JP7666013B2 - Processing device, processing method, and program - Google Patents

Description

The present invention relates to a processing device, a processing method, and a program.

There are chatbots (automatic conversation programs using text) that use artificial intelligence (AI) (see Patent Document 1).

International Publication No. 2019/186678

However, there is no chat or chatback (chat version of callback) that uses accumulated data to find a solution from a multi-layered neural network (deep neural network) that uses neural networks or deep learning, and works backwards from this already determined solution to find a solution. Hereinafter, "working backwards from an already determined solution to find a solution" will be referred to as "solving using the reverse direction."

The present invention was made in consideration of the above problems, and provides a processing device, processing method, and program that can derive a more appropriate solution.

In order to solve the above problem, the processing device of the present invention comprises a first processing unit that constructs a scenario consisting of a plurality of response stages and at least one or more first pieces of information used for responding at each response stage by machine learning past responses, a second processing unit that responds with a solution according to the scenario, and a third processing unit that responds with another solution by tracing backward through the response stages from a state in which a solution has already been determined to reaching the solution , and is characterized in that the third processing unit changes the scenario by linking two pieces of first information at different response stages, and responds with another solution according to the changed scenario.
In addition, the processing device of the present invention comprises a first processing unit that constructs a scenario consisting of a plurality of response steps and at least one or more first pieces of information used for responding at each response step by machine learning past responses, a second processing unit that responds with a solution according to the scenario, and a third processing unit that responds with another solution by tracing backward through the response steps from a state in which a solution has already been determined to arriving at the solution, and is characterized in that when it is determined that a solution cannot be obtained using the scenario, the third processing unit responds with a solution different from the solution derived by the scenario.

Moreover, the processing method of the present invention is a processing method in a processing device, the processing device having a first step of constructing a scenario consisting of a plurality of response stages and at least one or more first information used for responding at each response stage by machine learning past responses, a second step of responding with a solution according to the scenario, and a third step of responding with another solution by tracing backward through the response stages from a state in which a solution has already been determined to reaching the solution , the third step being characterized in that the scenario is changed by linking two pieces of first information at different response stages, and another solution is responded with according to the changed scenario.
Furthermore, the processing method of the present invention is a processing method in a processing device, the processing device having a first step of constructing a scenario consisting of a plurality of response steps and at least one or more first pieces of information used for responding at each response step by machine learning past responses, a second step of responding with a solution according to the scenario, and a third step of responding with another solution by tracing backward through the response steps from a state in which a solution has already been determined to arriving at the solution, characterized in that in the third step, if it is determined that a solution cannot be obtained using the scenario, the processing device responds with a solution different from the solution derived by the scenario.

In addition, the program of the present invention causes a computer to function as a first processing unit that constructs a scenario consisting of a plurality of response stages and at least one or more first pieces of information used for responding at each response stage by machine learning past responses, a second processing unit that responds with a solution according to the scenario, and a third processing unit that responds with another solution by tracing backward through the response stages from a state in which a solution has already been determined to reaching the solution , and the third processing unit changes the scenario by linking two pieces of first information at different response stages, and responds with another solution according to the changed scenario.
In addition, the program of the present invention causes a computer to function as a first processing unit that constructs a scenario consisting of a plurality of response steps and at least one or more first pieces of information used for responding at each response step by machine learning past responses, a second processing unit that responds with a solution according to the scenario, and a third processing unit that responds with another solution by tracing backward through the response steps from a state in which a solution has already been determined to arriving at the solution, and is characterized in that when it is determined that a solution cannot be obtained using the scenario, the third processing unit responds with a solution different from the solution derived by the scenario.

The present invention makes it possible to derive a more appropriate solution.

