JP7098928B2 - Question answer system, question answer method - Google Patents

Description

The present invention relates to a question-and-answer system and a question-and-answer method .

Call staff often handle inquiries from local governments and companies and accept applications. However, dealing with call staff requires a great deal of human resources. Therefore, it is required that local governments and companies appropriately manage personal information by utilizing networks to respond to inquiries and accept applications more efficiently. Further, as an interactive system that guides an answer to a question while interacting with a user, the ones shown in Patent Document 1, Patent Document 2, and Patent Document 3 have been proposed.

Japanese Unexamined Patent Publication No. 7-20937 Japanese Unexamined Patent Publication No. 11-445755 Japanese Patent Application Laid-Open No. 2003-58464

Chatbots are rapidly gaining attention due to the need for improved work quality and work efficiency, such as quick response to inquiries, expansion of the time available for inquiries, and stable response regardless of operator skill. Chatbots solve problems for user questions by interacting with the computer on behalf of the operator. However, in the conventional service using a chatbot, the answer is dry and monotonous, or the answer is patterned. For this reason, even if the amount of knowledge of chatbots is increased and the percentage of correct answers to inquiries is increased, customers (users) can obtain sufficient satisfaction if they want to gain familiarity and sympathy for chatbot answers. Not.

In view of the above-mentioned problems, the present invention estimates the emotional state of the customer from the question contents of the customer (user), and can give an answer including the emotional expression corresponding to the emotional state, and can improve the customer satisfaction. The purpose is to provide an answer system and a question answering method .

The question-and-answer system according to one aspect of the present invention is a question-and-answer system including a user interface server that provides an interactive user interface and a learning data generation system, and the learning data generation system provides information on answers to inquiries. In response to the above, the emotional expression unit that generates learning data to which information about the emotional expression corresponding to the user's emotions estimated from the inquiry is added, and the generated learning data are transmitted to the answer generation engine that derives the answer to the inquiry. The user interface server includes a learning data providing unit, an input unit for inputting an inquiry from a user, an output unit for outputting the input inquiry to the answer generation engine, and the answer generation in response to the inquiry. The receiver that receives the answer derived from the engine and the information of the emotional expression corresponding to the user's emotion estimated from the user's question content from the answer generation engine and the answer received from the answer generation engine. The emotional expression unit is provided with a transmission unit for transmitting to the user, and the emotional expression unit is not only an answer to which an emotional expression corresponding to an emotion estimated from the information of the question in the inquiry is added, but also an emotion different from the emotion. An answer to which an emotional expression corresponding to is added is generated, and the inquiry and an answer to which a plurality of different emotional expressions are added are generated as the learning data, and are estimated from the information of the question in the inquiry . It is characterized in that an answer to which an emotional expression corresponding to a certain emotion is added is a recommended answer.

The question-and-answer system according to one aspect of the present invention includes an interface in which the user interface provided by the user interface server can input emotional information, and the user interface server inputs the emotional information of the user input from the interface. It is characterized in that it is input together with the inquiry and the answer corresponding to the inquiry and the emotion information of the user derived by the answer generation engine is output to the user.

The question-and-answer system according to one aspect of the present invention is characterized in that the emotional expression unit generates the learning data using environmental parameters indicating environmental conditions including at least a date and time, a season, and weather.

The question-and-answer system according to one aspect of the present invention is characterized in that the emotional expression unit generates learning data including a character icon corresponding to the answer in addition to the answer to which the emotional expression information is added. ..

The question-and-answer method according to one aspect of the present invention is a question-and-answer method including a user interface server that provides an interactive user interface and a learning data generation system. The emotional expression process that generates learning data in which information on emotional expression corresponding to the user's emotions estimated from the inquiry is added to the information of the answer to the inquiry, and the learning data generated by the learning data providing unit are inquired. A learning data providing process to be transmitted to an answer generation engine for deriving an answer to an answer, an input process for inputting an inquiry from a user in the user interface server, and an output process for outputting the input inquiry to the answer generation engine. The reception process of receiving from the answer generation engine an answer to which the information of the emotional expression corresponding to the user's emotion estimated from the user's question content, which is derived by the answer generation engine in response to the inquiry, is added. An answer including an transmission process unit that transmits an answer received from the answer generation engine to the user, and an answer to which an emotional expression corresponding to an emotion estimated from the information of the question in the inquiry is added in the emotional expression process. Not only that, an answer to which an emotional expression corresponding to an emotion different from the emotion is added is generated, and the inquiry and an answer to which a plurality of different emotional expressions are added are generated as the learning data, and the inquiry is made. It is characterized in that the recommended answer is an answer to which an emotional expression corresponding to the emotion estimated from the information of the question in is added .

