JP2021157419A - Interactive business support system and interactive business support method - Google Patents

Abstract

To enable a person in charge of a business to produce a speech with an optimal content when he/she produces the speech responding to a speech of a customer in the business.SOLUTION: In an interactive business support system, voice of a customer is input from a customer terminal, voice of a person in charge is received from a person-in-charge terminal to be transmitted to a voice recognition server and a feeling analysis server. The voice recognition server converts the received voice into speech text, transmits the speech text to an interactive business support device, and the feeling analysis server calculates feeling scores representing feelings of the customer for every speech from the received voice to be transmitted to the interactive business support device. The interactive business support device calculates speech candidates, and prediction scores for every affirmation, negation, subject conversion as a response pattern of the person in charge to the speech of the customer by inference of machine learning for considering data of interaction between the customer and the person in charge with high skill in the past as teacher data on the basis of the speech text and the feeling scores, and displays the calculated speech candidates and the calculated prediction scores on the person-in-charge terminal.SELECTED DRAWING: Figure 8

Description

The present invention relates to an interactive business support system and an interactive business support method, and is particularly suitable for uttering the optimum content when a business person makes an utterance in response to a customer's utterance during business. Regarding interactive business support system and interactive business support method.

In the current customer service business, it is common for the person in charge of business (hereinafter, simply referred to as "person in charge") and the customer to communicate with each other to build a relationship with the customer and deal with troubles. In such a case, a person who is skilled in business or a person who has excellent communication ability is sufficient, but in the case of a person who cannot communicate well or a person who lacks skills, in dealing with the customer, the customer speaks. On the other hand, if appropriate measures cannot be taken and customer satisfaction declines, sales may decrease due to lost sales opportunities and complaints may occur.

In such an interactive business in which a customer and a person in charge have a dialogue, a system is known that supports the correspondence of the person in charge by an information processing device.

For example, Patent Document 1 discloses a dialogue scenario generator that generates answer candidates for a question sentence based on information on a ranking site and generates a dialogue scenario based on the question sentence and the answer.

Japanese Unexamined Patent Publication No. 2018-180936

In general, in the conventional technology, in order to prevent a decrease in customer satisfaction, there is an existing system that supports customer response by visualizing the emotional state of the customer by analyzing the voice generated by the customer and extracting the response sentence by the pattern. ..

However, even if the emotional state of the customer is visualized, the problem is that the person in charge does not know what to say next. Further, according to a technique such as Patent Document 1, in extracting a response sentence by a pattern, it is a problem that a positive response or a response can be made only according to a scenario.

An object of the present invention is to provide an interactive business support system and an interactive business support method capable of making an utterance having an optimum content when a business person makes an utterance in response to a customer's utterance during business. It is in.

The interactive business support system according to the present invention is preferably an interactive business support system that supports business by presenting an utterance when a person in charge of business responds to a customer's utterance, and is a voice from the customer. The customer terminal that inputs the voice, the person in charge terminal that inputs the voice from the person in charge and displays the speech candidate of the person in charge, the voice of the customer and the voice of the person in charge are received, converted into the voice text, and output. The recognition server receives the voice of the customer and the voice of the person in charge, converts it into an emotion score, and outputs it. The emotion analysis server, the voice output by the voice recognition server, and the emotion score output by the voice recognition server are input. Then, it has an interactive business support device that generates and displays a speech candidate of the person in charge, and the interactive business support device is a speech candidate for each classification of the response pattern of the person in charge to the customer's speech. , The predicted score, which is the evaluation value of the speech candidate, is displayed.
The present invention is also grasped as an interactive business support method.

According to the present invention, it is possible to provide an interactive business support system and an interactive business support method capable of making an utterance having an optimum content when a business person makes an utterance in response to a customer's utterance during business. Can be done.

It is a figure which shows the structure and data flow of an interactive business support system. It is a functional block diagram of an interactive business support device. It is a hardware / software configuration diagram of an interactive business support device. It is a figure which shows an example of a topic extraction model. It is a figure which shows an example of a scenario determination model. It is a figure which shows an example of the interest degree determination model. It is a figure which shows an example of the utterance candidate generation model. It is a figure which shows an example of the customer reaction prediction model. It is a figure which shows an example of the present state understanding information table. It is a figure which shows an example of a candidate generation table. It is a flowchart explaining a series of processing performed by an interactive business support apparatus. It is a figure which shows an example of the person in charge utterance support screen.

