JP6190984B1 - Question answer support device and question answer support system - Google Patents

Abstract

A question answering support device and a question answering support system that can provide an appropriate answer to a question from a user at an early stage and improve the satisfaction of the user. An acquisition unit for acquiring a question character string based on the user's question including at least one sentence, a derivation unit for deriving question data from the question character string, and a plurality of accumulated past questions Calculating a similarity between the corresponding reference data group and the database 17 in which answer character string groups corresponding to the accumulated past answers are stored, the question data, and the reference data group; A calculation unit 13 that selects first reference data included in the reference data group based on similarity, and a selection unit 14 that selects a first response character string included in the response character string group based on the first reference data. And an output unit 15 for outputting the first answer character string. [Selection] Figure 1

Description

  The present invention relates to a question answer support device and a question answer support system.

  Conventionally, as a system for transmitting an answer to a user's question via the Internet, for example, a Q & A system disclosed in Patent Document 1 and a question answering device disclosed in Patent Document 2 are disclosed. Further, for example, a help desk support system disclosed in Patent Document 3 is disclosed as a system that supports an operator who responds to a user inquiry.

  The Q & A system disclosed in Patent Literature 1 is a Q & A system that transmits and receives questions and answers to the questions between users via a plurality of client systems connected to a server connected via a communication network. The client system has a display means for displaying an image on a screen, an input means for inputting information, a designation means for designating a location on the screen, and a server via a communication network, (1) Search result question information or search result answer information obtained by separating associated and accumulated questions and answers and searching separately for each accumulated information, and (2) Searching for questions or answers A search information input area for inputting search information and a search result display area for displaying a plurality of searched questions or answers. Receiving means for receiving screen data to be displayed on the screen from the server, search result question information received from the receiving means, and a question answer display area for the searched question and answer based on the search result answer information Control means for controlling the display means so as to display on the screen, identification information for identifying a user, and search information for searching for a question or an answer input by the input means via the communication network And transmitting means for transmitting to the server.

  In the question answering device disclosed in Patent Document 2, in the question answering device, the question accepting unit accepts the question text, the partial question extracting unit extracts the partial question that is a part of the question text, and the answer obtaining unit is the partial Obtain the answer of the question, the question editor replaces the part of the partial question in the question sentence with the answer of the partial question, the answer acquisition unit acquires the answer of the replaced question, and the answer output unit replaces the question The answer is output.

  The help desk support system disclosed in Patent Document 3 is a help desk support system in which an operator who has received a telephone inquiry from a customer provides reference information for dealing with the customer, and the contents of the call between the customer and the operator A speech recognition server that converts speech data into text data, a corresponding text analysis unit that extracts one or more keywords from the text data acquired by the speech recognition server, and a keyword extracted by the corresponding text analysis unit A keyword search unit that searches the knowledge data to obtain reference information, and a display unit that displays the keyword and bibliographic information of the reference information corresponding to the keyword on an operator terminal.

JP 2013-214294 A JP 2014-229275 A JP 2014-174938 A

  Here, in a help desk provided in a company, a question from a user may include relatively specialized content. Further, help desk operators are required to answer questions from users at an early stage. For this reason, an operator with little experience may spend time before providing an answer to a question, or may not provide an answer desired by the user.

  As a countermeasure, for example, a method of accumulating a huge amount of data such as past questions and answers and acquiring a past data close to the user's question by keyword search or the like is used. However, since there are many relatively specialized questions in the help desk, there are cases in which completely different answers are given to questions including the same keyword. For this reason, it is difficult to limit the content of a question only by a simple keyword search. As a result, the answer desired by the user cannot be determined from the search result, or an incorrect answer may be provided, which may lead to a decrease in user satisfaction.

  In this regard, Patent Document 1 has a configuration in which the stored question and answer are separated and individually searched. When this configuration is diverted to a help desk, there is a possibility that a different answer is provided to the user for the same question as the content received in the past. For this reason, a unified answer cannot be provided in the help desk, which may lead to a decrease in user satisfaction.

  In Patent Document 2 and Patent Document 3, since a part of a question, a keyword, and the like are extracted to obtain an answer, there are many answer options, and it is difficult to specify an appropriate answer. For this reason, it is necessary to spend time searching for an answer desired by the user from the search result, which may lead to a decrease in user satisfaction.

  Due to the circumstances described above, it is desired that an appropriate answer to a question from a user can be provided at an early stage and the satisfaction of the user can be improved.

  Therefore, the present invention has been devised in view of the above-described problems, and the object of the present invention is to provide an appropriate answer at an early stage in response to a question from the user and improve the satisfaction of the user. An object is to provide a question answering support device and a question answering support system.

A question answering support device according to a first invention is a question answering support device for assisting an operator who answers a question from a user, and obtains a question character string based on the user's question including at least one sentence; A deriving unit for deriving question data from the question character string, a reference data group corresponding to the accumulated past questions, and an answer character string group corresponding to the accumulated past answers are stored. A calculation unit that calculates similarity between the database, the question data, and the reference data group, and selects first reference data included in the reference data group based on the similarity; and the first reference a selector for selecting the first answer string having the said answer character string groups based on the use data, the first and an output unit for outputting an answer string, in the database, the previous plurality of quality The reference matrix calculated based on is stored, and the derivation unit divides the question character string into morphemes, converts each of the divided morphemes into question matrices, and the reference matrix for the reference matrix matrix addition, subtraction, multiplication, and characterized that you derive the query data by performing at least one of the division.