The question-and-answer method according to one aspect of the present invention is a question-and-answer method including a user interface server that provides an interactive user interface and a learning data generation system. The emotional expression process that generates learning data in which information on emotional expression corresponding to the user's emotions estimated from the inquiry is added to the information of the answer to the inquiry, and the learning data generated by the learning data providing unit are inquired. The learning data provision process to be transmitted to the answer generation engine for deriving the answer to the answer, the input process for inputting the inquiry from the user in the user interface server, the output process for outputting the input inquiry to the answer generation engine, and the above. The reception process of receiving from the answer generation engine the answer to which the information of the emotional expression corresponding to the user's emotion estimated from the user's question content, which is derived by the answer generation engine in response to the inquiry, is added, and the above. Including the transmission process unit that transmits the answer received from the answer generation engine to the user, in the emotion expression process, only the answer to which the emotion expression corresponding to the emotion estimated from the information of the question in the inquiry is added. Instead, an answer to which an emotional expression corresponding to an emotion different from the emotion is added is generated, and the inquiry and an answer to which a plurality of different emotional expressions are added are generated as the learning data in the inquiry. The recommended answer includes an answer to which an emotional expression corresponding to the emotion estimated from the information of the question is added .

According to the present invention, a question-and-answer system and a question that can estimate the emotional state of the customer from the question contents of the customer (user), can answer including the emotional expression corresponding to the emotional state, and improve the customer satisfaction. It is possible to provide a learning method of the answer system. For example, when a question and an answer are interactively asked by an answer generation engine, a word of emotional expression indicating "empathy" can be added according to the "emotion" of the other party. As a result, the emotion of the user can be estimated and the customer satisfaction can be improved.

It is a block diagram which shows the outline of the question answer system which concerns on 1st Embodiment of this invention. It is a figure which shows the dialogue screen of the user terminal in the question answering system which concerns on 1st Embodiment of this invention. It is explanatory drawing of an example of the method of guiding a user at the end of a talk. This is an example of a word to be added among the emotional parameters and the information to be added. It is a figure of an example of the operation of the character icon which shows that the customer's emotion was understood. This is an example of a word added at the end of a call when the date and time are used as environmental parameters. It is a functional block diagram which shows the processing in a learning data generation system. It is a figure of an example of FAQ data. It is a figure of an example of learning data. It is a figure of an example of the answer of the answer generation engine to a question from a user. It is a figure which shows the dialogue screen of the user terminal in the question answering system which concerns on 2nd Embodiment of this invention.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing an outline of a question-and-answer system 1 according to a first embodiment of the present invention.

As shown in FIG. 1, the question-and-answer system 1 according to the embodiment of the present invention includes a user terminal 11, a user interface server 12, an answer generation engine 13, and a learning data generation system 14. The user terminal 11 and the user interface server 12 can be connected by the network 15.

The user terminal 11 is a terminal owned by the user (customer) 21. As the user terminal 11, a smartphone, a tablet terminal, a PC (Personal Computer), or the like can be used. When using a smartphone or tablet terminal, a dedicated application program is prepared. Further, as the user terminal 11, a smart speaker, a game machine, a dedicated terminal, or the like may be used.

The user interface server 12 provides a user interface that serves as a contact point for inquiries about businesses and applications. In the present embodiment, the user interface server 12 provides an interactive user interface between the user terminal 11 and the answer generation engine 13 that can solve a problem while enjoying a conversation with a character icon. The user interface server 12 also provides an interface with the call staff 22, and the user 21 can directly talk with the call staff 22 to solve a question that cannot be answered by the answer generation engine 13. If the answer generation engine 13 can solve all the problems, the call staff 22 can be unnecessary.

The answer generation engine 13 interactively outputs an answer to an inquiry from the user 21 while having a chat conversation with the user 21. When the answer generation engine 13 receives the text sentence of the question from the user terminal 11 via the user interface server 12, the answer generation engine 13 analyzes the input text sentence, derives (estimates) the optimum answer, and passes through the user interface server 12. The derived answer is returned to the user terminal 11. When deriving the optimum answer, the answer generation engine 13 performs, for example, morpheme analysis, semantic analysis, conversion processing of synonyms into words, and the like for the text sentence of the input inquiry to obtain keywords and meanings. Construct optimal data that fits and use it as input data to your own inference engine.

As the answer generation engine 13 that performs conversational processing, any format may be used as long as the optimum answer to the question can be returned. The answer generation engine 13 in the present embodiment performs rule-based inference including, for example, a general machine learning model using a neural network or the like as an estimation model, or a question unit and an answer unit for deriving an answer according to a predetermined rule. It is composed of functions using an AI (artificial intelligence) system such as a general inference engine.

In the present embodiment, the answer generation engine 13 uses FAQ (Frequently Asked Questions) data accumulated as a history of answers to past inquiries based on the customer-handling experience of the call staff, and uses the FAQ (Frequently Asked Questions) data to answer the questions in the inquiry. The answer is learned as learning data. For this reason, it is possible to improve the accuracy of answers by utilizing the FAQ data, which is the history of answers to customers corresponding to past questions, which has been accumulated over a long period of time by the call staff. Further, when the answer generation engine 13 is trained, the wording of the text sentence of the answer is converted for each character so as to match the image (impression) of the character of the character icon, and the emotional expression corresponding to the customer's emotion is shown. The FAQ data to which the information is added is used as the learning data. As a result, the affinity between the image of the character and the word can be improved, and the answer generation engine 13 can be constructed with a high degree of satisfaction when the customer obtains the answer to his / her own question.

