Information processing device, information processing method, and program

Abstract

The purpose is to provide an information processing device, an information processing method, and a computer program to support customer service provided by human operators. [Solution] The call management system 1 comprises a call management device 2, a call support AI 3, a caller call device 4, and an operator call device 5. The call management device 2 comprises a caller determination unit that identifies the caller, an operator call unit that manages the call between the caller and the operator, a call determination unit that determines whether the call between the caller and the operator is a predetermined call, a call intervention unit that intervenes in the call between the caller and the operator in response to the determination that the call is a predetermined call, and a call record management unit that records and manages the call history between the caller and the operator.

Description

Technical Field

[0001] The present invention relates to an information processing apparatus, an information processing method, and a program for supporting customer service.

Background Art

[0002] Many companies often install call centers as bases or windows for handling customers. These call centers handle a variety of matters, such as inquiries from customers and consumers, various applications and cancellations, support requests for malfunctions, and claims regarding products and services.

[0003] In recent years, it has been difficult to ensure and cultivate stable operators. The so-called customer harassment has exacerbated and accelerated this situation. Customer harassment hinders the long-term cultivation and stable employment of operators, and causes health damage and stress to operators. This is not a problem that can be solved by a company's sales efforts or management issues, but has become a social problem. Also, even if it cannot be called customer harassment, there are cases of out-of-hours responses where responses beyond normal working hours are required, and cases of disconnection timing where it is difficult to end the call. In order to solve such problems in telephone responses, technologies using computer systems are also known.

[0004] Patent Document 1 describes a system that always connects to the call line of a call center to enable detection and immediate response to customer harassment in the call center. The system includes means for determining a call corresponding to customer harassment from the content of the conversation between the customer and the CSR (Customer Service Representative), and means for issuing a warning to the on-site manager SV. This system is equipped with means for the manager to immediately intervene in calls determined to be customer harassment. It is also described that a function for recording the detected calls and creating detailed reports is provided to reduce the psychological stress of CSRs (Customer Service Representatives) engaged in customer service.

[0005] Patent Document 2 describes a data processing system that replaces supervisors with a super AI to eliminate the need for human operators, eliminates the need to operate centers in multiple countries by utilizing time differences, and enables call center operation without worrying about labor shortages or language barriers. This system includes converting customer inquiry voices into text information using a voice-to-text conversion function, having a generating AI analyze the text information, generating an appropriate response based on the analysis results, converting it back into voice information and providing it to the customer, and having the super AI take on the role of a supervisor who checks the appropriateness of the response generated by the generating AI, judging whether the response from the generating AI is appropriate, and correcting or regenerating the response as necessary.

[0006] Patent Document 3 describes a system that includes a conversational robot incorporating a generation AI, means for stationing the robot at a customer's location as a sales representative for a telecommunications carrier, means for listening to the customer's problems and providing answers in a conversational manner, means for connecting to the telecommunications carrier's center depending on the content of the request in the conversation, means for a call center or a corporate sales representative of the telecommunications carrier to handle matters that the robot cannot handle, and means for pre-configuring the robot's personality to quickly resolve various customer issues. [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2025-053478 [Patent Document 2] Japanese Patent Publication No. 2025-057479 [Patent Document 3] Japanese Patent Publication No. 2025-044945 [Overview of the Initiative] [Problems that the invention aims to solve]

[0008] Generally, it is the responsibility of the manager (supervisor) to handle problematic calls among the inquiries received by a call center. However, securing and training supervisors capable of such handling takes time and money. Furthermore, it is often difficult for managers to monitor all interactions between customers and operators and to ensure appropriate interruptions are made with consistent quality. In addition, while policies and rules are sometimes set up to protect operators, these measures alone are often insufficient to address problematic calls. To improve this situation, it is necessary to implement real-time preventative measures for individual phone calls rather than simply implementing general recurrence prevention measures. On the other hand, there are cases where the content of what the operator says to the customer is problematic and escalates into a complaint, and operator training may be necessary.

[0009] Patent Document 1 describes a system that determines whether a call constitutes customer harassment based on the content of the conversation between the customer and the operator, and allows the administrator to intervene immediately. However, it is not realistic for the administrator to respond immediately to all cases with a consistent level of quality. Patent Document 2 describes a system that eliminates the need for human operators, with a super AI checking the appropriateness of responses generated by the generation AI. However, it cannot address customer harassment in customer service interactions with human operators. Patent Document 3 describes a system where a conversational robot incorporating generation AI handles customer interactions, and matters that the robot cannot handle are handled by call centers or corporate sales representatives of telecommunications carriers. However, it cannot address customer harassment in customer service interactions with human operators.

[0010] This invention has been made in view of the above circumstances, and its purpose is to provide an information processing device, an information processing method, and a program for supporting customer service by human operators. [Means for solving the problem]

[0011] To solve the above problems, a first aspect of the information processing device according to the present disclosure is characterized by comprising: a call determination unit that determines whether a call between a caller and an operator is a predetermined call; and a call intervention unit that intervenes in the call between the caller and the operator in response to the determination that the call is a predetermined call.

[0012] A second aspect of the information processing device relating to this disclosure is characterized in that the predetermined call is a call that includes inappropriate call handling.

[0013] A third aspect of the information processing device relating to this disclosure is characterized in that the call determination unit includes a learning model trained to input call response information including the caller's statements, the operator's statements, or the operator's response time, and to output a call response score indicating the degree of inappropriateness of the call response.

[0014] A fourth aspect of the information processing device relating to this disclosure is characterized in that the predetermined call includes a call in which the call response score, which indicates the degree of inappropriateness of the call response, exceeds a predetermined threshold.

[0015] A fifth aspect of the information processing device relating to this disclosure is characterized in that the call intervention unit comprises a large-scale language model trained to output a response corresponding to the caller's statement.

[0016] A sixth aspect of the information processing device relating to this disclosure further comprises a caller determination unit that determines that the caller is a specific caller, and the caller determination unit activates the call intervention unit in response to determining that the caller is the specific caller.

[0017] A seventh aspect of the information processing device relating to this disclosure is characterized in that the specific caller is determined based on a specific caller score calculated based on the caller's call history. [Effects of the Invention]

[0018] According to the first aspect of the present disclosure, by having the call intervention AI intervene in the customer interaction with the caller, it is possible to prevent conflicts between the caller and the operator, thereby reducing the mental burden on individual operators and improving the working environment. In addition, since the generative AI can handle long-term customer interactions, it is possible to efficiently operate a call center with a small number of staff. Furthermore, since it is possible to create an environment where operators can work comfortably, an improvement in the awareness of society as a whole is expected.

[0019] According to the second aspect of the present disclosure, by monitoring the call between the caller and the operator, detecting inappropriate call responses, and being able to intervene in the call, it is possible to reduce the burden on the operator or the administrator for problematic calls.

[0020] According to the third aspect of the present disclosure, by training a learning model based on past call histories, it becomes possible to make highly accurate call judgments. In addition, since the learning model outputs a call response score indicating the degree of inappropriateness of the call, it is possible to numerically grasp the degree of inappropriateness of the call between the caller and the operator.

[0021] According to the fourth aspect of the present disclosure, it is possible to easily determine the degree of inappropriateness of the call response between the caller and the operator based on the call response score, and it is possible to shorten the time related to call judgment.