Question answer support apparatus according to the second invention, in the first invention, the calculating unit, and calculates an inner product of the reference data group and said query data as said degree of similarity.

In the question answering support device according to a third invention, in the first invention or the second invention , the calculation unit calculates a first similarity between the question data and the first reference data, and the question data , Calculating a second similarity with the second reference data included in the reference data group, and adding to the first reference data based on a relative value between the first similarity and the second similarity. And determining whether to select the second reference data.

A question answering support device according to a fourth aspect of the present invention is a question answering support device used in a help desk for answering a question from a user via a network, wherein a question character string based on the user's question including at least one sentence is obtained. An obtaining unit for obtaining, a deriving unit for deriving question data from the question character string, a reference data group corresponding to a plurality of accumulated past questions, and answer characters corresponding to a plurality of accumulated past answers A calculation unit that calculates a similarity between the database storing the column group, the question data, and the reference data group, and selects first reference data included in the reference data group based on the similarity; A selection unit that selects a first answer character string included in the answer character string group based on the first reference data; an output unit that outputs the first answer character string to the user terminal of the user; Provided, in the database, the previous reference matrix calculated based on the plurality of questions are stored, the deriving unit divides the question text into morphemes, questions matrix each of the divided said morphemes It was converted to the sum of the reference matrix to the question matrix, subtraction, multiplication, and characterized that you derive the query data by performing at least one of the division.

A question answering support system according to a fifth invention is a question answering support system for assisting an operator who answers a question from a user, and obtains a question character string based on the user's question including at least one sentence; , Deriving means for deriving question data from the question character string, a reference data group corresponding to the accumulated past questions, and an answer character string group corresponding to the accumulated past answers are stored. Calculation means for calculating similarity between the database, the question data, and the reference data group, and selecting first reference data included in the reference data group based on the similarity; and the first reference Selecting means for selecting the first answer character string of the answer character string group based on the data, and output means for outputting the first answer character string to the terminal used by the operator; Provided, in the database, the previous reference matrix calculated based on the plurality of questions are stored, said deriving means divides the question text into morphemes, questions matrix each of the divided said morphemes It was converted to the sum of the reference matrix to the question matrix, subtraction, multiplication, and characterized that you derive the query data by performing at least one of the division.

According to the first to fourth inventions, the calculating unit calculates the similarity between the question data and the reference data group, and selects the first reference data based on the calculated similarity. The selection unit selects the first answer character string based on the first reference data. For this reason, by using the similarity for each data, it is possible to search for past answers suitable for the content of the question, compared to, for example, searching for an answer based on a keyword or the like in the user's question. That is, it is possible to obtain a highly accurate answer to the user's question. Thereby, it is possible to provide an appropriate answer at an early stage with respect to the question from the user, and to improve the user's satisfaction.

According to the first to fourth aspects, the question data is derived from a question character string based on a user question including at least one sentence. For this reason, the number of cases searched as answer candidates can be reduced. That is, even when a vast amount of data such as past questions and answers is used or when there are many relatively specialized questions, it is possible to search for necessary answers in a short time. Thereby, it becomes possible to obtain an answer desired by the user in a short time.

According to the first to fourth inventions, the reference data group and the answer character string group correspond to the accumulated past questions and answers. For this reason, an answer character string corresponding to a question received in the past can be selected and an answer can be provided to the user. This makes it possible to provide appropriate answers that are unified within the help desk.

According to the first to fourth aspects, the derivation unit derives the question data by performing at least one of addition, subtraction, multiplication, and division of the reference matrix with respect to the question matrix. At this time, the question data can include a possibility that another morpheme is used before and after each morpheme included in the question character string. Therefore, even in the case of a question that uses the same morpheme but has slightly different contents, the reference data and the answer character string can be selected based on the context of each morpheme. Thereby, a highly accurate answer can be obtained with respect to a user's question, and a user's satisfaction can be improved further.

According to the second and third inventions, the calculating unit calculates the inner product of the question data and the reference data group as the similarity. The range of conditions for selecting reference data in the reference data group can be expanded. Thereby, the number of answer character strings selected according to the user's question can be arbitrarily changed, and the user's satisfaction can be further improved.

According to the third aspect , the calculating unit determines whether to select the second reference data based on the relative value between the first similarity and the second similarity. For this reason, for example, even when the second similarity is slightly inferior to the first similarity, the second reference data can be selected without leaking. Thereby, an optimal answer can be acquired with respect to a user's question, and it becomes possible to further improve a user's satisfaction.

According to the fifth aspect , the calculating means calculates the similarity between the question data and the reference data group, and selects the first reference data based on the calculated similarity. The selection means selects the first answer character string based on the first reference data. For this reason, since the degree of similarity for each data is used as compared with the case of searching for an answer based on, for example, a keyword in a user's question, a past answer suitable for the content of the question can be searched. That is, it is possible to obtain a highly accurate answer to the user's question. Thereby, it is possible to provide an appropriate answer at an early stage with respect to the question from the user, and to improve the user's satisfaction.