Hereinafter, each embodiment of the present invention will be described with reference to FIGS. 1 to 8.

First, the configuration of the interactive business support system will be described with reference to FIGS. 1 and 2.
In the interactive business support system of the present embodiment, when a customer conducts business while interacting with a person in charge, the person in charge indicates a candidate for utterance that is a response to the customer's utterance to provide business support. It is something to do.

In the interactive business support system of the present embodiment, as shown in FIG. 1, a customer terminal 10, a person in charge terminal 20, a communication server 40, an interactive business support device 100, a voice recognition server 60, and a sentiment analysis server 80 are networked. It is a form connected by.

The customer terminal 10 is a terminal for a customer to input voice and have a dialogue with a person in charge. The person in charge terminal 20 is a terminal for the person in charge to input voice and have a dialogue with a customer for business. The communication server 40 is a server for communicating voice with each other between the customer terminal 10 and the person in charge terminal 20. Here, the customer terminal 10 and the person in charge terminal 20 may be a telephone terminal, the communication server 40 may be a CTI server via an exchange, and the customer terminal 10 and the person in charge terminal 20 may be a PC. In an information processing device such as (Personal Computer), the customer and the person in charge may have a dialogue by means such as voice chat and TV telephone.

The voice recognition server 60 is a server that performs voice recognition on the customer's voice input from the customer terminal 10 and the voice of the person in charge input from the person in charge terminal 20, and generates a text. The voice recognition server 60 performs voice recognition on the voices of utterances transmitted from the customer terminal 10 and the person in charge terminal 20, converts each utterance into text in a format that can identify each utterance, and performs interactive work. It is transmitted to the support device 100. Here, the utterance is a unit that divides the call according to a certain existing state, and the voice recognition server 60 handles it as a unit for voice recognition.

The emotion analysis server 80 is a server that performs emotion analysis and calculates an emotion score from the feature amount (inflection, loudness, etc.) of the customer's voice input from the customer terminal 10. The emotion score is transmitted to the interactive business support device 100. The emotion score will be described in detail later.

Based on an existing model (details will be described later), the interactive business support device 100 seeks candidates to be spoken by the person in charge and additional information thereof in response to the customer's utterance from the dialogue between the customer and the person in charge. It is a device that displays them to the person in charge.

Next, the functional configuration of the interactive business support device 100 will be described.
As shown in FIG. 2, the interactive business support device 100 includes a topic extraction unit 110, a scenario determination unit 120, an interest level determination unit 130, an utterance candidate generation unit 140, a customer reaction prediction side unit 150, and a display output unit 160. It is composed of each component of the storage unit 170.

The topic extraction unit 110 is a functional unit that extracts the topic of the story from the input dialogue between the customer and the person in charge. The scenario determination unit 120 is a function unit that determines a scenario (daily conversation, business negotiation, customer situation, etc.) that is a major classification of utterances. The interest level determination unit 130 is a functional unit that determines the degree of interest in the content of the dialogue between the customer and the person in charge. The utterance candidate generation unit 140 is a functional unit that generates utterance candidates for which the person in charge should respond to the customer's utterance. The customer reaction prediction side unit 150 is a functional unit that quantifies and obtains the customer's reaction as a prediction score for the utterance candidate. The display output unit 160 is a functional unit that outputs and displays the utterance candidate and the predicted score that is the evaluation value of the utterance candidate to the person in charge. The storage unit 170 is a functional unit that stores data used in the interactive business support device.

As shown in FIG. 2, the storage unit 170 includes a topic extraction model 200, a scenario determination model 201, an interest degree determination model 202, an utterance candidate generation model 203, a customer reaction prediction model 204, a current status understanding information table 300, and a candidate generation. Table 301 is held. The details of each model and table will be described later.

Next, the hardware / software configuration of the interactive business support device will be described with reference to FIG.
The hardware configuration of the interactive business support device 100 is realized by, for example, a general information processing device such as the personal computer shown in FIG.

The interactive business support device 100 includes a CPU (Central Processing Unit) 402, a main storage device 404, a network I / F (InterFace) 406, a display I / F 408, an input / output I / F 410, and an auxiliary storage I / F 412 by bus. It is in a combined form.

The CPU 402 controls each part of the interactive business support device 100, and loads and executes a program required for the main storage device 404.