[0022] According to the fifth aspect of the present disclosure, by providing a generative AI that is fine-tuned to output a response corresponding to the caller's speech, it becomes possible to provide a highly accurate automatic response.

[0023] According to the sixth aspect of the present disclosure, by determining a caller who is likely to make an inappropriate call when receiving a call from the caller, it is possible to connect the corresponding inappropriate call from the caller to the generative AI before the operator receives the call, thereby further reducing the load on the operator.

[0024] According to the seventh aspect of the present disclosure, by calculating in advance the specific caller score of a caller based on the past call history, the caller can be quickly identified.

Brief Description of Drawings

[0025] [Figure 1] FIG. 1 is an overhead view showing an example of the overall configuration of a call management system 1 to which an information processing apparatus, an information processing method, and a computer program according to a first embodiment of the present disclosure are applied. [Figure 2] FIG. 2 is an operation flowchart showing an example of the outline of the operation of the call management system 1 according to the first embodiment of the present disclosure. [Figure 3] FIG. 3 is a diagram for explaining an example of the functional configuration of a call management apparatus 2 according to the first embodiment of the present disclosure. [Figure 4] FIG. 4 is a diagram showing an example of a caller determination table according to the first embodiment of the present disclosure. [Figure 5] FIG. 5 is a diagram for explaining an example of the learning of a call determination AI 31 according to the first embodiment of the present disclosure. [Figure 6] FIG. 6 is an operation flowchart showing an example of the outline of the operation of a call determination unit 213 according to the first embodiment of the present disclosure. [Figure 7] FIG. 7 is a diagram for explaining an example of the learning of a call intervention AI 32 according to the first embodiment of the present disclosure. [Figure 8] FIG. 8 is an operation flowchart showing an example of the outline of the operation of a call intervention unit 214 according to the first embodiment of the present disclosure. [Figure 9] FIG. 9 is a diagram for explaining an example of the learning of a litigation preparation AI 33 according to the first embodiment of the present disclosure. [Figure 10] FIG. 10 is an operation flowchart showing an example of the outline of the operation of a call record management unit 215 according to the first embodiment of the present disclosure.

Modes for Carrying Out the Invention

[0026] An information processing apparatus, information processing method, and computer program of the first embodiment of this disclosure will be described below with reference to the drawings. In the following description, components and elements that are the same or similar as those already described will be denoted by the same reference numerals and their descriptions will be omitted.

[0027] This disclosure uses telephone calls as an example of customer interaction, but is not limited to this. The interaction described in this disclosure may utilize any communication method that enables the caller and operator to communicate. The information used for the interaction may also be in any format, such as voice, video, or text. For example, the interaction described in this disclosure includes not only calls via telephone lines, but also calls via text messages, video calls, and voice calls. It may also include calls using video chat services that allow video and audio calls over the internet, or SNS calling services that allow voice and text calls.

[0028] In this disclosure, the following terms relating to telephone communication are used: "making a call" means making a call, "receiving a call" means having a call coming in, "answering a call" means answering a call that has come in, "talking" means talking on the telephone, and "hanging up" means hanging up the telephone. The terms making a call, receiving a call, talking, and hanging up also apply to calls made by any other means of communication.

[0029] (1) Call management system 1 (1-1) Overall configuration of call management system 1 Figure 1 is an overview view showing an example of the overall configuration of a call management system 1 to which the information processing device, information processing method, and computer program according to the first embodiment of this disclosure are applied. The call management system 1 shown in Figure 1 comprises a call management device 2, a call support AI 3, a caller call device 4, and an operator call device 5. The call management device 2, the caller call device 4, and the operator call device 5 are connected via a network 6 such as the Internet or a telephone network.

[0030] The call support AI 3 includes a call judgment AI 31, a call intervention AI 32, and a litigation preparation AI 33. The call support AI 3 is communicated with the call management device 2. The call support AI 3 may be located in close proximity to the call management device 2, or it may be communicated with via the network 6. The call judgment AI 31 estimates whether a call is an inappropriate call. The call intervention AI 32 is a generative AI trained to generate a call response with the caller. The litigation preparation AI 33 is a generative AI that generates a draft complaint based on call history, laws, precedents, etc., related to a case that may be subject to litigation when such a case occurs.

[0031] The call management device 2 is an information processing device for supporting customer service between callers and operators. A caller refers to any person who makes a phone call to the call center, and may be, for example, a general consumer or a customer with a specific business relationship. An operator refers to a person (a live operator) who engages in customer service through dialogue at the call center.

[0032] A call center is equipped with an operator communication device 5. The caller's communication device 4, used by the caller, and the operator's communication device 5, used by the operator, are connected in a way that enables communication between the caller and the operator. A call center may be established for a single company or for multiple companies. A call center typically receives calls from callers via telephone lines, but it may also receive calls using voice calls over the internet. Call centers are sometimes also called customer centers.

[0033] The call management device 2 determines whether the call between the caller and the operator is a predetermined call, and intervenes in the call between the caller and the operator in response to the determination that the call is a predetermined call. The call management device 2 monitors the call between the caller and the operator and checks whether the call between them includes a predetermined call. The call between the caller and the operator may be on any topic, such as opinions, comments, inquiries, complaints, consultations, offers, requests for improvement, or grievances.

[0034] In the first embodiment, a specified call is a call that requires intervention in a conversation between a caller and an operator, and is, for example, a call that includes inappropriate call handling (hereinafter also referred to as an "inappropriate call"). In this disclosure, inappropriate call handling means any call that is considered inappropriate as a telephone call according to generally accepted social norms. Examples of inappropriate calls include forcing unreasonable complaints, prolonged verbal abuse and insults, psychological pressure through intimidating words and actions, repeated excessive demands, and unreasonable orders that exceed the scope of work. It also includes any morally reprehensible words and actions or calls that disrupt business operations.

[0035] The call management device 2 can determine whether a call between a caller and an operator constitutes an inappropriate call by, for example, monitoring the statements made by the caller and the operator, as well as the call duration. Furthermore, it can detect inappropriate calls if the operator makes a specific message.

[0036] The call management device 2 can use a call judgment AI 31 to determine whether a call between a caller and an operator is an inappropriate call. The call judgment AI 31 is a learning model trained to take call response information as input and output a call response score indicating the degree of inappropriateness of the call. When call response information between a caller and an operator is input to the call judgment AI 31, it estimates whether the call response information constitutes an inappropriate call and outputs a call response score indicating the degree of inappropriateness of the call response.

[0037] Call response information is information obtained from monitored calls and may include, for example, the caller's statements, the operator's statements, or the operator's response time. Response time includes the time from when the operator receives the call until the caller hangs up. Call response information may also include identification information for the caller and the operator. Caller identification information may include, for example, the caller's phone number or user ID. Operator identification information may include the employee code assigned to the operator.

[0038] In the first embodiment, the call management device 2 is described as acquiring the statements of the caller and operator as voice signals (analog signals) and including them in the call response information, and inputting the call response information to the call judgment AI 31. However, the call management device 2 may also be configured to acquire the statements of the caller and operator as digitized data and input it to the call judgment AI 31. For example, the call management device 2 may, in place of or in conjunction with voice signals (analog signals), include the content of the statements (text data) obtained by speech recognition of the statements of the caller and operator, voice information (digital data) obtained by AD conversion of the statements, voice characteristics such as sound pressure and frequency, or call manners analyzed from the content of the statements in the call response information, and input the call response information to the call judgment AI 31.