According to the fifth aspect , the question data is derived from a question character string based on a user question including at least one sentence. For this reason, the number of cases searched as answer candidates can be reduced. That is, even when a vast amount of data such as past questions and answers is used or when there are many relatively specialized questions, it is possible to search for necessary answers in a short time. Thereby, it becomes possible to obtain an answer desired by the user in a short time.

According to the fifth aspect , the reference data group and the answer character string group correspond to the past questions and answers accumulated. For this reason, an answer character string corresponding to a question received in the past can be selected and an answer can be provided to the user. This makes it possible to provide appropriate answers that are unified within the help desk.

FIG. 1 is a block diagram illustrating an overall configuration of a question answer support system according to an embodiment. FIG. 2 is a schematic diagram illustrating an example of an answer character string output by the question answer support device according to the embodiment. FIG. 3A is a schematic diagram illustrating an example of the configuration of the question answering support apparatus according to the embodiment, and FIG. 3B is a schematic diagram illustrating an example of the function of the question answering support apparatus according to the embodiment. FIG. 4 is a schematic diagram illustrating an example of each function of the question answer support device according to the embodiment. FIG. 5 is a schematic diagram illustrating an example of question data in the embodiment. FIG. 6 is a schematic diagram illustrating an example of similarity in the embodiment. FIG. 7 is a flowchart illustrating an example of the operation of the question answer support system according to the embodiment.

  Hereinafter, an example of a question answering support system 100 according to an embodiment to which the present invention is applied will be described with reference to the drawings.

(Embodiment: Configuration of Question Answer Support System 100)
With reference to FIGS. 1-3, an example of a structure of the question answer support system 100 in this embodiment is demonstrated. FIG. 1 is a block diagram showing the overall configuration of a question / answer support system 100 in the present embodiment.

  The question answer support system 100 includes a question answer support device 1. The question answering support device 1 is connected to at least one of the server 2, the operator terminal 3, and the user terminal 5 through the public communication network 4, for example.

<Question answering support device 1>
The question answering support apparatus 1 according to the present embodiment is mainly used in a company help desk, and is used as an apparatus for assisting an operator who answers a question from a user. For example, as shown in FIG. 2A, an operator who has received a question from the user inputs a question character string based on the user's question to the operator terminal 3. After obtaining the question character string from the operator terminal 3, the question / answer support apparatus 1 selects an optimum answer character string for the question character string and outputs the answer character string to the operator terminal 3. After confirming the answer character string, the operator creates and sends a reply to the user. For example, as illustrated in FIG. 2B, the question answer support device 1 may acquire the question character string when the operator directly inputs the question character string into the question answer support device 1. In this case, the question answer support device 1 selects an answer character string that is optimal for the question character string, and outputs the answer character string.

  FIG. 3A is a schematic diagram illustrating an example of the configuration of the question answer support device 1. An electronic device such as a personal computer (PC) is used as the question answering support device 1. The question answer support device 1 includes a housing 10, a CPU 101, a ROM 102, a RAM 103, a storage unit 104, and I / Fs 105 to 107. Each component 101 to 107 is connected by an internal bus 110.

  A CPU (Central Processing Unit) 101 controls the entire question answer support device 1. A ROM (Read Only Memory) 102 stores an operation code of the CPU 101. A RAM (Random Access Memory) 103 is a work area used when the CPU 101 operates. The storage unit 104 stores various information such as answer character strings. As the storage unit 104, for example, an HDD (Hard Disk Drive), a data storage device such as an SSD (solid state drive) or a floppy disk is used. For example, the question answering support apparatus 1 may have a GPU (Graphics Processing Unit) (not shown). By having a GPU, it is possible to perform higher-speed arithmetic processing than usual.

  The I / F 105 is an interface for transmitting and receiving various types of information to and from the server 2 and the like via the public communication network 4. The I / F 106 is an interface for transmitting / receiving information to / from the input unit 108. For example, a keyboard is used as the input part 108, and an administrator of the question answering support system 100 inputs various information or a control command of the question answering support device 1 through the input part 108. The I / F 107 is an interface for performing transmission / reception of various information with the output unit 109. The output part 109 outputs various information stored in the storage unit 104 or the processing status of the question answer support apparatus 1. A display is used as the output portion 109, and for example, a touch panel type may be used.

  FIG. 3B is a schematic diagram illustrating an example of functions of the question answer support device 1. The question answer support device 1 includes an acquisition unit 11, a derivation unit 12, a calculation unit 13, a selection unit 14, an output unit 15, an input unit 16, and an information DB 17. Note that the function shown in FIG. 3B is realized by the CPU 101 executing a program stored in the storage unit 104 or the like using the RAM 103 as a work area. Moreover, each structure 11-17 may be controlled by artificial intelligence, for example. Here, “artificial intelligence” may be based on any known artificial intelligence technology.