As a method of adding information indicating an emotional expression corresponding to the customer's emotion, a word indicating the emotional expression corresponding to the customer's emotion may be added to the answer, or the wording may be modified to indicate the emotional expression. .. In addition, although the example using the character icon is explained here, when the character icon is not used, the process of converting the wording of the text sentence of the answer for each character so as to match the image (impression) of the character. Can be omitted. When a character is used, the answer generation engine 13 may derive the icon of the character to be displayed together with the derivation of the answer, and display the icon of the character corresponding to the generated answer content.

Further, as described above, in the present embodiment, the inference engine of the answer generation engine 13 is not trained by using the FAQ data in which a word indicating emotional expression is added to the answer as training data, but the answer generation engine 13 is used. Therefore, a database is provided, and words corresponding to characters or words indicating emotional expressions are stored in the above database according to each combination of words and emotional parameters extracted from the text sentence of the question. It may be configured. Then, the estimation engine in the answer generation engine 13 converts the text sentence of the question into a phrase corresponding to the extracted word and character, and adds a word indicating the emotional expression corresponding to the emotional parameter to the answer generation unit. It may be configured separately in the generation engine 13.

In the present embodiment, the question-and-answer system 1 is configured such that the user 21 and a virtual character talk with each other by, for example, chatting using a text sentence.
However, it may be configured to have a voice recognition function and output an answer to an inquiry from the user 21 by voice while having a voice conversation with the user 21. In this case, the voice of the answer may convert the text sentence into voice data having a frequency and intonation corresponding to the image of the character, and may further improve the affinity with the image of the character.

At this time, the user interface server 12 inputs the inquiry of the user 21 and converts the text sentence of the inquiry or the voice of this inquiry into a text sentence. Then, the user interface server 12 outputs the text sentence of the question or the text sentence of the converted question to the answer generation engine 13 or the learning data generation system 14. Further, when the answer generation engine 13 supplies the text sentence of the answer to the above question, the user interface server 12 converts the text sentence of the answer into voice data and causes the user terminal 11 to output the answer by voice. do.

As described above, the learning data generation system 14 transfers the accumulated call staff's past customer correspondence history to the answer generation engine 13 from the FAQ data corresponding to the call staff 22 in order to reflect it in the answer generation engine 13. The learning data of is generated and provided. That is, the call staff, for example, with the support of a BPO (Business Process Outsourcing) center or the like, directly talks with the user 21 by voice communication or text chat, and performs daily customer support work. When the BPO center receives information on the answers to new questions from the call staff, it accumulates the information on the questions / answers from the call staff and updates the history. Further, the learning data generation system 14 generates learning data from newly accumulated inquiries and their answers at predetermined intervals, and causes the answer generation engine 13 to relearn the inference engine. May be good.

In this way, in the learning data generation system 14, a large number of FAQ data based on the experience of the call staff are accumulated as a history, and the FAQ data of the accumulated history is gradually increased.
When the learning data generation system 14 generates the learning data of the answer generation engine 13, the learning data generation system 14 performs a process of converting the answer in the FAQ data accumulated as a history into a phrase suitable for the character, and also shows an emotional expression. The FAQ data to which the information of the words is added, the wording is converted, and the word indicating the emotional expression is added is stored in the database as learning data. The learning data generation system 14 converts the wording in this database and sends the FAQ data of the answer to which the data of one word indicating the emotional expression is added to the answer generation engine 13 as learning data at predetermined intervals, and generates the answer. The engine 13 is made to perform learning or re-learning using this learning data.

As a result, the answer generation engine 13 performs a process of converting into a phrase that matches the character, and learns the inference engine (machine learning or rule-based) using the FAQ data to which one word data indicating emotional expression is added as learning data. Rebuild the inference engine). By this learning, when the answer generation engine 13 receives the text sentence of the question from the user terminal 11 via the user interface server 12, the analysis result of the text sentence of the question is input and matched with the character of the character icon (described later) itself. An optimal answer to which a word of data indicating conversion to the wording and emotional expression is added is derived and returned to the user terminal 11 via the user interface server 12. That is, the user interface server 12 inputs an answer with a word indicating an emotional expression corresponding to the user's emotion, which is estimated from the user's question content in the inquiry, from the answer generation engine 13. Then, the user interface server 12 outputs to the user terminal 11 an answer to which a word indicating an emotional expression corresponding to the emotion of the user supplied from the answer generation engine 13 is added.

The network 15 is an IP (Internet Protocol) network called, for example, the Internet or an intranet. The configuration of lower layers such as the physical layer and data link layer includes LAN (Local Area Network), wireless LAN (IEEE802.11), mobile communication network, CATV (Community Antenna TeleVision) communication network, and virtual private network (Virtual Private). Network), telephone network, satellite communication network, etc. are assumed.

Next, the outline of the question-and-answer system 1 according to the first embodiment of the present invention will be described.

The question-and-answer system 1 according to the first embodiment of the present invention can be used for inquiries about services and procedures of companies and local governments, and inquiries about how to use applications. When making an inquiry about a business or an application, the user 21 accesses the URL (Uniform Resource Locator) of the user interface server 12 by his / her own user terminal 11.