[0039] The call management device 2 inputs call handling information between the caller and the operator to the call judgment AI 31 and obtains a call handling score indicating the degree of inappropriateness of the call, which is output by the call judgment AI 31. The call management device 2 determines whether the call between the caller and the operator falls under a predetermined call category by determining whether the obtained call handling score exceeds a predetermined threshold.

[0040] When the call management device 2 detects that the call handling score output by the call judgment AI 31 exceeds a predetermined threshold, it determines that the call between the caller and the operator contains inappropriate call handling and intervenes in the call between the two parties. The call management device 2 can use the call intervention AI 32 to intervene in the call between the caller and the operator.

[0041] The call intervention AI 32 is a large-scale language model trained to output responses corresponding to the caller's statements. The call intervention AI 32 is an AI that does its best to interact with humans until the end, learning from prepared Q&A, conversation order, and interaction history. For example, it is trained to provide explanatory answers to questions and to provide listening items to guide further questions. The call intervention AI 32 can take questions from the caller as input and determine whether it can be answered and what the answer should be. Also, in cases not covered by the response menu, the call intervention AI 32 can handle the call and end it without connecting to a human line. The call intervention AI 32 may be directly connected to the operator call device 5 or connected via the network 6.

[0042] The call management device 2 may further include a caller identification unit that identifies the caller when a call comes in from the caller. The caller identification unit determines whether the caller is a specific caller, and if it is determined that the caller is a specific caller, it can activate the call intervention unit. In other words, the call management device 2 can identify the caller at the stage when a call comes in from the caller, that is, before the operator answers the call from the caller, and if it determines that the caller is a specific caller that has been listed in advance, it can intervene in the call to the caller. As a result, if it is determined that there is a high possibility that the caller will make an inappropriate call, the call intervention AI 32, rather than the operator, can take over the call at the start of the call.

[0043] The call management device 2 may determine whether a caller is a specific caller by checking whether the caller is included in a caller determination table prepared based on the caller's past call history, or by checking whether the specific caller score recorded in the caller determination table exceeds a predetermined score value. The caller determination table is, for example, a list of callers who are likely to make inappropriate calls, extracted from past call history. The specific caller score is a score indicating the likelihood that each caller is a specific caller, and can be quantified and recorded in advance based on past call history. Here, a specific caller is a caller who is likely to make inappropriate calls.

[0044] In this disclosure, a caller identification table is used to determine that a caller is a specific caller. However, instead of a caller identification table, a specific caller may be identified using information that uniquely identifies the caller, such as the caller's telephone number, driver's license number, contract number, user ID, or email address. Alternatively, the caller's voice characteristics may be recorded, and the caller may be identified based on these voice characteristics.

[0045] The caller ID device 4 is a device used by callers when making a call to a call center. The caller ID device 4 can be any device that enables voice communication, and may be a landline telephone, a smartphone, or a PC. It is desirable that the caller ID device 4 has a function to notify the call center of the telephone number.

[0046] The operator call device 5 is a communication device with telephone functionality installed in the call center of a company or store. The operator call device 5 can communicate with the caller call device 4 via a network 6, which includes a telephone network or an internet communication network. The operator call device 5 may be directly connected to the call management device 2 for communication, or it may be connected for communication via the network 6.

[0047] Figure 1 illustrates three call support AIs 3, two caller communication devices 4, and two operator communication devices 5, but the number of these devices can be arbitrary. Furthermore, one device may be composed of multiple information processing devices, or multiple devices may be implemented by a single information processing device.

[0048] (1-2) Overall operation flow of call management system 1 Figure 2 is an operation flow diagram showing an example of the operation overview of the call management system 1. The caller intercom 4 places a call to the call center's telephone number at the caller's request (S310). The caller identification unit 211 of the call management device 2 recognizes the incoming call from the caller and obtains information to identify the caller (S320). The caller identification unit 211 determines whether the caller who made the call is a specific caller recorded in the caller identification table (S321). If the caller is not a specific caller (S321: No), the operator intercom 212 is notified that there is an incoming call from the caller. On the other hand, if the caller is a specific caller (S321: Yes), the call intervention unit 214 is notified that there is an incoming call from the specific caller. The determination in step S321 is performed before receiving the call from the caller, that is, before starting a conversation with the caller. Therefore, incoming calls from specific callers are received directly by the call intervention unit 214 without going through the operator.

[0049] When the operator call unit 212 is notified by the caller identification unit 211 that a call has been received from a caller, it establishes a session between the caller call device 4 and the operator call device 5, and the operator receives the call from the caller (S322). Once the operator receives the call, a conversation between the caller and the operator begins (S311, S323). The operator call unit 212 notifies the caller identification unit 213 of the conversation information, including what the caller and operator have said and the elapsed time since the call was received (hereinafter also referred to as "interaction time" or "call time").

[0050] The call judgment unit 213 inputs the call handling information between the caller and the operator, obtained from the operator call unit 212, into the call judgment AI 31, causing the call judgment AI to estimate the degree of inappropriateness of the call handling. Here, it is assumed that the call judgment AI 31 is trained to output a call handling score of "1" when call handling information corresponding to an inappropriate call is input. The trained call judgment AI 31 estimates the degree of inappropriateness corresponding to the input call handling information and outputs a call handling score between "0" and "1" according to the call handling information.

[0051] The call handling score can be any score that can indicate the degree of inappropriateness. For example, the score range can be arbitrarily set from "0" indicating the lowest degree of inappropriateness to "10" indicating the highest degree of inappropriateness. Alternatively, the call handling score may not be a value within a specific range, but rather a cumulative value obtained by accumulating the degree of inappropriateness over a predetermined period.

[0052] The call determination unit 213 determines that if the call handling score is below a predetermined threshold, the call between the caller and the operator is not an inappropriate call (hereinafter referred to as a "normal call") (S324). The call determination unit 213 determines whether the call has ended based on whether the caller has hung up (S325). If the call has not ended (S325: No), the call determination (S324) is repeated. On the other hand, if the call has ended (S325: Yes), the call history is notified to the call record management unit 215. The predetermined threshold can be arbitrarily set depending on whether or not to strictly determine the degree of inappropriateness, but for example, 0.8 may be set as the predetermined threshold.

[0053] The predetermined threshold is not a fixed value, but may be set to vary depending on the sender's attributes (such as gender and age), the content of the inquiry, the field, and the product. For example, even if the response time is long, the threshold may be raised to more leniently determine the degree of inappropriateness for inquiries about products that require a lot of explanation. Similarly, for senders who are registered members, their gender, age, purchase history, hobbies, etc., can be extracted from their member data, so the degree of inappropriateness may be determined more leniently for fields that the sender is interested in. On the other hand, for inquiries about products that are unlikely to be purchased by the sender, the threshold may be lowered to more strictly determine the degree of inappropriateness.

[0054] The call determination unit 213 determines that a call between the caller and the operator is an inappropriate call if the call handling score exceeds a predetermined threshold. The call determination unit 213 notifies the call intervention unit 214 of the identification information of the operator involved in the inappropriate call or the identification information of the corresponding operator call device 5, and requests that the unit intervene in the caller's call on behalf of the operator (hereinafter this request is also referred to as the "intervention request").