The main storage device 404 is usually composed of a volatile memory such as a RAM, and stores a program executed by the CPU 402 and data to be referred to.

The network I / F406 is an interface for connecting to a network.

The display I / F 408 is an interface for connecting a display device 420 such as an LCD (Liquid Crystal Display).

The input / output I / F 410 is an interface for connecting an input / output device. In the example of FIG. 3, the keyboard 430 and the mouse 432 of the pointing device are connected.

The auxiliary storage I / F412 is an interface for connecting an auxiliary storage device such as an HDD (Hard Disk Drive) 450 or an SSD (Solid State Drive).

The HDD 450 has a large storage capacity, and stores a program for executing the present embodiment. The topic extraction program 461, the scenario determination program 462, the interest level determination program 463, the utterance candidate generation program 464, the customer reaction prediction side program 465, and the display output program 466 are installed in the interactive business support device 100.

The topic extraction program 461, the scenario determination program 462, the interest level determination program 463, the utterance candidate generation program 464, the customer reaction prediction side program 465, and the display output program 466 are the topic extraction unit 110, the scenario determination unit 120, and the interest level determination unit, respectively. This is a program that executes the functions of 130, the utterance candidate generation unit 140, the customer reaction prediction side unit 150, and the display output unit 160.

Further, the HDD 450 stores a topic extraction model 200, a scenario determination model 201, an interest degree determination model 202, an utterance candidate generation model 203, a customer reaction prediction model 204, a current status understanding information table 300, and a candidate generation table 301.

Next, the data structure used in the interactive business support device will be described with reference to FIGS. 4A to 6.
The topic extraction model 200 is a model for finding a topic by a machine learning method with teacher data, and as shown in FIG. 4A, the utterance text is composed of the utterance subject, the utterance text, and the teacher data consisting of the topic items. It is a model that associates the correct answer data of the topic. In the utterance subject, either a "customer" or a "person in charge" is stored according to the utterance subject. In the utterance text, the text brought from the utterance history with the customer arranged in chronological order of the person in charge (so-called expert) with high skill in the past is stored. The topic stores the topic labeled for the spoken text.

The utterance subject and the utterance text are the same for the following models.

The scenario judgment model 201 is a model for obtaining a scenario by a machine learning method with teacher data, and as shown in FIG. 4B, the utterance text is composed of the utterance subject, the utterance text, and the teacher data consisting of the scenario items. It is a model that associates the correct answer data of the scenario. As already explained, the scenario is a major classification of utterances such as daily conversation, business negotiations, and customer circumstances.

The interest degree determination model 202 is a model for obtaining an emotion score by a machine learning method with teacher data, and as shown in FIG. 4C, from teacher data consisting of an utterance subject, an utterance text, and an emotion score item, It is a model that associates correct answer data of utterance text and emotion score.

The emotion score is a score that quantifies the emotion of how much the customer is interested in the content of the dialogue with the person in charge, and the numerical value increases when the customer is interested. Further, the emotion score of the teacher data is valid only when the uttering subject is the "customer", and becomes "-" (not applicable) when the uttering subject is the "person in charge".

The utterance candidate generation model 203 is a model for obtaining utterance candidates of the person in charge by a machine learning method with teacher data, and as shown in FIG. 4D, items of utterance subject, utterance text, scenario, and presence / absence of interest level. It is a model that associates the correct answer data of the utterance text with the teacher data consisting of. The presence or absence of interest is set from the value of the emotion score. When the emotion score is above a certain threshold, the presence or absence of interest is considered to be "yes", and when the emotion score is below a certain threshold, the presence or absence of interest is determined. , "None". The degree of interest in the teacher data is valid only when the uttering subject is the "customer", and is "-" (not applicable) when the uttering subject is the "person in charge".

The interactive business support device 100 holds different models for the cases where the classification of the customer's response is "affirmation", "counter-argument", and "topic change".

The customer reaction prediction model 204 is a model for obtaining a prediction score by a machine learning method with teacher data, and as shown in FIG. 4E, the utterance subject, the utterance text, the scenario, the items of the degree of interest, and the utterance. It is a model that associates the correct answer data of the predicted score with the teacher data consisting of candidates.

The predicted score is an evaluation score when the person in charge selects the utterance of the utterance candidate, and the higher the predicted score, the more appropriate the score is evaluated. Further, the predicted score of the teacher data is valid only when the uttering subject is the "person in charge", and becomes "-" (not applicable) when the uttering subject is the "customer".