When the URL of the user interface server 12 is accessed by the user terminal 11, the dialogue screen 100 as shown in FIG. 2 is displayed on the user terminal 11. FIG. 2 is a diagram showing an example of a dialogue screen of the user terminal 11 in the question-and-answer system according to the first embodiment of the present invention. As shown in FIG. 2A, a character icon 101 is displayed in the center of the dialogue screen 100. A text representing guidance or the like is displayed in the balloon 102 of the character icon 101. An input field 103 is provided below the dialogue screen 100. As shown in FIG. 2B, the user 21 inputs a text sentence (or voice) in the input field 103.

The user interface server 12 transmits a text sentence (or a text sentence converted from voice) input by the user 21 supplied from the user terminal 11 to the answer generation engine 13. The answer generation engine 13 analyzes the text sentence of the inquiry input by the user 21, derives the optimum answer to the question of this inquiry by the inference engine, and outputs it to the user interface server 12. Then, the user interface server 12 returns the answer derived from the answer generation engine 13 to the user terminal 11. As a result, as shown in FIG. 2C, the text 105 input by the user 21 and the text 106 indicating the answer by the character icon 101 are alternately displayed on the dialogue screen 100.

In this embodiment, the character icon 101 is displayed in the center of the left and right sides of the screen so as not to be hidden by a balloon. This is because the screen configuration, which is located in the center of the left and right sides of the screen and is always observable, has a psychological effect of giving the user 21 a sense of security due to the symmetry effect. Of course, the character icon 101 may be arranged on one of the left and right sides, and the text indicating the answer by the character icon 101 and the input text of the user 21 may be separated into left and right. In this case, the text indicating the answer and the input text of the user 21 can be understood at a glance.

As described above, in the present embodiment, the user 21 can obtain an answer to an inquiry about a business or an application while having a conversation with the character icon 101. Further, in the present embodiment, when the answer generation engine 13 is trained, the wording is converted according to the character icon 101, and the FAQ data to which the emotional expression is added is used as the learning data. Therefore, an emotional expression is added to the answer to the question of the user 21.

For example, when the user 21 inputs a text sentence "Please tell me the procedure when getting married" in the input field 103, the answer generation engine 13 expresses the feeling of "joy" of "Wow, congratulations" and then. Answers such as "Please contact the Citizen's Division of each city for marriage registration, etc." are answered as the words of the character icon 101.

In addition, the answer generation engine 13 sends a message of concern at the end of the call. That is, the answer generation engine 13 says, for example, "Do your best today" in the morning, "Do your best this week" on Monday morning, and "Let's take a rest at this time" at night. It is added at the end as a word of the character icon 101.

As a method of guiding the user at the end of the call, the answer generation engine 13 asks the user 21 a question that guides his / her impressions and opinions about the service, analyzes a word returned from the user 21, and makes one at the end of the call. You may decide what to say. FIG. 3 is an explanatory diagram of an example of such a method of guiding a user at the end of a call. In this example, the answer generation engine 13 asks the user 21 an impression or opinion about the service, such as "Is it useful? Is there any other consultation?" 3 seconds after the business question.

Then, when the user 21 returns a positive word such as "Thank you", the answer generation engine 13 returns a word such as "Wow, you are welcome ..." to the user 21. When the user 21 returns a negative word such as "It is not useful", the answer generation engine 13 returns a word such as "Garn, sorry for lack of power, ..." to the user 21.

As described above, in the present embodiment, the answer generation engine 13 is made to output the text of the human answer in consideration of emotions. As a result, the answer from the answer generation engine 13 to the customer's question content is brought closer to the answer from the call staff, and the emotional expression corresponding to the emotion at the time of inquiry is exchanged according to the image of the character displayed on the screen. The user 21 can obtain an answer with a word added, and the satisfaction level of the user 21 with respect to the answer to the inquiry can be improved. This will be further described.

As the components of customer satisfaction in the question-and-answer system 1, "accuracy of answers" and "quick follow-up", as well as "knowledge and ability of responders" and "friendly response" are important elements. In particular, "friendly response" such as observing etiquette, estimating the customer's situation and feelings (hereinafter sometimes referred to as understanding) and answering inquiries is very important as a component of customer satisfaction. It has become. In the case of the call staff 22, the customer is dealt with with due consideration of "friendly correspondence". When the answer generation engine 13 responds, as in the case of the call staff 22, "accuracy of answer", "quick follow-up", and "friendly response" that understands the customer's situation and emotions are required. Be done.

It is considered important to add a word as an emotional expression that shows "empathy" according to the "emotion" of the other party as a customer response that makes you feel "friendly correspondence". That is, human emotions include "trouble," "anger," "joy," and "sadness." If you add a word as an emotional expression that matches the other person's emotions, such as "I'm in trouble" if the other person's emotions are "trouble", "I'm sorry" if the other person's emotions are "anger", etc. Psychological effects improve customer satisfaction.