[0055] If the call determination unit 213 determines that the call between the caller and the operator is excessively inappropriate, it may terminate the call with the caller without sending an intervention request to the call intervention unit 214. For example, if the call handling score output by the call determination AI exceeds a threshold (second threshold) that is higher than a predetermined threshold (first threshold), it may send a call termination message to the caller and forcibly disconnect the session between the caller call device 4 and the operator call device 5.

[0056] When the call intervention unit 214 is activated by an intervention request from the caller determination unit 211 or the call determination unit 213, it performs call receiving processing by the call intervention AI (S330). If the caller determination unit 211 makes an intervention request, the caller determination unit 211 has received a call from the caller but has not received it. Therefore, the call intervention unit 214 establishes a session between the caller and the call intervention AI 32, and the call intervention AI 32 receives the call. On the other hand, if the call intervention unit 214 makes an intervention request from the call determination unit 213, since a session has been established between the caller call device 4 and the operator call device 5, the call intervention unit 214 disconnects this session, establishes a session between the caller call device 4 and the call intervention AI 32, and the call intervention AI 32 receives the call.

[0057] The call intervention unit 214 instructs the call intervention AI 32 to answer the call from the caller. The call intervention AI 32 receives the voice signal from the caller and generates a voice response to the caller's statements. The call intervention unit 214 transmits the generated voice response to the caller's call device 4 (S331). The call intervention unit 214 determines whether the call has ended based on whether the caller has hung up (S332). If the call has not ended (S332: No), the call intervention AI 32 repeats the call (S331). On the other hand, if the call has ended (S332: Yes), the call history is notified to the call record management unit 215.

[0058] The call record management unit 215 records and manages identification information that identifies the caller and operator, as well as the call history between the caller and the operator. After a call ends, the call record management unit 215 retrieves the call history between the caller and the operator or between the caller and the call intervention AI 32 from the call determination unit 213 or the call intervention unit 214, and records it as the call history in the call history DB 221. The call record management unit 215 also has the function of creating a draft complaint based on past call history, laws, precedents, etc., by using the litigation preparation AI 33.

[0059] (2) Hardware configuration (2-1)Call management device 2 The call management device 2 comprises a processor 21, memory 22, storage 23, a communication interface (communication I / F) 24, an input / output interface (input / output I / F) 25, and a bus 26 connecting these. The processor 21 controls the operation of the entire call management device 2. The processor 21 may be a general-purpose processor such as a CPU, MPU, or GPU, but is not limited to a general-purpose processor and may be a dedicated processor such as an ASIC or FPGA. The memory 22 is the main memory and includes RAM, etc. The storage 13 is an auxiliary storage device and includes a non-volatile storage device such as a hard disk drive (HDD) or solid-state drive (SSD). The communication interface (communication I / F) 24 is a wired or wireless communication interface and is a module for communicating with devices such as the call support AI 3, caller ID device 4, and operator ID device 5 via a communication network including network 6. Network 6 may be, for example, the internet and may include access networks such as LAN, WAN, mobile communication network, wired telephone network, FTTH, CATV network, etc. The input / output interface (input / output I / F) 25 takes in input data from an external source and outputs output data to an external source.

[0060] (2-2) Call support AI3 The call support AI 3 is an information processing device used by the call management device 2 to determine call responses, intervene in calls, or prepare for litigation. This disclosure includes a learning model or a generative model trained according to each purpose of use. The call support AI 3 in this disclosure includes a call judgment AI 31, a call intervention AI 32, and a litigation preparation AI 33. The call judgment AI 31 includes a learning model trained to take information relating to a call between a caller and an operator as input and output a call response score representing the degree of inappropriateness of the call response. The call intervention AI 32 includes a generative model trained to take a speech signal as input and generate a speech response to that speech. Here, the call intervention AI 32 is described as receiving a speech signal (analog signal) as input, but it may also receive a speech signal that has been digitized, or a speech that has been converted into text data by speech recognition. The litigation preparation AI 33 includes a generative model trained to take a specific caller's call history as input and generate a draft complaint.

[0061] Generative models include, for example, natural language generation models or natural language processing models. Generative models include general-purpose natural language processing learning models such as Large Language Models (LLMs) that have been trained on vast amounts of data. Generative models may also be language models that have been fine-tuned for specific purposes such as customer service or customer interaction. Generative models may also be language models that can handle various tasks without fine-tuning. Generative models are not limited to those described above. Generative models may also be a combination of text generation AI and speech generation AI. Generative models take a command (prompt) about a task as input and generate a response about the task based on the command. Generative models may perform additional learning based on feedback on the generated responses.

[0062] (2-3) Caller's communication device 4 The caller-to-call device 4 is a device equipped with telephone functionality that allows a caller to make a call to a call center and have a voice conversation with an operator via a telephone network or the internet. The caller-to-call device 4 includes, for example, a landline telephone, a smartphone, or a tablet device.

[0063] (2-4) Operator communication device 5 The operator communication device 5 used in a call center includes functions for making calls to and from external lines, a PBX (Private Branch Exchange) function for distributing incoming calls to multiple internal telephones, a call recording function, and a function for managing customer information and call history. However, in this disclosure, it is described as a device that has at least a telephone function for operators to communicate with callers.

[0064] The operator communication device 5 is connected to the call management device 2, and is configured so that the call management device 2 can monitor the call between the caller and the operator. The operator communication device 5 is also configured to perform control to establish and terminate the session between the caller and the operator.

[0065] (3) Functional configuration of the call management device 2 Figure 3 is a diagram illustrating an example of the functional configuration of a call management device 2, which is an information processing device according to the first embodiment of this invention. The call management device 2 of this first embodiment has a functional configuration comprising a call management unit 210 and a storage unit 220, and is communicatively connected to a call support AI 3, a caller call device 4, and an operator call device 5. The call management device 2 is connected to the caller call device 4 via a network 6, which includes a telephone line network. The call management device 2 may be directly connected to the call support AI 3 and the caller call device 4, or it may be connected via the network 6.

[0066] (3-1) Storage section 220 The memory unit 220 includes a call history DB 221, a caller identification DB 222, a learning data DB 223 for the call determination AI, a knowledge DB 224 for the call intervention AI, and a knowledge DB 225 for the litigation preparation AI. The call history DB 221 is a database that stores the past call history of calls made to the call center, including identification information of the caller and operator, statements made by the caller and operator, voice information, call etiquette, call duration, and the activation status of the call intervention AI 32.

[0067] The caller identification DB222 is primarily accessed by the caller identification unit 211 and stores caller identification tables and other data used to determine whether a caller is a specific caller. The call judgment AI learning data DB223 is primarily accessed by the call judgment unit 213 and records the call judgment AI 31 (learning model) and training data for training the call judgment AI 31. The training data is extracted from past call history to show call histories of callers who have made typical inappropriate calls. Alternatively, the training data may also be extracted from past call history to show call histories of callers who have made normal calls. The call intervention AI knowledge DB224 and the litigation preparation AI knowledge DB225 record the call intervention AI 32 (generating AI model), the litigation preparation AI 33 (generating AI model), and knowledge data for fine-tuning these generating AI models, respectively.

[0068] (3-2) Call management department 210 The call management unit 210 includes a caller identification unit 211, an operator call unit 212, a call determination unit 213, a call intervention unit 214, and a call record management unit 215.

