction of the intention feature amount and the used vocabulary feature amount. The knowledge amount information given in advance as the knowledge amount indicating the knowledge amount of the inquirer assumed from the dialogue indicated by the text data is used as learning data, so that the knowledge amount corresponding to the text data is used for estimation. A knowledge amount estimation information generation step for generating estimation information;
A knowledge amount estimation information generation method comprising: A knowledge amount estimation method in a knowledge amount estimation device including an estimation target utterance intention determination unit, an estimation target feature amount extraction unit, and a knowledge amount estimation unit,
A storage step of storing in advance in a storage device the estimation information generated by the knowledge amount estimation information generation method according to claim 11;
The estimation target utterance intention discrimination means discriminates an utterance indicating the intention of the utterance corresponding to the question and explanation to the other party from the text data of the utterance string in the dialogue between the inquirer and the respondent as the estimation target of the knowledge amount. An estimation target utterance intention determination step;
The estimation target feature amount extraction unit extracts an estimation target intention feature amount indicating a feature amount related to a generation state in the text data of the utterance determined in the estimation target utterance intention determination step. An estimation target feature amount extraction step for extracting the total number of types of words used by the inquirer as an estimation target use vocabulary feature amount; and
By using the estimation target intention feature quantity and the estimation target used vocabulary feature quantity extracted in the estimation target feature quantity extraction step by the knowledge quantity estimation means, and the estimation information stored in the storage device , A knowledge amount estimation step for estimating the knowledge amount corresponding to the text data;
A knowledge amount estimation method comprising:   A computer for causing the computer to function as the knowledge amount estimation information generation device according to any one of claims 1 to 5 or the knowledge amount estimation device according to any one of claims 6 to 10. program.