Therefore, in the present embodiment, when learning the answer generation engine 13, the learning data generation system 14 has an emotional parameter for the FAQ data in which the inquiry and the operator's answer to the inquiry are accumulated as a history. Is added and output to the answer generation engine 13 as learning data. The answer generation engine 13 estimates the user's emotion according to the user's emotion parameter by learning the learning data to which such an emotion parameter is added, and adds information as an emotion expression to answer. As the information to be added, as described above, in addition to adding a word as an emotional expression, a part of the text sentence may be modified to a wording indicating the emotional expression.

FIG. 4 is an example of a word to be added among the emotion parameters and the information to be added. For example, for the emotional parameter of "joy", add the word "Wow, congratulations". For the emotional parameter of "sadness", add the word "looks sad". For the emotional parameter of "trouble", add the word "looks like you are in trouble". For the emotional parameter of "anger", add the word "Don't get angry. I'm sorry". As a result, even when the response generation engine 13 corresponds to the user 21, a word indicating "empathy" can be added according to the "emotion" of the other party, and customer satisfaction can be improved.

In the above example, the response generation engine 13 estimates the customer's emotion corresponding to the emotion parameter, adds a word corresponding to this emotion, and returns the response to the user. However, not only the emotional expression may be expressed by a character string, but also the character icon 101 may be moved so as to perform an action indicating the emotional expression, as the answer generation engine 13 understands the customer's emotion. FIG. 5 is an example of the operation of the character icon 101 indicating that the customer's emotions are understood. By moving the character icon 101 so as to perform an action indicating emotional expression, emotions can be conveyed to the user by body language, and customer satisfaction is improved. FIG. 5A shows a character icon 101 having a normal facial expression. Further, FIG. 5B shows a character icon 101 waving and greeting. FIG. 5C shows a character icon 101 with a troubled expression.

In this case, the learning data generation system 14 causes the answer generation engine 13 to learn the animation data (moving image data) of the motion showing the above-mentioned emotional expression in response to the answer in the FAQ data as the learning data. As a result, the response generation engine 13 derives the animation of the character that operates as an emotional expression corresponding to the customer's emotion, together with the text sentence of the answer corresponding to the customer's emotion, and outputs it to the user interface server 12. Then, the user interface server 12 transmits to the user terminal 11 an animation of a character that operates as an emotional expression corresponding to the customer's emotion, together with an answer to which the emotional expression supplied from the answer generation engine 13 is added.

Also, humans most impressively remember the last letter or word in conversation. From this, it is important to add words that indicate memorable emotional expressions at the end of the conversation. Therefore, in the present embodiment, the learning data generation system 14 adds a word indicating the environmental condition such as the date and time, the season, and the weather at the end of the conversation with the user as a word indicating the emotional expression. Then, this is output to the answer generation engine 13 as learning data. The answer generation engine 13 learns the learning data to which a word of emotional expression corresponding to such an environmental parameter is added, so that it corresponds to the environmental situation at that time and the emotion of the user, and is effective at the end of the call. Add a word to show the emotional expression of.

In addition, when deriving the answer from the learning data generated using the environmental parameters, the environmental parameters such as date and time, season, and weather are acquired from the clock function built in the user interface server 12 or the external weather database, and the answer is given. It may be transmitted to the generation engine 13 and the answer generation engine 13 may derive the answer using the environmental parameters, or when the answer generation engine 13 derives the answer, the clock function built in the answer generation engine 13 or an external one. Environmental parameters such as date and time, season, and weather may be acquired from a weather database or the like, and an answer using the environmental parameters may be derived.

FIG. 6 is an example of a word added at the end of a call when the date and time are used as environmental parameters. For example, for the environmental parameter of "morning", add the word "Do your best today as well". For the environmental parameter of "Monday morning", add the word "Do your best this week". For the environmental parameter of "night", add the word "Please take a rest at this time". For the environmental parameters of "Friday afternoon-Saturday morning", add the word "We have a good weekend for each other".

As mentioned above, the environmental parameters also include seasonal information. For example, if the season is "winter", add the word "Don't catch a cold because it's cold!" ..
Further, in addition to the above-mentioned environmental parameters, the weather during conversation may be extracted from a weather forecast or the like from a network such as the Internet and used as a weather parameter. In this case, add a word corresponding to the weather during conversation, for example, if it is going to rain, add a word "Don't forget your umbrella!".
In addition, since the letters or words shown at the beginning of the conversation may be impressively remembered, a word with environmental parameters added at the start of the conversation with the user is added, and this is used as learning data for the answer generation engine. It may be output to 13. For example, if the season is "winter," add the phrase "Good morning, it's cold today."

FIG. 7 is a functional block diagram showing processing in the learning data generation system 14. As shown in FIG. 7, the learning data generation system 14 includes an expression conversion unit 151, an emotion expression unit 152, and a learning data providing unit 153.

The expression conversion unit 151 converts the wording of the answer text in the FAQ data in the history of the inquiry and the answer of the call staff to this inquiry into the text of the wording that matches the character image of the character icon 101. ..