1 is a schematic configuration diagram of a processing apparatus according to an embodiment of the present invention; FIG. 2 is an image diagram of a scenario used in the processing device according to the embodiment of the present invention. 4 is a data configuration example of customer information stored in the processing device according to the embodiment of the present invention. 4 is a data configuration example of response history information held by the processing device according to the embodiment of the present invention. This is a diagram showing the flow (1) of a scenario (guidance) using "solving using backwards" constructed using a neural network (deep neural network). This is a diagram showing the flow (2) of a scenario (guidance) using "solving using backwards" constructed using a neural network (deep neural network). This is a diagram showing a flow (3) of a scenario (guidance) using a "reverse solution" constructed using a neural network (deep neural network).

Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings as appropriate.
Each figure is merely a schematic illustration to the extent that the present invention can be fully understood. Therefore, the present invention is not limited to only the illustrated examples. In addition, in this embodiment, explanations of configurations that are not directly related to the present invention or well-known configurations may be omitted. In addition, in each figure, common components or similar components are given the same reference numerals, and duplicate explanations thereof will be omitted.

<Configuration of response system including processing device according to embodiment>
A response system 1 including a processing device 2 according to an embodiment will be described with reference to Fig. 1. Fig. 1 is a schematic configuration diagram of the response system 1 including the processing device 2 according to an embodiment.

The response system 1 shown in FIG. 1 is a system that responds to inquiries from customers. The response system 1 is, for example, a contact center system, and the following description is based on the assumption that the response system 1 is used for responding to inquiries about financial products. There are no particular limitations on the field in which the response system 1 can be applied. As will be described in detail later, the response system 1 can respond to customer inquiries in real time, and can also later send a chatback (a chat version of a callback) from the response system 1. The response system 1 can respond both by chat (for example, see reference number 135 in FIG. 1) and by phone (for example, see reference number 130 in FIG. 1). The response system 1 can also specify the corresponding time according to the customer, and specify an operator according to the operator's operating hours.

The response system 1 includes a processing device 2. The processing device 2 is, for example, an application server, and is configured as a computer having a CPU (Central Processing unit), memory, a storage means (storage unit, storage device) such as a hard disk, and a network interface. Various functions of this computer are realized by the CPU executing programs loaded onto the memory.

The processing device 2 mainly includes a processing unit 3 having an AI chatbot 100, a database 200, and a neural network 210. The database 200 stores response history information 220 and customer information 230. The neural network 210 is generated from the information stored in the database 200, for example.
Note that the neural network 210 is an example of the “first processing unit”, and the example of the “first processing unit” is not limited to the neural network 210.
Furthermore, the AI chatbot 100 is an example of a "second processing unit" and a "third processing unit," and examples of the "second processing unit" and the "third processing unit" are not limited to the AI chatbot 100.

The AI chatbot 100 is a function that responds to inquiries from customers by chat. The AI chatbot 100 accepts inquiries by chat and responds based on a scenario derived by, for example, the neural network 210. The neural network 210 may be configured as a multi-layer neural network (deep neural network) using deep learning. The neural network 210 is constructed, for example, by machine learning past responses with customers. Note that a scenario is a flow (plot) in an automatic voice response. A scenario is composed of, for example, multiple pieces of guidance. The guidance may be, for example, the content of an inquiry (question) to a customer or a suggestion to a customer, but is not limited to this. This guidance may have different content depending on the situation.

2 shows an image of a scenario (i.e., transition of guidance) used by the processing device 2. Fig. 2 is an image diagram of a scenario used by the processing device 2. The scenario shown in Fig. 2 is derived by a neural network 210.
In FIG. 2, a square with an English letter inside indicates a guidance, and a straight line connecting the guidance indicates a transition of the guidance. The scenario shown in FIG. 2 has a tree structure branched by guidance and its response result. That is, the scenario is composed of a plurality of response stages and at least one or more guidances used for response at each response stage. The scenario shown in FIG. 2 approaches a final solution (answer or proposal) to the customer's inquiry as it progresses from the left side to the right side of the paper. That is, the guidances "E" to "L" are final solutions (answers or proposals) to the customer's inquiry, and the guidances "A" to "J" are inquiries (questions) for obtaining the final solution (answer or proposal). For example, the AI chatbot 100 proceeds through a question that is the guidance "A", a question that is the guidance "B", a question that is the guidance "C", and finally answers the customer with a proposal that is the guidance "E". In the following, in the scenario shown in FIG. 2, the left side of the paper may be expressed as "upstream" and the right side of the paper may be expressed as "downstream".
Note that the guidance constituting the scenario is an example of the "first information", and the example of the "first information" is not limited to the guidance.