<Acquisition unit 11>
The acquisition unit 11 acquires a question character string based on a user question including at least one sentence. As the acquisition unit 11, for example, as shown in FIG. 4, the user asks the operator a content such as “What should I do if △ Δ is changed to □□?”. Based on this question, the operator creates a question character string in the operator terminal 3 or the like, so that the acquisition unit 11 acquires the question character string. The acquisition unit 11 acquires, for example, a question character string created by the operator based on a user's question sentence sent from the user terminal 5, for example, and directly acquires a question sent from the user as a question character string, for example. Alternatively, a question character string created by the operator based on a voice dialogue with the user may be acquired, or a question selected by the user from a question prepared in advance may be acquired as the question character string. Here, “one sentence” includes, for example, a character string having at least a subject and a predicate, and may include a character string that can understand the intention of a question from the user.

  The question character string acquired by the acquisition unit 11 may have user information such as the sex and age of the user, for example. Thereby, the question answer support apparatus 1 can select an answer corresponding to the user information.

<Deriving unit 12>
The deriving unit 12 derives question data from the question character string. The question data is represented by a matrix as shown in FIG. 5, for example. Here, the “matrix” includes not only a matrix indicated by n rows and m columns but also a vector indicated by 1 row and n columns or n rows and 1 column, and the same applies to the following.

  The derivation unit 12 divides the sentence of the question character string into any one or more units among all grammatical structural units (hereinafter referred to as morphemes) such as morphemes, words, phrases, and clauses. At this time, for example, a sentence included in the question character string is divided into m morphemes and the like.

  The deriving unit 12 converts each of m divided morphemes of the m question character strings into m question matrices (first question matrix, second question matrix,..., M-th question matrix). The deriving unit 12 refers to, for example, the reference matrix and converts each of the divided morphemes of the question character string into a question matrix represented as a matrix (vector) of 1 row and n columns. Note that the derivation unit 12 may convert, for example, into a single question matrix represented by m rows and n columns based on the morphemes of the m question character strings divided.

  The reference matrix has information in which a matrix is assigned to each of a plurality of morphemes acquired in advance. The reference matrix has a plurality of 1-row n-column matrices in which the column corresponding to the morpheme or the like is “1” and the other columns are “0”. As the reference matrix, a morpheme extracted from a plurality of past questions acquired in advance, a morpheme set in advance by an administrator, or the like is used. Note that the value of n depends on the number of morphemes or the like acquired in advance as a reference matrix.

  The deriving unit 12 derives the question data by performing at least one of addition, subtraction, multiplication, and division of the reference matrix with respect to the question matrix. At this time, for example, each question matrix of 1 row and n columns is multiplied by a reference matrix, and m question vectors represented by 1 row and n columns (first question vector, second question vector,. Question data having a question vector) is derived. Note that the derivation unit 12 may derive one piece of question data represented by m rows and n columns by multiplying, for example, an m row and n column question matrix by a reference matrix.

  The reference matrix is represented by, for example, a co-occurrence matrix of n rows and n columns. The number of rows and columns of the reference matrix corresponds to the number (n) of morphemes or the like acquired in advance in the reference matrix. The value of the reference matrix indicates the possibility that another morpheme or the like is used before or after a specific morpheme or the like when, for example, a morpheme or the like in the reference matrix is used in a sentence. For this reason, for example, the value of the first question vector (5, 13, 2, 20, 3,...) Is changed to “□□”, “XX” before and after “OO” corresponding to the first question vector. ”,“ ΔΔ ”,“ no ”, and the like. The reference matrix is a matrix calculated based on a plurality of accumulated past questions.

  In addition to deriving the question data using the above-described standard matrix and reference matrix, the deriving unit 12 may derive the question data using, for example, standard image data and reference image data. In this case, a plurality of morphemes acquired in advance and an image (question image) assigned to each morpheme are used as the reference image data, and assigned to the question image and the question image as reference image data. A plurality of matrices represented by 1 row and n columns are used.

  The deriving unit 12 converts each morpheme or the like obtained by dividing the question character string into a corresponding question image with reference to the reference image data. The deriving unit 12 refers to the reference image data, and derives the question data by selecting a matrix corresponding to the converted question image. Even in this case, the content of the reference image data indicates the possibility that other morphologies etc. may be used before and after a specific morpheme, for example, when a morpheme assigned to a question image is used in a sentence.

  The reference image data is formed by assigning a question image to a morpheme extracted from a past question acquired in advance or a morpheme set in advance by an administrator or the like. The reference image data is formed by assigning a matrix calculated based on a morpheme extracted from a past question acquired in advance to a question image.

<Calculation unit 13>
As shown in FIG. 4, the calculation unit 13 calculates the similarity between the question data and the reference data group, and selects the reference data included in the reference data group based on the similarity. The reference data group is a data group corresponding to a plurality of accumulated past questions, and has a plurality of reference data.

  Similarity may be calculated in three stages or more, such as 100 minutes, 10 stages, or 5 stages, or may be calculated in, for example, two stages based on a threshold. For example, as the degree of similarity increases, the question character string corresponding to the question data is likely to be similar to a past question corresponding to the reference data.