The emotion expression unit 152 adds one word data (emotion parameter) of emotion expression to the text sentence of the answer to the user according to the wording of the character image. At this time, the emotion expression unit 152 extracts information indicating an emotion parameter from the text sentence of the inquiry by emotion analysis in the text sentence. In this sentiment analysis, for example, the text sentence of the inquiry in the FAQ data is morphologically analyzed, and the negative / positive analysis (whether the sentence has a negative meaning or not) shows whether the word string consisting of the obtained words is negative or positive. It is done by analyzing whether it has a positive meaning). For example, if there is a keyword such as "marriage" in the text sentence of the inquiry, the emotional expression unit 152 can presume that it is "joy", and in this case, the data of one word of the emotional expression of "joy". Is added as an emotion parameter.

Similarly, if there is a keyword of "sadness", the emotional expression unit 152 adds data of one word of emotional expression of "sadness", and if there is a keyword of "trouble", one of the emotional expressions of "trouble". Add word data, and if there is a keyword of "anger", add one word data of emotional expression of "anger". As for the sentiment analysis, not only negative or positive negative / positive analysis for the above-mentioned word sequence but also various kinds are proposed. The emotional analysis in the emotional expression unit 152 may be performed in any way. In addition, rigorous sentiment analysis requires human assistance. The sentiment analysis in the sentiment expression unit 152 may be supported by the call staff or the operator of the BPO center.

The learning data providing unit 153 converts the wording according to the character by the expression conversion unit 151, and sends the FAQ data including the answer to which the emotion parameter is added by the emotion expression unit 152 to the answer generation engine 13 as learning data, and answers. The generation engine 13 is trained.

FIG. 8 is an example of FAQ data in which the answers to the inquiries of the call staff are accumulated as a history. As shown in Fig. 8, for example, in response to the question "Please tell me the procedure when you get married", the frequently asked questions such as "Please contact the Citizen's Division of each city for marriage notification etc." The answer is stored as FAQ data in the database of the training data generation system 14 as training data. Since this FAQ data is intended for the call staff 22, it is generally used in a formal language as shown in FIG.

FIG. 9 is an example of learning data sent from the learning data generation system 14 to the response generation engine 13. As shown in FIG. 9, the answer in the FAQ data, which is a formal wording, is converted into a phrase matching the image of the character icon 101 by the expression conversion unit 151. As a result, the expression conversion unit 151 can display the FAQ data of the formal expression, for example, "Please contact the Citizen's Division of each city for marriage notification, etc." as shown in Fig. 9, "Marriage notification, etc. of each city. Please contact the Citizen's Division, etc. ", and convert it to a phrase that matches the image of the character of the character icon 101.

Further, the emotion expression unit 152 adds an emotion parameter corresponding to the user's emotion in the inquiry to the response of the FAQ data. As a result, as shown in FIG. 9, for example, in the question "Please tell me the procedure when getting married", the emotion of "joy" is estimated from the keyword such as "marriage", and the emotion parameter of "joy" is set. Will be added. The FAQ data, which is converted into a phrase matching the character icon 101 and has an emotional parameter added, is sent to the answer generation engine 13 as learning data.
That is, the learning data generation system 14 extracts words and the like from the text sentence of the inquiry in the FAQ data, and adds a question data indicating the question content and an emotion parameter indicating the emotion of the customer who asked the question to the answer generation engine 13. It is used as learning data for. Here, the learning data generation system 14 converts the text sentence of the answer to the inquiry in the learning data into a phrase corresponding to the image of the character, and adds a word corresponding to the emotion parameter to this text. , It is used as a derivation result which is an answer output by the answer generation engine 13.

FIG. 10 is an example of the answer of the answer generation engine 13 to the question from the user 21. As shown in FIG. 9, the answer generation engine 13 learns the FAQ data to which the emotion parameter is added, which is converted into the wording according to the character icon 101, as the learning data. Further, the answer generation engine 13 adds a word as shown in FIG. 4 to the emotion parameter. Therefore, the answer of the answer generation engine 13 to the question from the user 21 is shown in FIG.

For example, when a question such as "Please tell me the procedure when you get married" is asked from the user terminal 11, the answer generation engine 13, as shown in FIG. 9, "The marriage registration etc. is the citizen's section of each city etc." Search for the answer "please contact us" and the emotional parameter of "joy". Since the emotion parameter of "joy" was searched, the answer generation engine 13 derived the word "Wow, congratulations" corresponding to the user's emotion from the additional word shown in FIG. 4, and this one. After answering the words, the text text of the answer with the phrase corresponding to the image of the character "Please contact the Citizen's Division of each city for marriage registration etc." shown in Fig. 9 is output as the derivation result.

As a result, as shown in FIG. 10, the answer generation engine 13 extracts emotional parameters from the text sentence of this question in response to a question such as "Please tell me the procedure when getting married" from the user terminal 11. However, by using this emotional parameter as input data together with the word indicating the question content for other questions, the character's wording and emotions such as "Wow, congratulations. Please contact the Citizen's Division of each city for marriage notification etc." An answer with a word corresponding to the parameter is derived, and the answer as the derivation result is returned to the user terminal 11 via the user interface server 12.