[0069] (3-2-1) Caller identification unit 211 The caller identification unit 211 refers to the caller identification table recorded in the caller identification DB 222 to determine whether the caller who made a call to the call center from the caller identification device 4 is a caller who has made an inappropriate call in the past. When a call is coming in, i.e., when the call is received, the caller identification unit 211 determines from the caller's telephone number whether the caller is recorded in the caller identification table. Alternatively, in addition to determining whether the caller is recorded in the caller identification table, it may also determine whether the specific caller score is a predetermined score value.

[0070] If the caller determination unit 211 determines that the caller is not a designated caller, it sends an incoming call notification to the operator call unit 212. On the other hand, if it determines that the caller is a designated caller, it sends an intervention request to the call intervention unit 214. In other words, it initiates a call using the call intervention AI 32 without initiating a call with the operator.

[0071] Figure 4 shows an example of a caller identification table. The caller identification table registers call histories that fall under the category of inappropriate calls, based on past call histories recorded in the call history DB221. The caller identification table records the personal information of the caller who made the inappropriate call, the duration of the past call, the number of calls made, the number of times certain words or tones were deemed inappropriate, the caller's status on other companies' blacklists, and a specific caller score.

[0072] In the example caller identification table in Figure 4, personal information such as name, address, and telephone number can be registered. Personal information is not limited to the example in Figure 4; any identifier that uniquely identifies an individual, such as a telephone number or driver's license number, is acceptable. Personal information can be pre-registered identification information such as a common ID or My Number, or it can be unregistered identification information. Past call time records the total call time per month, the average call time per call, the shortest call time, the longest call time, and the median call time. The number of calls made per month is the number of times the caller has called the call center. The number of NGs records the number of times inappropriate words were detected and the number of times inappropriate vocal characteristics (tones) were detected. The blacklist records whether or not other companies have blacklisted the caller. Note that past call time, number of calls made, and number of NGs may be calculated from the previous month's call history, or from the average of the call history for each month over a year.

[0073] The specific caller score can be calculated as a function of past call duration, number of calls made, number of rejected calls, and blacklist information. For example, for callers with a history of inappropriate calls, the coefficients for past call duration, number of calls made, number of rejected calls, and blacklist information can be determined so that the specific caller score calculated from the caller's call history is 1. Alternatively, the specific caller score for each individual caller can be calculated using a learning model that has been trained to output 1 as the specific caller score when given past call duration, number of calls made, number of rejected calls, and blacklist information for inappropriate calls. By recording the specific caller score, the likelihood of a caller making inappropriate calls can be quantified in advance. Note that the specific caller score is not limited to a score in the range of 0 to 1, but may be a numerical range of, for example, 1 to 100, or it may be a cumulative value obtained by accumulating the specific caller score over a predetermined period.

[0074] In the caller identification table in Figure 4, for example, looking at caller BBbb, the longest call duration each month was 50 minutes, the total monthly call duration was 900 minutes, the number of calls was 100, the number of NG words was 100, the number of NG tones was 30, and furthermore, the caller is listed on the blacklists of companies A and B, resulting in the highest possible specific caller score of 1. The specific caller scores for callers AAaa, CCcc, and DDdd are calculated as 0.6, 0.7, and 1.0, respectively. These values ​​may be updated monthly or in real time. Of the NG counts, NG words refer to the number of times inappropriate terms were used, and NG tones refer to the number of times inappropriate sounds such as shouting were recorded.

[0075] When a call is received from a caller, the caller identification unit 211 may determine whether or not the caller is a specific caller by considering the caller's personal information in addition to the specific caller score. The caller identification unit 211 may, for example, refer to a member table or customer table using the caller's telephone number to obtain the caller's personal information. If the caller is a specific individual, such as a member participating in the company's service or a customer of the company, the caller identification unit 211 may connect the call to the operator call device 5 without activating the call intervention AI 32. In this case, the operator may be notified of the personal information.

[0076] (3-2-2) Operator's communication section 212 The operator call unit 212 manages the call between the caller and the operator. When the operator call unit 212 receives an incoming call notification from the caller determination unit 211, it establishes a session between the caller call device 4 and the operator call device 5 and sets up the system so that the caller and the operator can communicate. The operator operates the operator call device 5 to receive the call from the caller and communicate with the caller. The operator call unit 212 notifies the call determination unit 213 of the call response information between the caller and the operator. Alternatively, instead of the operator call unit 212 notifying the call determination unit 213 of the call response information, the call determination unit 213 may be configured to obtain the call response information by querying the operator call unit 212.

[0077] The operator communication unit 212 continues the call between the caller and the operator until it receives notification of call termination or a session disconnection request. When the operator communication unit 212 receives notification of call termination from the call determination unit 213, or when it receives notification from the call intervention unit 214 of a request to disconnect the session between the caller and the operator, it disconnects the session between the caller and the operator and terminates the call.

[0078] (3-2-3) Call determination section 213 The call determination unit 213 monitors the call in a manner that does not affect the call between the operator and the caller. The call determination unit 213 obtains call response information, including caller statements, operator statements, or response time, from the operator call unit 212, and determines whether the call is an inappropriate call. The call determination unit 213 inputs the obtained call response information into the call determination AI 31 and obtains a call response score from the call determination AI 31.

[0079] The call determination unit 213 determines whether a call between the operator and the caller is an inappropriate call based on whether the call response score obtained from the call determination AI 31 exceeds a predetermined threshold. For example, if the call response score is less than 0.8, it determines that a normal call is taking place. In this case, the call determination unit 213 determines whether the call has ended, that is, whether the caller has hung up, and if the call has not ended, it continues to monitor the call between the operator and the caller. On the other hand, if it determines that the call has ended, it notifies the operator call unit 212 of the end of the call. The call determination unit 213 requests the call record management unit 215 to record the call history.

[0080] On the other hand, if the call judgment unit 213 obtains a call response score from the call judgment AI 31 and the call response score is equal to or greater than a predetermined threshold (for example, 0.8; hereinafter also referred to as the "first threshold"), it determines that the call between the operator and the caller is an inappropriate call and sends an intervention request to the call intervention unit 214 to intervene in the call between the caller and the operator. The call judgment unit 213 may also notify the administrator that an inappropriate call has occurred and provide the call response score along with the intervention request to the call intervention unit 214. Alternatively, the call response score corresponding to the call between the operator and the caller may be displayed to the administrator in real time, regardless of whether an intervention request has been made.

[0081] If the call handling score exceeds a second threshold (for example, 0.95) that is higher than a predetermined threshold, the call determination unit 213 may determine that it cannot continue the call, notify the caller of a message that it cannot handle the call any further, and forcibly terminate the session to end the call. In this case, the call determination unit 213 notifies the operator call unit 212 of the end of the call and requests the call record management unit 215 to record the call history.

[0082] The call judgment AI 31 is trained to use typical normal and typical inappropriate calls between callers and operators as training data, and to output 1 if it is a typical inappropriate call and 0 if it is a typical normal call. Once trained, the call judgment AI 31 is configured to output a value between 0 and 1 as a call response score when it receives call response information, including caller statements, operator statements, or response time.

[0083] The call judgment AI 31 may estimate the response time corresponding to the content of the spoken words and output the expected response time. For example, if the caller requests a detailed explanation of a product or service, the response time is often long, even in a normal call. For this reason, the call judgment AI 31 can learn the relationship between past caller statements and actual response times and output an expected response time from the input caller statements. The call judgment unit 213 may decide whether to intervene in the call based on the monitored response time included in the call handling information and the estimated response time output by the call judgment AI 31.