In this embodiment, it is assumed that the AI chatbot 100 or the operator 110 responds to a chat or phone call sent from the customer to the processing device 2 as the first response and conveys a solution (answer or proposal), and then, as the second response, the processing device 2 sends a chat or phone call to the customer to convey a new solution (answer or proposal) different from the solution given in the first response. For example, as the first response, the AI chatbot 100 makes inquiries and proposals in the order of ("A" → "B" → "C" → "E"), and then makes inquiries and proposals in the order of ("E" → "C" → "B" → "A" → "I" → "J" → "K"). In such a situation, for example, an "insurance contract" recommended to the general public is answered with guidance "E", but considering the occupation, age, etc., it is more appropriate to answer the "insurance contract" recommended with guidance "K". In this embodiment, we focus on the processing during chatback (a call from the processing device 2 to the customer), and will explain it on the premise that the first response sent from the customer to the processing device 2 is, in principle, completed.

Next, the customer information 230 stored in the database 200 will be described with reference to Fig. 3. Fig. 3 shows an example of the data structure of the customer information 230 held by the processing device 2.
3 mainly includes, for example, a customer identification field 231, a customer category field 232, a gender field 233, an age field 234, a region field 235, and an occupation field 236. Note that the configuration shown here is merely an example, and other configurations may be used.
The customer identification column 231 stores information that can identify a customer.
The customer classification column 232 stores information that classifies customers based on preset criteria. The customer classification column 232 may store composite information, and in FIG. 3, information such as "rank", anonymous/general/important, etc. is stored.
The gender column 233 stores information regarding the gender of the customer.
The age column 234 stores information regarding the customer's age.
The area column 235 stores information about the area where the customer is located (the area where the customer's activities are based).
The occupation column 236 stores information about the customer's occupation.

The customer information 230 shown in Fig. 3 is used to determine how to respond to a customer. In other words, the processing device 2 manages customer information (e.g., rank, anonymous/general/important, gender, age, occupation, etc.) in the customer information 230, and determines how the chat operator will respond based on this information. In this embodiment, the following response is determined based on information in the customer category column 232, etc. Note that the response methods shown here are merely examples. Details of the response methods will be described later.
-Customer A decides to handle the situation using only the chatbot.
-Customer B decides to handle the situation using "a chatbot and/or an operator."
-Customer C decides to handle the situation using "a chatbot and/or an operator."
-Customer D decides to handle the situation using "chatbot only."
-Customer E decides to have the call handled by an operator only.

Next, referring to FIG. 4, the response history information 220 stored in the database 200 will be described. FIG. 4 shows an example of the data configuration of the response history information 220 held by the processing device 2. The response history information 220 includes the response history between the operator (person) and the customer, and the response history between the AI chatbot 100 and the customer. The response history between the operator (person) and the customer is mainly used as information for generating the neural network 210. On the other hand, the response history between the AI chatbot 100 and the customer is used as basic information for processing during chatback (transmission from the processing device 2 to the customer). Note that FIG. 4 shows a case where customer C has requested chatback from an operator, customer E has requested chatback specifying a time and operator, and there are no particular requests from other customers.