As described above, in the question-and-answer system 1 according to the first embodiment of the present invention, when the answer generation engine 13 is made to perform learning, the FAQ data stored in the database of the learning data generation system 14 is dealt with. And the emotional parameters extracted from the text of the question are added. When the FAQ data to which such an emotion parameter is added and the wording corresponding to the character and the word corresponding to the above emotion parameter are added to the text sentence of the answer is learned as learning data, the answer generation engine 13 responds to the emotion parameter. Then, you will be able to answer by adding a word that understands the customer's emotions. As a result, when the answer generation engine 13 corresponds to the user 21, a word indicating "empathy" can be added according to the "emotion" of the other party, and the customer satisfaction with the answer corresponding to the inquiry is improved. be able to.

Further, in the question / answer system 1 according to the first embodiment of the present invention, the answer generation engine 13 sends a message of concern at the end of the call. Customers are more likely to remember the last word most impressively, and when they send a caring message at the end of the call, they recognize that the virtual character has "empathy" for the customer's emotions. This creates an intimacy effect and can improve customer satisfaction with the virtual character's answer to the question.

<Second embodiment>
Next, a second embodiment of the present invention will be described. In the first embodiment described above, in the learning data generation system 14, the user's emotion is estimated from the text sentence of the question in the FAQ data, and the wording and the customer are used as emotion parameters for the text sentence of the answer in the FAQ data. By adding the information corresponding to the emotion of the user, the answer generation engine 13 can respond according to the emotion of the user.
On the other hand, in this second embodiment, when the wording indicating the emotion parameter and the information corresponding to the customer's emotion are added, the information indicating a plurality of emotions different from the customer's emotion is added corresponding to each different emotion. Then, it is used as training data. For example, the learning data generation system 14 indicates emotional expressions such as "sadness", "trouble", and "anger" that are different from this emotion when the user's emotion is "joy", which is estimated from the user's inquiry. Answers with information added are also used as learning data. Then, the learning data generation system 14 learns an emotional expression corresponding to the emotion estimated from the user's inquiry as a recommended answer to this inquiry.
Further, the user interface server 12 provides a user interface capable of inputting emotions to the user terminal 11 so that the user 21 can directly transmit the emotion information of the user 21 in the inquiry to the response generation engine 13. Other configurations are the same as those in the first embodiment described above.

FIG. 11 shows a dialogue screen 200 in the question-and-answer system according to the second embodiment of the present invention. As shown in FIG. 11, a character icon 101 is displayed in the center of the dialogue screen 200. A text representing guidance or the like is displayed in the balloon 102 of the character icon 101. Further, the dialogue screen 200 is provided with an input field 103. These configurations are the same as the dialogue screen 100 in the first embodiment shown in FIG.

In the present embodiment, an emotion selection field 210 is further provided below the dialogue screen 200. In the emotion selection column 210, icons 211a, 211b, 211c, and 211d representing the emotions of "joy", "sadness", "trouble", and "anger" are arranged, respectively. These icons 211a, 211b, 211c, 211d provide a user interface for inputting emotions. When the user 21 inputs the text of the question in the input field 103, the user 21 selects an icon close to his / her emotion from the icons 211a to 211d arranged in the emotion selection field 210. Then, the user terminal 11 transmits the information indicated by the icon selected by the user 21 from the icons 211a to 211d to the user interface server 21 as information indicating the emotion of the user 21. As a result, the user interface server 12 transmits the information indicating the emotion of the user 21 transmitted from the user terminal 11 to the answer generation engine 13 together with the text sentence of the inquiry question.

As described above, in the present embodiment, the user 21 can send the emotion at that time to the answer generation engine 13 together with the question. The answer generation engine 13 can acquire the emotional information of the user 21 based on the information of the selected icons 211a to 211d. When answering the question of the inquiry from the user 21, the answer generation engine 13 derives an answer from the database to which a word of emotion corresponding to the icon selected by the user 21 is added as an answer to the inquiry. return it. On the other hand, if the user 21 does not select any of the icons at this time, the response generation engine 13 derives a recommended answer, that is, derives an answer corresponding to the emotion of the user 21 extracted from the text sentence of the inquiry.

That is, it is assumed that the question from the user 21 is "Please tell me the procedure when getting married", and the icon selected at that time is the icon 211d indicating "anger". In this case, the answer generation engine 13 adds the word "I'm sorry" in response to anger instead of the word "Wow, congratulations" that shares "joy", and "marriage registration etc. is for each city. Please contact the Citizen's Division, etc. " Further, the answer generation engine 13 determines that the emotion for the question "Please tell me the procedure when getting married" is not "joy" but "anger" by the icon 211d selected by the user 21.

On the other hand, it is assumed that the question from the user 21 is "Please tell me the procedure when getting married", and the icon selected at that time is the icon 211c indicating "trouble". In this case, the answer generation engine 13 adds the word "it looks like you are in trouble" to share "trouble", and gives the answer "please contact the citizens' section of each city for marriage registration etc." return. In addition, the answer generation engine 13 does not know how to proceed with the question "Please tell me the procedure when you get married" because the icon 211c is selected for the user 21. Determine that you are in trouble.