[0084] Figure 5 is a diagram illustrating an example of the learning process for the call detection AI 31. Figure 5(a) shows an example of a judgment criteria table in which criteria for determining inappropriate calls are set for extracting training data from past call history. The judgment criteria table sets the target for judgment and the criteria for determining inappropriate calls for each of the following: call duration, caller statements, and operator statements. Based on these judgment criteria, positive training data corresponding to inappropriate calls and negative training data corresponding to normal calls can be extracted from past call history recorded in the call history DB221.

[0085] In the example in Figure 5(a), the response time is determined by the response time for each service, and the criteria for determining an inappropriate call are set at 30 minutes for service A and 60 minutes for service B. For example, if service A is a service that does not require telephone explanation by an operator, and service B is a service that requires telephone explanation by an operator, then it is expected that the telephone response time for service B will be longer than that for service A. Therefore, in the past call history, for calls related to service A, calls with a response time exceeding 30 minutes are extracted as training data for inappropriate calls related to service A. On the other hand, for service B, calls with a response time exceeding 60 minutes are extracted as training data for inappropriate calls related to service B.

[0086] Regarding caller statements, criteria for determining inappropriate calls are defined based on the content of the statement, voice characteristics (sound pressure, frequency, speed), and call etiquette. For the content of the statement, abusive language, threats, insults, discrimination, and circular reasoning are defined as criteria for judgment. Training data can be generated by extracting calls similar to these statement contents from past call history recorded in the call history DB221. For example, past statement contents that will serve as training data can be extracted by calculating the similarity (e.g., cosine similarity) between the feature vector of the caller statement (text data) recorded in the call history DB221 and the feature vector of text data predefined as abusive language, threats, insults, discrimination, and circular reasoning.

[0087] When a caller's speech is shouting, intimidating, or otherwise offensive, it exhibits different vocal characteristics such as sound pressure, frequency, and speed compared to normal conversations. Therefore, by setting vocal characteristics (sound pressure, frequency, and speed) that constitute inappropriate communication as criteria, and extracting call responses with these characteristics from past call history, training data can be generated. These vocal characteristics can be extracted from both the caller's and operator's statements.

[0088] Call etiquette can also be used as a criterion for determining inappropriate calls. Inappropriate calls include call etiquette such as interrupting and talking over the other person. Therefore, call etiquette that constitutes an inappropriate call can be set as a criterion, and call responses that exhibit such call etiquette can be extracted from past call history and used as training data.

[0089] Regarding operator statements, criteria for determining inappropriate calls are defined based on the content of the statements, voice characteristics, and activation switches. For the content of the statements and voice characteristics, prohibited phrases and words are set as criteria for determining inappropriate calls. These prohibited phrases and words are those that should not be used by operators.

[0090] As a criterion for determining whether a call is inappropriate based on operator statements, operator intervention request messages can also be set. For example, operator intervention request messages are called activation switches. When an operator makes a specific statement requesting escalation of the call, it is determined that an activation switch has been activated. For example, statements such as "I will transfer you to the customer complaints department" or "I will consult with my supervisor" can be registered as operator activation switches. The call judgment AI31 can determine that an inappropriate call has occurred if the operator's statement contains an activation switch.

[0091] The criteria shown in Figure 5 are typical, but other criteria can also be used. For example, conditions could be set for connecting and disconnecting calls between callers and operators. For instance, a distinction could be made between toll-free numbers and premium-rate numbers, allowing longer calls for premium-rate numbers, varying the time until the call is disconnected depending on whether it's a toll-free or premium-rate number, or estimating the quality of the caller based on their membership rank and setting criteria for judgment.

[0092] As another example, the call judgment AI31 can output a call handling score based on the compatibility between the speaker and the operator. In other words, if a particular operator handles a call from a certain caller, inappropriate calls may be less likely to occur. Therefore, by including the caller's identification information and the operator's identification information in the training data, it is possible to train the AI ​​while also considering the compatibility between the caller and the operator.

[0093] As another example, the call judgment AI31 can output a call handling score based on the operator's experience and evaluation. In other words, if the operator has extensive customer service experience or a high customer service evaluation, inappropriate calls tend to occur less frequently. Therefore, by including the operator's customer service experience and evaluation (years, performance, etc.) in the training data, it is possible to train the AI ​​while also considering the operator's experience and evaluation.

[0094] Figure 5(b) is a diagram illustrating the training of the call detection AI 31 using positive and negative training data. From the past call history recorded in the call history DB221, data that matches the criteria in the judgment criteria table is extracted as positive training data, and data that does not match the criteria in the judgment criteria table is extracted as negative training data. This positive and negative training data is then input into the call judgment AI31. The call judgment AI31 is then trained to output a call response score of "1" when the positive training data "inappropriate call" is input, and to output a call response score of "0" when the negative training data "normal call" is input. In this disclosure, an example is shown in which the conversation content judgment AI31 is trained using positive and negative training data, but the call judgment AI31 may also be trained using only positive training data.

[0095] Figure 6 is an operation flowchart showing an example of the operation overview of the call determination unit 213. When the call determination unit 213 obtains call response information from the operator call unit 212, it inputs the call response information to the call determination AI 31 (S50). The call determination unit 213 obtains the call response score output from the call determination AI 31 (S51). The call determination AI 31 outputs a call response score based on the input call response information of the caller. The call response score takes a value between 0 and 1, with a score close to 0 if the call is appropriate and a score close to 1 if the call is inappropriate.

[0096] The call determination unit 213 determines whether a call between the caller and the operator is an inappropriate call based on whether the call response score obtained from the call determination AI 31 is greater than a first threshold (for example, 0.8) (S52). If it is determined that the call is not an inappropriate call (No in S52), it determines whether the call has ended, and if not, it continues to monitor the call. The determination of whether the call has ended can be made by detecting that the caller has hung up.

[0097] The call determination unit 213 determines that an inappropriate call is taking place if the call response score obtained from the call determination AI 31 exceeds a first threshold, for example, if it exceeds 0.8 (Yes in S52). Next, the call determination unit 213 determines whether the call is a call that should be terminated (S54). If it determines that an inappropriate call is taking place, the call intervention unit 214 can be activated immediately, but if the degree of inappropriateness is extremely high and the call intervention AI 32 cannot provide an appropriate response, the call may be terminated immediately. If the call response score exceeds a second threshold, for example, if it exceeds 0.95 (Yes in S54), the call determination unit 213 terminates the call and forcibly ends the call with the caller. At this time, a message informing the caller that the call has been terminated may be sent. The call determination unit 213 reports this call history to the call record management unit 215.

[0098] If the call determination unit 213 determines that the call response score is between a first threshold and a second threshold, for example, if the call response score is between 0.8 and 0.95 (No. in S54), it sends an intervention request to the call intervention unit 214 to intervene in the call between the caller and the operator (S56).

[0099] (3-2-4) Call intervention department 214 The call intervention unit 214 receives an intervention request from the caller determination unit 211 or the call determination unit 213. When the call intervention unit 214 receives an intervention request from the caller determination unit 211, the call intervention AI 32 receives the incoming call from the caller instead of the operator receiving it. On the other hand, when the call intervention unit 214 receives an intervention request from the call determination unit 213, it sends a disconnection request notification to the operator call unit 212 to disconnect the session between the caller and the operator, and establishes a session between the caller and the call intervention AI 32.