The interactive business support device 100 holds different models for the cases where the classification of the customer's response is "affirmation", "counter-argument", and "topic change".

The current state understanding information table 300 associates the results obtained by inference with the utterances of the past N cases (N is a positive integer determined by the operation of the system) actually performed so far by the customer and the person in charge. It is a table, and as shown in FIG. 5, it is composed of each item of the utterance subject 300a, the utterance content 300b, the topic 300c, the label 300d, and the presence / absence of interest 300ce.

In the utterance subject 300a, either the "customer" or the "person in charge" of the utterance subject in the dialogue between the customer and the person in charge is stored. The utterance content 300b stores the text of the utterance in the dialogue. The topic 300c stores the value of the topic obtained by inference from the topic extraction model 200 by the machine learning method with teacher data. In the scenario 300d, the value of the scenario obtained by inference from the scenario determination model 201 is stored by the method of machine learning with teacher data. The interest level presence / absence 300 stores the value of the interest level presence / absence obtained by inference from the interest level determination model 202 by the method of machine learning with teacher data.

The candidate generation table 301 is a table that holds the utterance candidates obtained by inference from the utterance candidate generation model 203 and the prediction score obtained by inference from the customer reaction prediction model 204 by the method of machine learning with teacher data. As shown in FIG. 6, it is composed of each item of classification 301a, candidate utterance text 301b, and prediction score 301c. A value of any one of "affirmation", "counter-argument", and "topic change" is stored in the classification 301a. "Affirmation" means that the utterance of the candidate person has a positive meaning to the utterance of the previous customer. "Counter-argument" is the opposite, meaning that the utterance of the candidate person in charge is a counter-argument to the utterance of the previous customer. "Topic change" means that the topic is changed from the previous customer's utterance. The candidate utterance text 301b stores the utterance candidate text obtained by inference from the utterance candidate generation model 203 by a machine learning method with teacher data. In the prediction score 301c, the value of the prediction score obtained by inference from the customer reaction prediction model 204 by the machine learning method with teacher data is stored.

The candidate generation table 301 in FIG. 6 shows one utterance candidate for each of “affirmation”, “counter-argument”, and “topic change”, but there may be a plurality of each, and the value of the predicted score is above a certain threshold value. Large utterance candidates may be retained.

Next, the processing performed by the interactive business support device will be described with reference to FIGS. 7 and 8.

First, the interactive business support device 100 performs the current status understanding process (S100).
In the current situation understanding process, the topic, scenario, and the value of the degree of interest are obtained by inference using each model by the machine learning method with teacher data for the dialogue between the customer and the person in charge so far. Is a process of storing in the current state understanding information table 300 shown in FIG.

The topic, the scenario, and the presence / absence of the degree of interest are obtained by inference in which the utterance text is input to the topic extraction model 200, the scenario determination model 201, and the interest degree determination model 202, respectively.

Next, the interactive business support device 100 performs a candidate generation process (S101).
In the candidate generation process, the data of the record of the immediately preceding N cases (N = 4 in the example shown in FIG. 5) of the current state understanding information table 300 is input to the utterance candidate generation model 203 as a parameter, and the candidate utterance text is input. It is a process to obtain. The candidate utterance text is stored in the candidate generation table 301 shown in FIG. 6 in the corresponding item for each classification.

Next, the interactive business support device 100 performs customer reaction prediction processing (S102).
The customer reaction prediction process is a process of inputting the data of the records of the immediately preceding N current situation understanding information table 300 and the candidate utterance text obtained by inference in S101 into the customer reaction prediction model as parameters to obtain a prediction score. Is. The predicted score is stored in the corresponding item in the candidate generation table 301 shown in FIG. 6 for each classification.

Next, the interactive business support device 100 performs display output processing (S103).
The display output process is a process of displaying and outputting the candidate utterance text obtained by the inference of S101 and S102, the predicted score, and the process associated therewith to the person in charge. For example, it is shown on the display device in FIG. The person in charge speech support screen 500 is displayed.

The person in charge utterance support screen 500 displays and outputs the candidate utterance text obtained by inference, the predicted score, and the processing associated therewith to the person in charge, and supports the customer in charge of the response. It is a screen, and as shown in FIG. 8, it is composed of an utterance candidate display area 410 and a time series score graph display area 420.