4 mainly includes, for example, a customer identification field 221, a reception time field 222, a response time field 223, a chatback information field 224, and a response content field 225. Note that the configuration shown here is merely an example, and other configurations may be used.
The customer identification column 221 stores information that can identify a customer.
The reception time column 222 stores the time when an inquiry was made by a customer.
The response time column 223 stores the time when a response was made to a customer (or the time when a response will be made), and the like.
The chatback information field 224 stores information related to chatback (e.g., the time of chatback, whether operator assistance is required, and the method of response). For example, customer C is an important customer and has received a request for operator assistance, so the status is "operator assistance required." When the status is "operator assistance required," for example, the AI chatbot 100 requests an operator to respond (see reference numeral 120 in FIG. 1).
The response content column 225 stores the response content between the operator (human) or AI chatbot 100 and the customer.

5 to 7 (and also FIG. 1 to FIG. 4 as appropriate), a characteristic process in the processing device 2 during chatback will be described.
Fig. 5 is a diagram showing a flow (1) of a scenario (guidance) by "solution using backwards" constructed using the neural network 210 (deep neural network). Fig. 6 is a diagram showing a flow (2) of a scenario (guidance) by "solution using backwards" constructed using the neural network 210 (deep neural network). Fig. 7 is a diagram showing a flow (3) of a scenario (guidance) by "solution using backwards" constructed using the neural network 210 (deep neural network). "Solution using backwards" here means "solving by tracing backwards from a state where a solution has already been determined".

In this embodiment, when the AI (processing unit 3) helps the operator solve the problem, the AI solves the connection between the "content of the inquiry" from the "AI" to the "customer" in each of the following "A" to "L" (see Figure 5) and the flow of the scenario (guidance) by working backwards from the state where the solution has already been determined, based on the solution obtained from a multi-layered neural network 210 (deep neural network) using a neural network or deep learning, using already accumulated data (i.e., solving using the reverse direction).

For example, in the state shown in Figure 5, if the content of the inquiry is already in state "E" and the problem can be solved by taking it to "K", the AI (processing unit 3) determines that it is best to guide it to state "K" and guides it accordingly (see the thick and dotted arrows in Figure 5). In this case, the AI chatbot 100 responds to the inquiry by saying that it is best to solve it in the order ("E" → "C" → "B" → "A" → "I" → "J" → "K"). In other words, it returns upstream once and then guides it downstream via another route to respond with a new solution.

Also, as shown in FIG. 6, if the AI (processing unit 3) determines that it is not possible to help the operator resolve the problem using the existing flow, either the operator or the AI (processing unit 3), or both, may provide new guidance links to guide the operator to a solution (see the thick line connecting "C" in FIG. 6 with "I", which is the response step one step upstream). In this case, the operator will respond to the inquiry via the AI chatbot 100, for example, that the best way to resolve the issue is ("E" → "C" → "I" → "J" → "K"). Note that the example given here is a case where "B" and "A" have already been determined to be unnecessary or impossible.

Also, as shown in FIG. 7, when the AI (processing unit 3) determines that a new solution is necessary, either the operator or the AI (processing unit 3), or both, may set a new solution (see "X" in FIG. 7). In this case, the operator responds to an inquiry, for example, via the AI chatbot 100, that it would be best to resolve the issue using another method "X." Note that the example shown here is a case where the scenario (guidance) flow constructed using the current neural network 210 (deep neural network) has already been determined to be unnecessary or unacceptable.

With reference to Figures 1 to 7, an example of the operation during chatback in the processing device 2 is shown. Note that the operation examples shown here are merely examples. Also, it is not necessary to perform all of the operation examples shown here, and it is possible to select one or several of them, for example.

(Operation example 1)
In the case of "Customer A" in the customer information 230 shown in FIG. 3, when it has been decided to handle the case "only with the chatbot" because the "rank" is low and the customer is anonymous, for example, the AI chatbot 100 is made to respond according to the "flow of a scenario (guidance) constructed using the neural network 210 (deep neural network) (see FIG. 2)." In this case, only the AI chatbot 100 responds, and the operator does not respond according to the flow. In addition, guidance and response using a solution using the reverse direction (see FIG. 5) are not performed.