[0100] The call intervention unit 214 instructs the call intervention AI 32 to output a response result to the caller's input. The call intervention unit 214 generates a response to the caller's statement and sends the generated response back to the caller as an audio signal. The call intervention unit 214 repeats this operation until the call ends. When the call ends, the call intervention unit 214 activates the call record management unit 215 and requests that the call history be recorded.

[0101] The call intervention AI 32 may have a function to confirm the main points of the caller's inquiry, extract the points of the complaint, and report those points to the administrator. If the call intervention AI 32 determines that the complaint is severe and difficult for a human to handle, it may have a function to forcibly end the call after confirming the content of the inquiry, with a response such as "We have received your feedback." In response, the administrator can take measures such as checking the call history obtained from the call intervention AI 32 and responding to the caller or leaving a record of the interaction after the call has ended.

[0102] Figure 7 illustrates an example of the learning process for the call intervention AI 32. The call intervention AI 32 is a large-scale language model called a generative AI. The call intervention AI 32 can be fine-tuned to suit the specific tasks of a call center by using a pre-trained model. The knowledge used for fine-tuning is pre-recorded in the call intervention AI knowledge database 224 and includes pre-prepared Q&A, conversation order, interaction history, and warnings for inappropriate phrases and words.

[0103] The call intervention AI32 takes the caller's statement as input and outputs a response to that statement. For example, if the caller inputs a question, it will refer to the Q&A knowledge base and output a response in accordance with the FAQ. If the caller requests an explanation about a product or service, it can provide a complex explanation or an operation manual. If the caller uses inappropriate phrases or words, it may output a request for the statement to be retracted.

[0104] Figure 8 is an operation flowchart showing an example of the operation overview of the call intervention unit 214. When the call intervention unit 214 receives an intervention request from the call determination unit 213, it performs call receiving processing using the call intervention AI 32 (S70). The call intervention unit 214 notifies the operator call unit 212 of the end of the call. Next, the call intervention unit 214 intervenes in the call between the caller and the operator. The call intervention unit 214 establishes a session between the caller and the call intervention AI 32, enabling the caller and the call intervention AI 32 to communicate. Note that if the call intervention unit 214 receives an intervention request from the caller determination unit 211, it performs call receiving processing using the call intervention AI 32 (S70), but does not notify the operator call unit 212 of the end of the call.

[0105] The call intervention unit 214 inputs the caller's statement to the call intervention AI 32 (S71). The call intervention unit 214 causes the call intervention AI 32 to generate a response to the statement. The call intervention unit 214 transmits the voice response output by the call intervention AI 32 to the caller (S72).

[0106] The call intervention unit 214 determines whether the call has ended or not (S73). If the call has not ended (S73: No), it repeatedly executes the exchange between the caller and the call intervention AI 32 (S71, S72). On the other hand, if the call has ended (S73: Yes), it requests the call record management unit 215 to record the call history.

[0107] (3-2-5) Call Record Management Department 215 The call record management unit 215 receives the call history between the caller and the operator from the call judgment unit 213 and records it in the call history DB 221. The call history from the call judgment unit 213 includes call response information (caller's statements, operator's statements or operator's response time, caller identification information, operator identification information, call response score output by the call judgment AI 31, etc.).

[0108] When the call intervention unit 214 is activated, the call record management unit 215 receives the call history between the caller and the call intervention AI 32 from the call intervention unit 214 and records it in the call history DB 221. The call record from the call intervention AI 32 includes call response information including the caller's statements, the call intervention AI 32, and the response time, as well as caller identification information and call intervention AI identification information.

[0109] The call record management unit 215 analyzes past call history at the operator's instruction or at any time, extracts call history of recipients with high call handling scores, and updates the caller identification table (Figure 4). The call record management unit 215 may also statistically process the monthly status of inappropriate calls based on the call history and create reports.

[0110] The call record management unit 215 can also prepare documents for litigation in cases of inappropriate conduct that may warrant legal action. The call record management unit 215 reads knowledge for the litigation preparation AI from the litigation preparation AI knowledge database 225 and fine-tunes the litigation preparation AI 33. When a case that may warrant legal action is input to the fine-tuned litigation preparation AI 33, it outputs applicable legal provisions, the application of the case to the legal provisions, and the legal effects, as well as creating a draft complaint regarding the inappropriate call case. Examples of cases that may warrant legal action include "repeatedly making phone calls to disrupt business operations," "repeatedly using abusive language," and "making unreasonable demands for money," but it may also be limited to cases where the operator is not at fault.

[0111] Figure 9 illustrates an example of the training of the complaint preparation AI 33. The knowledge for the complaint preparation AI can include legal provisions related to inappropriate phone calls, judgments related to cases involving inappropriate phone calls, and sample complaints that serve as templates for complaints. Examples of legal provisions include the crime of intimidation (Article 222 of the Penal Code), the crime of coercion (Article 223 of the Penal Code), the crime of obstruction of business by force (Article 234 of the Penal Code), and the crime of extortion (Article 249 of the Penal Code). The judgments include the facts of the case, applicable laws and regulations, requirements, and effects, and may also include the complaint concerning the case.

[0112] The call record management unit 215 extracts the call history of the case to be sued as instructed by the operator from the call history DB 221, inputs it into the complaint preparation AI 33, and commands it to generate a draft complaint. Based on the call history of the case to be sued, the complaint preparation AI 33 identifies applicable laws (articles), estimates their application to legal requirements, outputs the estimated legal effects, and generates a draft complaint for the case to be sued.

[0113] Figure 10 is an operation flow diagram showing an example of the operation overview of the call record management unit 215. The call record management unit 215 collects the call history between the caller and the operator from the operator call unit 212. In addition, if the call intervention unit 214 is activated, it collects the call history between the caller and the call intervention AI 32 from the call intervention unit 214 (S90).

[0114] The call record management unit 215 analyzes the recorded call history (S91). The call record management unit 215 can perform statistical analysis on a yearly, monthly, and service-specific basis regarding the purpose of calls to the call center (inquiries, complaints, etc.), the number of calls made, and the average call duration. Statistical analysis may be performed periodically or when instructed by the operator. In addition, for callers with high call handling scores among the collected call history, the call record management unit 215 may analyze the monthly call duration, number of calls made, number of NG words, number of NG tones, and blacklist status, update the caller identification table (Figure 4), and record the results in the caller identification DB 222.

[0115] The call record management unit 215 can create reports based on statistical analysis (S92). Administrators can refer to the reports to understand the operational status of the call center and to check for any cases that may constitute customer harassment. The call record management unit 215 may also extract maladaptive cases from the call history, create reports, and provide feedback for operator training. Furthermore, the training data for the call judgment AI 31 and the call intervention AI 32 may be updated based on the maladaptive cases.

[0116] The call record management unit 215 determines whether or not there are cases that can be sued (S93). Here, cases that can be sued are not limited to cases that are actually planned to be sued, but may also include cases for which the possibility of suing should be determined. For example, if the administrator wants to determine whether or not a case can be sued in accordance with the law (evaluation of win or loss from a legal and technical standpoint), or whether or not a case can be sued in accordance with the content of the case (humanitarian, social, and economic necessity), the administrator can present the case as a case that can be sued. If the administrator does not specify a case that can be sued, the unit determines that there are no cases that can be sued (S93: No) and terminates the process. On the other hand, if a case that can be sued is specified, the unit determines that there is a case that can be sued (S93: Yes), and the call record management unit 215 obtains the call history of the specified case that can be sued, and instructs the litigation preparation AI 33 to input the call history of the case that can be sued and generate a draft complaint for the case that can be sued (S94).