  The reference data corresponds to a past question including at least one sentence. In addition to being represented by a matrix, the reference data is, for example, “○ / / / △△ / is / □□ / is / / but /, / What should I do? Thus, the question may be stored in a form in which the question is divided into morphemes or the like, or in a form displayed in a superordinate concept of a plurality of questions such as “question about □□”. When the reference data is a matrix, it is represented by a plurality of 1-row n-column matrices, for example.

  For example, as illustrated in FIG. 6, the calculation unit 13 may calculate the inner product of the question data and the reference data group as the similarity (cosine similarity). In this case, the similarity value is calculated as a value between −1 and 1. As each reference data (first reference data, second reference data,...) Of the reference data group, a matrix based on a morpheme or the like divided in advance is used. The matrix of each reference data is derived from the character string of the past question by the derivation unit 12 as in the case where the question data is derived from the question character string described above, for example.

  For example, each of the first question vector to the m-th question vector in the question data is D11 to D1m, and each matrix in the first reference data is D21 to D2m. In this case, the calculation unit 13 calculates a value obtained by dividing the absolute value of D11 and D21 with respect to the inner product of D11 and D21. The calculation unit 13 performs the same calculation as described above for each of D12 to D1m and D22 to D2m, and adds the obtained values to obtain the similarity between the question data and the first reference data. calculate. Similar to the above-described content, the calculation unit 13 calculates the similarity between the question data and each reference data, and calculates the similarity between the question data and the reference data group. Note that the calculation unit 13 may perform the above-described calculation using a matrix in which the question data and each reference data are standardized in advance. In that case, there is no need to divide the absolute value for the inner product. If the numbers of the question data and the reference data group do not match, the numbers are calculated according to the small number, and the extra number is excluded from calculation.

  The calculation unit 13 may select a plurality of reference data. In this case, the similarity is preferably calculated in three or more stages. For this reason, for example, when setting to select a plurality of reference data having a high similarity to the question data, the selection method can be greatly increased. That is, the calculation unit 13 can select a plurality of reference data in descending order of similarity, and can also select reference data indicating relatively similar similarity as one group from the reference data group. For this reason, the method for selecting the reference data can be arbitrarily set according to the field of the corresponding question such as a help desk in which the question answering support apparatus 1 is used. Thereby, it is possible to obtain a highly accurate answer corresponding to the field of the question, and it is possible to further improve user satisfaction.

  For example, the calculation unit 13 selects reference data having a similarity of 60% or more. In this case, it is possible to limit the number of selections of the reference data as compared with the conventional keyword search or the like. Further, the calculation unit 13 selects, for example, three reference data in descending order of similarity. In this case, selection of reference data with high accuracy can be realized.

  For example, the calculation unit 13 may select reference data by comparing the calculated values of similarities. For example, the calculation unit 13 calculates the first similarity between the question data and the first reference data and the second similarity between the question data and the second reference data, and the first similarity and the second similarity are calculated. Whether to select the second reference data in addition to the first reference data may be determined based on the relative value to the degree.

  For example, the calculation unit 13 may calculate the similarity with the question data based on the appearance frequency of a specific morpheme or the like in the reference data group. In this case, the usage frequency of a specific morpheme or the like is extracted from the question data and compared with the appearance frequency of the specific morpheme or the like in the reference data. For this reason, compared with the case where the inner product mentioned above is calculated, the load accompanying the calculation process can be reduced. In the reference matrix, the value of the matrix corresponding to each morpheme or the like may be set to be large according to the importance. As a result, calculation corresponding to morphemes that are important in the field of questions from users can be performed.

<Selection unit 14>
As shown in FIG. 4, the selection unit 14 selects an answer character string included in the answer character string group based on the reference data selected by the calculation unit 13 (broken line frame in FIG. 4). The answer character string group is a data group corresponding to a plurality of accumulated past answers, and a plurality of answer character strings (first answer character string, second answer character string, third answer character string,...). Have Each answer character string is associated with corresponding reference data based on past questions and answers. One reply character string may have a plurality of answers, for example, “Reply 1 XX is recommended.” And “Please answer 2 XX.”

<Output unit 15>
The output unit 15 outputs the answer character string selected by the selection unit 14. The output unit 15 may output the answer character string to the operator terminal 3 via the I / F 105, or may output the answer character string to the output part 109 via the I / F 107, for example.

<Input unit 16>
The input unit 16 receives a question character string or the like input to the operator terminal 3 or the question answer support device 1. The input unit 16 receives various types of information input from the operator terminal 3 via the I / F 105 and receives various types of information input from the input unit 108 via, for example, the I / F 106.

<Information DB 17>
The information DB 17 stores various types of information such as a standard matrix, a reference matrix, a reference data group, and a reply character string group, and stores standard image data and reference image data according to settings. The information DB 17 may store, for example, a question character string, a question matrix, question data, similarity, and the like. Various types of information are stored as a database of various types of information in the storage unit 104 embodied by an HDD, SSD, or the like.