(Operation example 2)
In the case of "Customer B" in the customer information 230 shown in FIG. 3, when it has been decided that the case will be handled by "either or both of a chatbot and an operator" because the "rank" and other information have been determined, for example, the AI chatbot 100 and the operator will be made to respond in the order of "Flow (1) (see FIG. 5) of scenario (guidance) using "solution using backward directions" constructed using the neural network 210 (deep neural network)" → "Flow (2) (see FIG. 6) of scenario (guidance) using "solution using backward directions" constructed using the neural network 210 (deep neural network)".

For example, the problem is that the content of the inquiry is already in the "E" state, and the AI (processing unit 3) has determined that it would be best to guide the customer to the "K" state, and considering the customer's "gender, age, region, and occupation," these match the "gender, age, region, and occupation" assumed in the inquiry content of "K" more than the "gender, age, region, and occupation" assumed in the inquiry content of "E."
In this case, for example, the AI chatbot 100 may first provide guidance and answer using flow (1), then provide guidance and answer using flow (2), and if the problem is still not resolved, an operator may handle the individual chat or phone call.

(Operation example 3)
In the case of "Customer C" in the customer information 230 shown in FIG. 3, where the "rank" and other information have been determined as an "important customer" and therefore it has been decided that the customer should be handled by "one or both of the chatbot and the operator," for example, the AI chatbot 100 and the operator are made to respond in the following order: "Flow (1) of scenario (guidance) by "solution using backward directions" constructed using the neural network 210 (deep neural network) (see FIG. 5)" → "Flow (2) of scenario (guidance) by "solution using backward directions" constructed using the neural network 210 (deep neural network) (see FIG. 6)" → "Flow (3) of scenario (guidance) by "solution using backward directions" constructed using the neural network 210 (deep neural network) (see FIG. 7)."

For example, the problem is that the content of the inquiry is already in the "E" state, and the AI (processing unit 3) has determined that it would be best to guide the customer to the "K" state, and considering the customer's "gender, age, region, and occupation," these match the "gender, age, region, and occupation" assumed in the inquiry content of "K" more than the "gender, age, region, and occupation" assumed in the inquiry content of "E."
In this case, for example, the AI chatbot 100 may first provide guidance and answer using flow (1), then provide guidance and answer using flow (2), and then provide guidance and answer using flow (3). If the problem is still not resolved, an operator may respond with an individual chat or phone call.

(Operation example 4)
In the case of "Customer D" in the customer information 230 shown in FIG. 3, where the "rank" and other information have been determined and it has been decided to handle the case "only with the chatbot," for example, the AI chatbot 100 is made to handle the case with "flow (1) (see FIG. 5) of a scenario (guidance) using "solution using backwards" constructed using the neural network 210 (deep neural network)."

For example, the problem is that the content of the inquiry is already in the "E" state, and the AI (processing unit 3) has determined that it is best to guide the customer to the "K" state. Considering the customer's "gender, age, region, and occupation," these partially match the "gender, age, region, and occupation" assumed in the inquiry content of "K" rather than the "gender, age, region, and occupation" assumed in the inquiry content of "E."
In this case, for example, the AI chatbot 100 may first provide guidance and provide a response using flow (1), and if that still does not resolve the issue, an operator may provide an individual chat response.

(Operation example 5)
Like "Customer E" in the customer information 230 shown in FIG. 3, when it is decided that the designated "operator alone" will handle the case because the time and operator to handle the case have been designated, for example, the operator is made to handle the case using "flow (1) (see FIG. 5) of a scenario (guidance) using "solution using backward directions" constructed using the neural network 210 (deep neural network)."
In this case, for example, an operator may first provide guidance and a response according to flow (1), and if that does not resolve the issue, the operator may respond to the customer's individual chat or phone call.