As described above, in the present embodiment, the response generation engine 13 directly acquires the emotions of the user 21 and adds a word of emotional expression that understands the emotions of the customer. As a result, the user 21 can obtain an answer to which a word corresponding to his / her emotion is added, and can further improve the customer satisfaction with the answer to the question as compared with the first embodiment. .. That is, in the present embodiment, the answer generation engine 13 directly acquires the emotion of the user 21 from the emotional state indicated by the icon selected by the user 21, so that the emotion for the question can be determined with high accuracy, and the question can be answered. Customer satisfaction with the answer can be improved.

Further, in the second embodiment, the user interface server 12 transmits information indicating the emotion selected by the user 21 with the icon to the answer generation engine 13 together with the text sentence of the inquiry question, but the user interface. The server 12 may be configured to send only the text text of the inquiry question to the answer generation engine 13. In the case of this configuration, the answer generation engine 13 provides a recommended answer to which a word of emotion extracted from the text sentence of the question is added, and answers of emotional expressions corresponding to other emotions to the user interface server 12. Send. Then, the user interface server 12 selects an answer to which a word of emotional expression corresponding to the emotion selected by the user 21 by the icon is added from a plurality of answers to the inquiry, and transmits the answer to the user terminal 11.

All or part of the question-and-answer system 1 in the above-described embodiment may be realized by a computer. In that case, a program for realizing this function may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read by a computer system and executed. The term "computer system" as used herein includes hardware such as an OS and peripheral devices. Further, the "computer-readable recording medium" refers to a portable medium such as a flexible disk, a magneto-optical disk, a ROM, or a CD-ROM, and a storage device such as a hard disk built in a computer system.

Further, a "computer-readable recording medium" is a communication line for transmitting a program via a network such as the Internet or a communication line such as a telephone line, and dynamically holds the program for a short period of time. It may also include a program that holds a program for a certain period of time, such as a volatile memory inside a computer system that is a server or a client in that case. Further, the above program may be for realizing a part of the above-mentioned functions, and may be further realized for realizing the above-mentioned functions in combination with a program already recorded in the computer system. It may be realized by using a programmable logic device such as FPGA.

Although the embodiments of the present invention have been described in detail with reference to the drawings, the specific configuration is not limited to this embodiment, and includes designs and the like within a range that does not deviate from the gist of the present invention.

11: User terminal, 12: User interface server, 13: Answer generation engine, 14: Learning data generation system

Claims (5)

A question-and-answer system including a user interface server that provides an interactive user interface and a learning data generation system.
The learning data generation system
An emotional expression unit that generates learning data by adding information on emotional expressions corresponding to the user's emotions estimated from the inquiries to the information of the answers to the inquiries.
It is equipped with a learning data providing unit that sends the generated learning data to an answer generation engine that derives an answer to an inquiry.
The user interface server
An input section for inputting inquiries from users, and
An output unit that outputs the input inquiry to the answer generation engine,
A receiving unit that receives from the answer generation engine an answer to which information of an emotional expression corresponding to the user's emotion estimated from the user's question content, which is derived by the answer generation engine in response to the inquiry, is added.
It is provided with a transmitter for transmitting the answer received from the answer generation engine to the user.
The emotional expression part
Not only the answer to which the emotional expression corresponding to the emotion estimated from the information of the question in the inquiry is added, but also the answer to which the emotional expression corresponding to the emotion different from the emotion is added is generated. A question-and-answer system that generates answers with different emotional expressions as the learning data, and recommends answers to which emotional expressions corresponding to emotions estimated from the information of the question in the inquiry are added. The user interface provided by the user interface server includes an interface capable of inputting emotional information.
The user interface server
The user's emotional information input from the interface is input together with the inquiry.
The question-and-answer system according to claim 1 , wherein the answer corresponding to the inquiry of the user and the emotional information derived by the answer generation engine is output to the user. The question-and-answer system according to claim 1 or 2 , wherein the emotional expression unit generates the learning data by using an environmental parameter indicating an environmental condition including at least a date and time, a season, and a weather. One of claims 1 to 3 , wherein the emotional expression unit generates learning data including a character icon corresponding to the answer in addition to the answer to which the emotional expression information is added. The question and answer system described in. It is a question-and-answer method including a user interface server that provides an interactive user interface and a learning data generation system.
In the learning data generation system
An emotional expression process in which the emotional expression unit generates learning data in which information on emotional expression corresponding to the user's emotion estimated from the inquiry is added to the information of the answer to the inquiry.
A learning data providing process in which the learning data providing unit transmits the generated learning data to an answer generation engine that derives an answer to an inquiry.
In the user interface server
The input process for inputting inquiries from users and
The output process that outputs the input inquiry to the answer generation engine, and
The reception process of receiving from the answer generation engine an answer to which the information of the emotion expression corresponding to the user's emotion estimated from the user's question content, which is derived by the answer generation engine in response to the inquiry, is added.
Including a transmission process unit that transmits an answer received from the answer generation engine to the user.
In the emotional expression process,
Not only the answer to which the emotional expression corresponding to the emotion estimated from the information of the question in the inquiry is added, but also the answer to which the emotional expression corresponding to the emotion different from the emotion is added is generated. A question including that an answer to which the emotional expression is added is generated as the learning data, and an answer to which the emotional expression corresponding to the emotion estimated from the information of the question in the inquiry is added is a recommended answer. Answer method.

Images