  The reference matrix, the reference matrix, the reference data group, the answer character string group, the reference image data, and the reference image data are constructed using machine learning based on, for example, past questions or answers acquired in advance. A known technique can be used for machine learning, and a neural network, a support vector-machine, a naive Bayes method, deep learning, or the like may be used. Note that the standard matrix, reference matrix, reference data group, answer character string group, standard image data, and reference image data are saved in, for example, an Excel file or csv file format, and are convenient for the administrator. It may be saved in any file format according to.

  Each structure 11-16 memorize | stores various information in information DB17 as needed, or takes out various information. Note that various types of information such as a reference matrix, a reference matrix, a reference data group, and an answer character string group may be constructed based on, for example, questions and answers published on the Internet via the public communication network 4.

<Server 2>
The server 2 stores data (database) related to various types of information. In this database, for example, information sent via the public communication network 4 is accumulated. The server 2 may store information similar to the information DB 17, for example, and may transmit and receive various types of information to and from the question answer support device 1 via the public communication network 4. As the server 2, for example, a database server on a network may be used. The server 2 may be used instead of the storage unit 104 and the information DB 17 described above.

<Operator terminal 3>
The operator terminal 3 is installed in a help desk that uses the question answering support device 1, and may be owned by a help desk contractor, for example. As the operator terminal 3, for example, an electronic device such as a personal computer is used.

<Public communication network 4>
The public communication network 4 (network) is an Internet network or the like to which the question answering support device 1 and the operator terminal 3 are connected via a communication circuit. The public communication network 4 may be a so-called optical fiber communication network. The public communication network 4 is not limited to a wired communication network, and may be realized by a wireless communication network.

<User terminal 5>
As the user terminal 5, a mobile phone is mainly used, and other than that, it is embodied in all electronic devices other than electronic devices such as a smartphone, a tablet-type terminal, a wearable terminal, a personal computer, and an IoT (Internet of Things) device. Things may be used. A user transmits a question to operator terminal 3 using user terminal 5, for example.

  According to the present embodiment, the calculation unit 13 calculates the similarity between the question data and the reference data group, and selects reference data (for example, first reference data) based on the calculated similarity. The selection unit 14 selects an answer character string (for example, a first answer character string) based on the reference data. For this reason, by using the similarity for each data, it is possible to search for past answers suitable for the content of the question, compared to, for example, searching for an answer based on a keyword or the like in the user's question. That is, it is possible to obtain a highly accurate answer to the user's question. Thereby, it is possible to provide an appropriate answer at an early stage with respect to the question from the user, and to improve the user's satisfaction.

  According to the present embodiment, the question data is derived from a question character string based on a user question including at least one sentence. For this reason, the number of cases searched as answer candidates can be reduced. That is, even when a vast amount of data such as past questions and answers is used or when there are many relatively specialized questions, it is possible to search for necessary answers in a short time. Thereby, it becomes possible to obtain an answer desired by the user in a short time.

  Further, according to the present embodiment, the reference data group and the answer character string group correspond to the accumulated past questions and answers. For this reason, an answer character string corresponding to a question received in the past can be selected and an answer can be provided to the user. This makes it possible to provide appropriate answers that are unified within the help desk.

  Further, according to the present embodiment, the derivation unit 12 derives the question data by performing at least one of addition, subtraction, multiplication, and division of the reference matrix with respect to the question matrix. At this time, the question data can include the possibility that other morphemes and the like are used before and after each morpheme included in the question character string. Therefore, even in the case of a question that uses the same morpheme but has slightly different contents, the reference data and the answer character string can be selected based on the context of each morpheme. Thereby, a highly accurate answer can be obtained with respect to a user's question, and a user's satisfaction can be improved further.

  Further, according to the present embodiment, the calculation unit 13 calculates the inner product of the question data and the reference data group as the similarity. For this reason, the range of conditions for selecting the reference data included in the reference data group can be expanded. Thereby, the number of answer character strings selected according to the user's question can be arbitrarily changed, and the user's satisfaction can be further improved.

  Further, according to the present embodiment, the calculation unit 13 determines whether to select the second reference data based on the relative value between the first similarity and the second similarity. For this reason, for example, even when the second similarity is slightly inferior to the first similarity, the second reference data can be selected without leaking. Thereby, an optimal answer can be acquired with respect to a user's question, and it becomes possible to further improve a user's satisfaction.

  In addition, according to the present embodiment, by using the similarity calculated in three or more stages, the calculation unit 13 is similar to the question data in addition to the selection of reference data that completely or partially matches the question data ( Reference data including morphemes and the like (including the same concept, synonyms, etc.) can be selected. For this reason, an optimal answer character string can be acquired even for a question with unprecedented content. Thereby, a user's satisfaction can further be improved.

(First Embodiment: Operation of Question Answering Support System 100)
Next, an example of the operation of the question answer support system 100 in the present embodiment will be described. FIG. 7 is a flowchart showing an example of the operation of the question answer support system 100 in the present embodiment.

<Acquisition means: S110>
As shown in FIG. 7, first, a question character string based on the user's question is obtained (acquisition means: S110). The acquisition unit 11 acquires a question character string input or received by the question answer support device 1.

  The question character string is transmitted from the operator terminal 3 to the question answer support device 1 by the operator, or may be directly input to the question answer support device 1 by the operator, for example.