The utterance candidate display area 510 is an area for displaying the contents of the candidate generation table 301, that is, the utterances that are candidates for the person in charge for each classification and their predicted scores.

The time-series score graph display area 520 is an area for displaying the time-series score graph 521 that visually shows the flow of dialogue between the customer and the person in charge.

The time-series score graph 521 is a line graph in which the horizontal axis is the flow of time and the vertical axis is the score (emotion score and predicted score). From the current marker 522, the left side 523 shows the flow of the dialogue until just before, and the customer's emotion score is displayed at the corresponding point. On the right side 524 of the current marker 522, the predicted score for the utterance candidate performed by the person in charge is displayed for each "affirmation", "counter-argument", and "topic change".

The person in charge utterance support screen 500 allows the person in charge to visually and easily understand the utterance to be made and the predicted score of the prospect of making the utterance.

As described above, according to this embodiment, specific candidates for utterances to be made by the person in charge can be selected as "affirmation" and "counter-argument" by the method using machine learning using the correspondence of the person in charge of veteran work as teacher data. , The person in charge selects a more appropriate utterance by showing the utterance of the person in charge by showing each classification of the response that is called "topic change" and quantitatively showing the concrete predicted score of the utterance. be able to. That is, it is possible to predict the customer's reaction (positive / negative) to the generated candidate utterance, and to propose a response in which the customer's reaction is most positive or a positive timing for the customer's proposal depending on the situation.

10 ... Customer terminal, 20 ... Person in charge terminal, 40 ... Communication server, 60 ... Voice recognition server, 80 ... Sentiment analysis server, 100 ... Interactive business support device

Claims (7)

It is an interactive business support system that supports business by presenting the utterance when the person in charge of business responds to the customer's utterance.
A customer terminal that inputs voice from a customer and
The person in charge terminal that inputs voice from the person in charge and displays the utterance candidates of the person in charge,
A voice recognition server that receives the voice of the customer and the voice of the person in charge, converts it into spoken text, and outputs it.
An emotion analysis server that receives the voice of the customer and the voice of the person in charge, converts it into an emotion score, and outputs it.
It has an interactive business support device that inputs the voice output by the voice recognition server and the emotion score output by the voice recognition server to generate and display utterance candidates of the person in charge.
The interactive business support device is characterized in that it displays a utterance candidate and a predicted score as an evaluation value of the utterance candidate for each classification of the utterance response pattern of the person in charge to the customer's utterance. Business support system. The interactive business support system according to claim 1, wherein the classification of the response pattern of the utterance of the person in charge to the utterance of the customer is an affirmation, a counterargument, and a topic change to the utterance of the customer. The interactive business support device is
The interactive business support system according to claim 1, wherein the utterance candidate is obtained by machine learning inference based on a model with teacher data of a customer and a person in charge of business. The interactive business support system according to claim 2, wherein the model with teacher data of the person in charge of the customer and the business includes the topic, the scenario, and the degree of interest in the dialogue between the customer and the business. The interactive business support device is
The interactive business support system according to claim 1, wherein the predicted score is obtained by machine learning inference based on a model with teacher data of a customer and a person in charge of business. The interactive business support device is
Display a screen showing utterance candidates and predicted scores for each classification,
The interactive business support system according to claim 1, wherein the screen shows the customer's emotion score for each utterance and the predicted score of the utterance candidate of the person in charge for each classification in chronological order. It is an interactive business support method that supports business by presenting the utterance when the person in charge of business responds to the customer's utterance.
The step that the customer terminal inputs voice from the customer,
The step that the person in charge terminal inputs voice from the person in charge,
A step in which the voice recognition server receives the voice of the customer and the voice of the person in charge, converts it into utterance text, and outputs it.
The step that the sentiment analysis server receives the voice of the customer and the voice of the person in charge, converts it into an emotion score, and outputs it, and the interactive work support device are based on a model with teacher data of the customer and the person in charge of work. The step of obtaining the speech candidate of the person in charge by the reasoning of machine learning,
A step in which the interactive work support device obtains a predicted score as an evaluation value of the utterance candidate by machine learning inference based on a model with teacher data of a customer and a person in charge of work.
As a classification of the response pattern of the utterance of the person in charge to the utterance of the customer, the dialogue is characterized by having a utterance candidate and a step of displaying the predicted score for each affirmation, counterargument, and topic change to the customer's utterance. Type business support method.

Images