As described above, the processing device 2 according to the embodiment solves the problem by working backwards from the state where the solution has already been determined, based on the solution obtained from the multi-layered neural network 210 (deep neural network) using a neural network or deep learning (i.e., solving using the backwards direction). For example, in the state shown in FIG. 5, if the problem is that the content of the inquiry is already in state "E" and can be solved by taking it to "K", it is determined that it is best to guide it to state "K" and guides it accordingly (see the thick and dotted arrows in FIG. 5). Therefore, the processing device 2 can derive a more appropriate solution.

In addition, the processing device 2 according to the embodiment responds by changing its operation based on the customer information 230 shown in FIG. 3. This makes it possible to derive a more appropriate solution for each customer.

[Modification]
Although the embodiment of the present invention has been described above, the present invention is not limited to this, and can be practiced within the scope of the claims.

Reference Signs List 1 Response system 2 Processing device 3 Processing unit 100 AI chatbot 110 Operator 200 Database 210 Neural network 220 Response history information 230 Customer information

Claims (7)

A first processing unit that constructs a scenario including a plurality of response stages and at least one or more first information used for a response in each response stage by machine learning of past responses;
A second processing unit that responds with a solution according to the scenario;
a third processing unit that responds with another solution by tracing the response steps from a state in which a solution has already been determined to a state in which the solution has been reached in a backward direction;
Equipped with
The third processing unit changes the scenario by combining two pieces of the first information in different response stages, and responds with another solution according to the changed scenario.
A processing device comprising: A first processing unit that constructs a scenario including a plurality of response stages and at least one or more first information used for a response in each response stage by machine learning of past responses;
A second processing unit that responds with a solution according to the scenario;
a third processing unit that responds with another solution by tracing the response steps from a state in which a solution has already been determined to a state in which the solution has been reached in a backward direction;
Equipped with
When the third processing unit determines that the problem cannot be solved by the scenario, the third processing unit responds with a solution different from the solution derived by the scenario.
A processing device comprising: The first processing unit constructs the scenario using a neural network.
3. The processing device according to claim 1 or 2 . A processing method in a processing device, comprising:
The processing device includes:
A first step of constructing a scenario including a plurality of response stages and at least one or more first pieces of information used for the response in each response stage by machine learning of past responses;
A second step of responding with a solution according to said scenario.
A third step of responding with another solution by tracing backward the response steps from an already determined solution to the solution in question;
having
In the third step, the scenario is changed by combining the two pieces of first information at different response stages, and another solution is responded according to the changed scenario.
A processing method comprising: A processing method in a processing device, comprising:
The processing device includes:
A first step of constructing a scenario including a plurality of response stages and at least one or more first pieces of information used for the response in each response stage by machine learning of past responses;
A second step of responding with a solution according to said scenario.
A third step of responding with another solution by tracing backward the response steps from an already determined solution to the solution in question;
having
In the third step, if it is determined that the problem cannot be solved by the scenario, a response is made with a solution different from the solution derived by the scenario.
A processing method comprising: Computer,
a first processing unit that constructs a scenario including a plurality of response stages and at least one or more pieces of first information used for the response in each response stage by machine learning of past responses;
a second processing unit that responds with a solution according to the scenario;
a third processing unit that responds with another solution by tracing back through the response steps from a state in which a solution has already been determined to the solution in question;
Function as a
The third processing unit changes the scenario by combining two pieces of the first information in different response stages, and responds with another solution according to the changed scenario.
A program characterized by: Computer,
a first processing unit that constructs a scenario including a plurality of response stages and at least one or more pieces of first information used for a response in each response stage by machine learning of past responses;
a second processing unit that responds with a solution according to the scenario;
a third processing unit that responds with another solution by tracing back through the response steps from a state in which a solution has already been determined to the solution in question;
Function as a
When the third processing unit determines that the problem cannot be solved by the scenario, the third processing unit responds with a solution different from the solution derived by the scenario.
A program characterized by:

Images