  The acquisition unit 11 may store the acquired question character string in the information DB 17.

<Deriving means: S120>
Next, the question data is derived from the question character string acquired by the acquisition unit 11 (derivation means: S120). The derivation unit 12 acquires a question character string from the acquisition unit 11 or the information DB 17, and acquires a reference matrix and a reference matrix from the information DB 17.

  As shown in FIG. 5, for example, the derivation unit 12 divides the sentence of the question character string into morphemes and converts each morpheme and the like into a question matrix based on the reference matrix. For example, the first morpheme “XX” of the question character string is converted into the first question matrix (1, 0, 0, 0,...) Based on the reference matrix, and the second morpheme of the question character string is obtained. “No” is converted into the second question matrix (0, 0, 0, 0, 1,...) Based on the reference matrix. The derivation unit 12 converts each morpheme and the like into a question matrix (first question matrix, second question matrix,..., M-th question matrix) represented by 1 row and n columns, and also supports each morpheme, for example. It may be converted into a matrix of m rows and n columns including a 1 row and n column matrix.

  Thereafter, the deriving unit 12 derives the question data by performing at least one of addition, subtraction, multiplication, and division of the reference matrix with respect to the question matrix. At this time, for example, in addition to deriving question data having each question vector (first question vector, second question vector,..., M-th question vector) obtained by multiplying each question matrix by a reference matrix, For example, the query data represented by a matrix of m rows and n columns may be derived by multiplying a reference matrix by a question matrix of m rows and n columns.

  The derivation unit 12 may store the derived question data and the converted question matrix in the information DB 17.

<Calculation means: S130>
Next, as shown in FIG. 4, the similarity between the question data and the reference data group is calculated, and reference data (for example, first reference data) included in the reference character string group is selected based on the similarity. (Calculation means: S130). The calculation unit 13 acquires question data from the derivation unit 12 or the information DB 17 and acquires a reference data group from the information DB 17.

  For example, as illustrated in FIG. 6, the calculation unit 13 calculates the inner product of the question data and the reference data group as the similarity. In this case, the calculation unit 13 calculates the inner product of each reference data with respect to the question data. The calculation unit 13 selects reference data based on the calculated inner product value (similarity). For example, the calculation unit 13 may select reference data having the highest degree of similarity, or may select any number of reference data in descending order of similarity, and may include a plurality of references included in groups having similarities. Data may be selected, and the criteria for selection can be arbitrarily set by an administrator or the like.

  Note that the calculation unit 13 may store the selected reference data and the calculated similarity in the information DB 17. At this time, the question data whose similarity is calculated may also be stored in the information DB 17.

<Selection means: S140>
Next, as shown in FIG. 4, an answer character string (for example, a first answer character string) included in the answer character string group is selected based on reference data (for example, first reference data) (selecting means: S140). The selection unit 14 acquires reference data from the calculation unit 13 or the information DB 17 and acquires a reply character string group from the information DB 17.

  For example, the selection unit 14 selects an answer character string corresponding to the question based on the reference data “question about □□”. The answer character string has an answer such as “recommended to be xx”, and may have a plurality of answers, for example. In this case, the selection unit 14 can select a plurality of answers for one reference data. The criterion for selecting an answer by the selection unit 14 can be arbitrarily set by an administrator or the like.

  The selection unit 14 may store the selected answer character string in the information DB 17. At this time, reference data corresponding to the answer character string may also be stored in the information DB 17.

<Output means: S150>
Next, an answer character string (for example, a first answer character string) is output (output means: S150). The output unit 15 acquires a response character string from the selection unit 14 or the information DB 17.

  In addition to outputting the answer character string to the operator terminal 3, the output unit 15 may output the answer character string to the output part 109, for example. For example, the output unit 15 may output past questions corresponding to the reference data in addition to the answer character string. For example, the output unit 15 may select and output an answer included in the answer character string. A criterion for selecting an answer by the output unit 15 can be arbitrarily set by an administrator or the like.

  Thereby, the operation of the question answer support system 100 in the present embodiment is completed. Note that the operator may answer the user after considering the answer based on the answer character string output by the output unit S150, and the answering method is arbitrary. Further, based on the output answer character string, an operator or the like may further create a question character string and perform the acquisition unit S110.

  According to the present embodiment, the calculation unit S130 calculates the similarity between the question data and the reference data group, and selects reference data (for example, first reference data) based on the calculated similarity. The selection unit S140 selects an answer character string (for example, a first answer character string) based on the reference data. For this reason, since the degree of similarity for each data is used as compared with the case of searching for an answer based on, for example, a keyword in a user's question, a past answer suitable for the content of the question can be searched. That is, it is possible to obtain a highly accurate answer to the user's question. Thereby, it is possible to provide an appropriate answer at an early stage with respect to the question from the user, and to improve the user's satisfaction.

  According to the present embodiment, the question data is derived from a question character string based on a user question including at least one sentence. For this reason, the number of cases searched as answer candidates can be reduced. That is, even when a vast amount of data such as past questions and answers is used or when there are many relatively specialized questions, it is possible to search for necessary answers in a short time. Thereby, it becomes possible to obtain an answer desired by the user in a short time.