[0117] In step S93 of Figure 10, the determination of whether or not there are cases subject to litigation was made based on the presentation of cases subject to litigation by the administrator. However, the determination of whether or not there are cases subject to litigation may also be made by the litigation preparation AI 31 searching for cases subject to litigation. For example, stricter criteria for determining cases subject to litigation than the criteria for determining inappropriate calls (Figure 5) may be set in advance, and the litigation preparation AI 33 may extract call histories from the call history DB 221 that satisfy the criteria for determining cases subject to litigation and determine that there are cases subject to litigation. Alternatively, call response scores that correspond to cases subject to litigation may be set in advance, and the litigation preparation AI 33 may extract call histories from the call history DB 221 that satisfy the call response scores that correspond to cases subject to litigation and determine that there are cases subject to litigation.

[0118] The litigation preparation AI33 obtains applicable legal provisions, legal requirements, and legal effects for the case to be sued, and outputs a generated draft complaint (S95).

[0119] As described above, according to the first embodiment, the call judgment AI 31 monitors the call between the caller and the operator, and when an inappropriate call occurs, the call intervention AI 32 intervenes in the call and takes over from the operator, enabling intervention at the appropriate time and continuous call handling. Specifically, the call judgment AI 31 grasps the call between the caller and the operator in real time, and if the requirements for an inappropriate call are met, the call intervention AI 32 can forcibly take over from the operator and take over the subsequent interaction. In addition, by setting a limit on the operator's response time and intervening with the call intervention AI 32 when that limit is exceeded, the burden of long-term handling can be reduced. Furthermore, it is also possible for the call intervention AI 32 to directly handle calls from specific callers that have been identified in advance. By having the call intervention AI 32 intervene, the burden on operators can be reduced, and it becomes possible to address the social issue of so-called customer harassment.

[0120] <Second Embodiment> In the first embodiment, when a call between a caller and an operator is deemed inappropriate, the call intervention AI 32 intervenes in the call to reduce the operator's burden. In the second embodiment, the system is configured to allow the call intervention AI 32 to intervene in calls other than inappropriate calls in order to reduce the operator's burden.

[0121] When an operator receives an inquiry from a caller, for example, about a product or service, and there is no specialized department for handling such inquiries, the operator may have to answer the inquiry themselves. In such cases, the operator will have to provide a lengthy response while referring to product or service manuals, etc. The second embodiment can be used to avoid such lengthy responses.

[0122] In the second embodiment, the call intervention AI 32 can intervene in a call when a caller requests a detailed explanation from an operator, that is, when a given call requires a detailed explanation from an operator. A call requiring a detailed explanation from an operator is, for example, a call that includes an explanation based on a product or service manual.

[0123] The call determination unit 213 monitors the call between the caller and the operator, and if it determines that the call between the caller and the operator includes an activation switch, it can activate the call intervention AI 32. Examples of activation switches include statements made by the operator such as "I will explain the product details," "I will introduce the product manual," or "I will explain a specific example of the service." The call intervention AI 32 has learned detailed information or manuals for products and services in advance, and can provide detailed explanations of products and services or present manuals in response to inquiries from the caller.

[0124] According to the second embodiment, in response to specialized questions or requests for detailed explanations from the caller, the operator can activate the call intervention AI 32 and have the AI ​​32 provide the explanation. This allows the operator to avoid lengthy conversations and also ensures that the caller receives an accurate answer.

[0125] This disclosure is not limited to the embodiments described above, and the components can be modified and implemented in practice without departing from the gist of the invention. Furthermore, various inventions can be formed by appropriately combining the multiple components disclosed in the embodiments. For example, some components may be deleted from all the components shown in the first embodiment. Moreover, components from different embodiments may be appropriately combined. [Explanation of Symbols]

[0126] 1...Call management system, 2...Call management device, 3...Call support AI, 4...Caller call device, 5...Operator call device, 6...Network, 21...Processor, 22...Memory, 23...Storage, 24...Communication I / F, 25...Input / Output I / F, 31...Call judgment AI, 32...Call intervention AI, 33...Litigation preparation AI, 41...Caller call device a, 42...Caller call device b, 51...Operator call device A, 52...Operator call device B, 210...Call management unit, 211...Caller judgment unit, 212...Operator call unit, 213...Call judgment unit, 214...Call intervention unit, 215...Call record management unit, 220...Storage unit, 221...Call history DB, 222...Caller judgment DB, 223...Call judgment AI learning data DB, 224...Call intervention AI knowledge DB, 225...Litigation preparation AI knowledge DB

Claims

1. A call determination unit that determines whether a call between a caller and a human operator includes inappropriate call handling, In response to the determination that the call includes the inappropriate call handling, a call intervention unit is provided in which a generating AI, trained to generate a call handling response with the caller, intervenes in the call with the caller on behalf of the human operator. Equipped with, The call determination unit includes a learning model trained to generate call responses with the caller, input call response information including the caller's statements, the human operator's statements, or the human operator's service-specific response time, and output a call response score indicating the degree of inappropriateness of the call response. Information processing device.

2. The information processing apparatus according to claim 1, wherein the call including the inappropriate call handling includes a call in which the call handling score, which indicates the degree of inappropriateness of the call handling, exceeds a predetermined threshold.

3. The information processing apparatus according to claim 1, wherein the call intervention unit comprises a large-scale language model trained to output a response corresponding to the caller's statement.

4. A caller determination unit that determines that the caller is a specific caller, A call determination unit that determines whether the call between the caller and the human operator includes inappropriate call handling, In response to the determination that the call includes the inappropriate call handling, a call intervention unit is provided in which a generating AI, trained to generate a call handling response with the caller, intervenes in the call with the caller on behalf of the human operator. Equipped with, If the caller determination unit determines that the caller does not belong to the specified caller, it initiates a call between the caller and the human operator, and in response to determining that the caller is the specified caller, it activates the call intervention unit. Information processing device.

5. The information processing apparatus according to claim 4, wherein the specific caller is determined based on a specific caller score calculated based on the caller's call history.

6. An information processing method performed by an information processing device, To determine whether a call between a caller and a human operator includes inappropriate call handling, In response to the determination that the call includes the inappropriate call handling, a generative AI trained to generate call handling responses with the caller intervenes in the call with the caller on behalf of the human operator, Equipped with, The determination includes making the determination using a learning model trained to input call response information, including the caller's statements, the human operator's statements, or the human operator's service-specific response time, as a generative AI trained to generate call responses with the caller, and outputting a call response score indicating the degree of inappropriateness of the call response. Information processing methods.

7. An information processing method performed by an information processing device, Determining that the sender is a specific sender, To determine whether a call between a caller and a human operator includes inappropriate call handling, In response to the determination that the call includes the inappropriate call handling, a generative AI trained to generate call handling responses with the caller intervenes in the call with the caller on behalf of the human operator, Equipped with, The determination of whether the caller is a specific caller is configured such that, if it is determined that the caller is not a specific caller, a call is initiated between the caller and the human operator, and in response to the determination that the caller is a specific caller, intervention in the call is performed. Information processing methods.

8. A program that realizes the functions of each part of the information processing apparatus described in any one of claims 1 to 5.

Images