  Further, according to the present embodiment, the reference data group and the answer character string group correspond to the accumulated past questions and answers. For this reason, an answer character string corresponding to a question received in the past can be selected and an answer can be provided to the user. This makes it possible to provide appropriate answers that are unified within the help desk.

  Note that, according to the present embodiment, the acquisition unit 11 may acquire, for example, an answer character string output by the output unit 15 or an answer character string that the operator has answered to the user with reference to the answer character string. . In this case, the question character string and the answer character string obtained by the obtaining unit 11 are accumulated in the information DB 17, and based on the accumulated question character string and the answer character string, a reference matrix, a reference matrix, a reference data group, At least one of the answer character string groups may be updated. Thereby, the accuracy of the output answer can be improved, and the user's satisfaction can be further improved.

  In addition to the above, according to the present embodiment, the user may directly send a question to the question answer support device 1 via the public communication network 4 using the user terminal 5. In this case, the acquisition unit 11 of the question answering support apparatus 1 that has received the user's question acquires the user's question as a question character string. Then, the operation | movement mentioned above is performed and the output part 15 may output the reply character string based on a question character string to the user terminal 5. FIG. For this reason, the user can obtain an appropriate answer to the question at an early stage without using an operator. Thereby, it becomes possible to improve a user's satisfaction. In addition, since the number of operators in the help desk can be reduced, personnel costs can be reduced.

  Although several embodiments of the present invention have been described, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the invention described in the claims and the equivalents thereof.

1 ... Question answer support device,
10 ... Case,
11 ... acquisition part,
12 ... Deriving part,
13 ... calculation part,
14 ... selection part,
15 ... output part,
16 ... input part,
17 ... Information DB,
100 ... Question answer support system,
101 ... CPU,
102 ... ROM,
103 ... RAM,
104 ... storage part,
105-107 ... I / F,
108 ... input part,
109 ... the output part,
110 ... Internal bus,
2 ... server,
3 ... operator terminal,
4 ... Public communication network,
5 ... User terminal

Claims (5)

A question answering support device for assisting an operator who answers a question from a user,
An acquisition unit for acquiring a question character string based on the user's question including at least one sentence;
A derivation unit for deriving question data from the question character string;
A reference data group corresponding to a plurality of accumulated past questions, and a database storing answer character string groups corresponding to a plurality of accumulated past answers;
A calculation unit for calculating a similarity between the question data and the reference data group, and selecting first reference data included in the reference data group based on the similarity;
A selection unit for selecting a first answer character string of the answer character string group based on the first reference data;
An output unit for outputting the first answer character string ,
In the database, a reference matrix calculated based on the plurality of past questions is stored,
The derivation unit includes:
Dividing the question string into morphemes;
Convert each of the divided morphemes into a query matrix,
The addition of the reference matrix to the question matrices, subtraction, multiplication, and question answering support device characterized that you derive the query data by performing at least one of the division. The calculating unit, the question answers support apparatus of claim 1, wherein the calculating the inner product of the reference data group and said query data as said degree of similarity. The calculation unit includes:
Calculating a first similarity between the question data and the first reference data;
Calculating a second similarity between the question data and the second reference data included in the reference data group;
The determination as to whether or not to select the second reference data in addition to the first reference data is made based on a relative value between the first similarity and the second similarity. The question answering support apparatus according to 1 or 2 . A question answering support device used in a help desk that answers a question from a user via a network,
An acquisition unit for acquiring a question character string based on the user's question including at least one sentence;
A derivation unit for deriving question data from the question character string;
A reference data group corresponding to a plurality of accumulated past questions, and a database storing answer character string groups corresponding to a plurality of accumulated past answers;
A calculation unit for calculating a similarity between the question data and the reference data group, and selecting first reference data included in the reference data group based on the similarity;
A selection unit for selecting a first answer character string of the answer character string group based on the first reference data;
An output unit that outputs the first answer character string to the user terminal of the user ;
In the database, a reference matrix calculated based on the plurality of past questions is stored,
The derivation unit includes:
Dividing the question string into morphemes;
Convert each of the divided morphemes into a query matrix,
The addition of the reference matrix to the question matrices, subtraction, multiplication, and question answering support device characterized that you derive the query data by performing at least one of the division. A question answering support system for assisting an operator who answers a question from a user,
Obtaining means for obtaining a question character string based on the user's question including at least one sentence;
Derivation means for deriving question data from the question character string;
A reference data group corresponding to a plurality of accumulated past questions, and a database storing answer character string groups corresponding to a plurality of accumulated past answers;
A calculation means for calculating a similarity between the question data and the reference data group, and selecting first reference data included in the reference data group based on the similarity;
Selecting means for selecting a first answer character string of the answer character string group based on the first reference data;
Output means for outputting the first answer character string to a terminal used by the operator ;
In the database, a reference matrix calculated based on the plurality of past questions is stored,
The derivation means includes
Dividing the question string into morphemes;
Convert each of the divided morphemes into a query matrix,
Addition of the reference matrix to the question matrix, subtraction, multiplication, and question answering support system characterized that you derive the query data by performing at least one of the division.

Images