Control program and control method

Abstract

To solve the following problems in which there are training tools for elderly people who are susceptible to deception and show higher levels of cognitive bias, to experience a simulated special fraud in order to raise awareness of fraud prevention, however, readily available training tools for the latest scenario of various special frauds are desired, the tools being configured to use an AI virtual fraud to talk with an elderly person along a scenario of a special fraud for experiencing a simulated special fraud.SOLUTION: A control program is executed by an interaction support apparatus which supports interactions between a virtual agent representing a person who executes a special fraud and a subject who experiences the special fraud. The program includes: generating a scenario for the special fraud using special fraud damage information; and causing a computer to perform processing to output the generated scenario, as dialogue information of the virtual agent, to a terminal used by the subject.SELECTED DRAWING: Figure 6

Description

Technical Field

[0001] The present invention relates to a control program and a control method.

Background Art

[0002] Elderly people who are easily deceived have a high cognitive bias (confirmation bias). For example, they believe that they will not be victimized by special fraud. Therefore, for example, by having such elderly people experience a simulated special fraud and actually be deceived, their awareness of fraud prevention can be enhanced.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, for example, in order to have someone experience a simulated special fraud, a training tool can be considered that makes a virtual fraudster speak according to a scenario of special fraud by AI (Artificial Intelligence) and have a conversation with the person experiencing it. However, there are various cases of special fraud. Also, for example, in special fraud, when countermeasures against the modus operandi of special fraud are taken, new modus operandi of special fraud keep emerging one after another. For this reason, for example, with a training tool, it is impossible to effectively train against the modus operandi of special fraud, and it has been difficult to enhance the training effect of the trainer.

[0005] In one aspect, an object is to provide a control program and a control method that can effectively train a trainer.

Means for Solving the Problems

[0006] In one aspect, the control program is a control program executed by a dialogue support device that supports the dialogue between a virtual agent indicating a special fraud perpetrator and a special fraud subject, and uses the special fraud damage information to generate a scenario related to the special fraud, and causes a computer to execute a process of outputting the generated scenario to a terminal used by the subject as the speech information of the virtual agent.

Effect of the Invention

[0007] In one aspect, it aims to enable effective training for the trainer.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

[0009] Hereinafter, a control program according to this embodiment and an example of a control method will be described in detail with reference to the drawings. Note that this embodiment is not limited by this example. Also, each example can be appropriately combined within a non - contradictory range.

[0010] [Configuration of Information Processing System 1] First, an information processing system for implementing this embodiment will be described. FIG. 1 is a diagram showing a configuration example of an information processing system 1 according to this embodiment. As shown in FIG. 1, the information processing system 1 is a system in which, for example, an interaction support device 10 and a trainer terminal 100 are communicably connected to each other via a network 50.

[0011] For the network 50, various communication networks such as the Internet can be adopted regardless of whether they are wired or wireless. Also, the network 50 may not be a single network, and for example, an intranet and the Internet may be configured via a network device such as a gateway or other devices (not shown).

[0012] The interaction support device 10 is an information processing device managed, for example, by a service administrator who supports the interaction between a virtual agent indicating a perpetrator of an event such as special fraud and a trainer for an event such as special fraud. The interaction support device 10 may be, for example, a desktop PC (Personal Computer), a notebook PC, or a server computer. Note that the virtual agent is also called, for example, a virtual agent, and is an automatic conversation program in a virtual space using AI or machine learning, and conducts a conversation with a user (human) in real time. Note that in the description of this embodiment, special fraud is used as an example of an event, but the event is not limited to special fraud.

[0013] The dialogue support device 10 generates a scenario related to an event such as special fraud using the occurrence information of the event such as the damage information of special fraud, and outputs the generated scenario to the trainer terminal 100 used by the trainer as the speech information of the virtual agent. Further, the dialogue support device 10 receives the speech information of the trainer from the trainer terminal 100, for example, and analyzes the speech information to support the dialogue between the virtual agent on the trainer terminal 100 and the trainer. Further, the dialogue support device 10 analyzes the mental state of the trainer using, for example, the physiological reaction of the trainer in a predetermined section where the dialogue executed between the virtual agent and the trainer is being carried out. Then, the dialogue support device 10 evaluates the risk for an event such as the risk of the trainer being deceived by special fraud using, for example, the analyzed mental state.

[0014] In addition, in FIG. 1, the dialogue support device 10 is shown as one computer, but it may be a distributed computing system composed of a plurality of computers. Further, the dialogue support device 10 may be a cloud computer device managed by a service provider that provides cloud computing services.

[0015] The trainer terminal 100 may be, for example, a desktop PC, a notebook PC, a mobile terminal such as a smartphone or a tablet PC, or a telephone used by the trainer for an event such as special fraud. For example, the trainer can have a conversation with a virtual agent indicating the perpetrator of an event such as special fraud via the trainer terminal 100 and can pseudo-experience an event such as special fraud.

[0016] [Functional Configuration of Dialogue Support Device 10] Next, the functional configuration of the dialogue support device 10 will be described. FIG. 2 is a diagram showing a configuration example of the dialogue support device 10 according to the present embodiment. As shown in FIG. 2, the dialogue support device 10 includes a communication unit 20, a storage unit 30, and a control unit 40.

[0017] The communication unit 20 is a processing unit that controls communication with other devices such as the trainer terminal 100, and is, for example, a communication interface such as a network interface card or a USB (Universal Serial Bus) interface.

[0018] The storage unit 30 has a function of storing various data and programs executed by the control unit 40, and is realized by a storage device such as a memory or a hard disk, for example. The storage unit 30 stores, for example, event occurrence information 31, scenario information 32, model information 33, and digital twin information 34.

[0019] The event occurrence information 31 stores, for example, occurrence information of events such as special fraud. The occurrence information of the event may be, for example, damage information of special fraud for an arbitrary period in the target area extracted from a database of public security information.

[0020] The scenario information 32 stores, for example, information about scenarios generated for events such as special fraud. Details of the generation of the scenario will be described later, but the scenario is selected based on a predetermined selection condition for establishing the event from the occurrence information of the event such as damage information of special fraud, and each element selected, and is generated using a generation AI (model) for scenario generation.

[0021] The model information 33 stores, for example, information about a machine learning model for generating a scenario related to an event such as special fraud, model parameters and training data for constructing the machine learning model, etc. The machine learning model may be generated by machine learning using, as feature amounts, each element selected based on a predetermined selection condition for establishing the event from the occurrence information of the event such as damage information of special fraud, and using the scenario related to the event as a correct label.

[0022] In addition, the model information 33 stores, for example, information regarding a machine learning model for estimating the mental state of a trainer, model parameters for constructing the machine learning model, training data, and the like. The machine learning model may be generated by machine learning using, for example, the vital information of a person as a feature amount and the mental state of the person as a correct label.

[0023] Here, the vital information of the trainer may include, for example, the trainer's pulse (heartbeat), respiration (rate), blood pressure, body temperature, sweating, etc., and further may include various information that can be obtained from the trainer's body, such as the trainer's voice and voice tone. Note that the trainer's pulse, etc., may be obtained from a sensing device such as a millimeter wave sensor that is communicably connected to, for example, the dialogue support device 10 or the trainer terminal 100. Also, the trainer's voice, etc., may be obtained from a microphone or the like that is communicably connected to, for example, the dialogue support device 10 or the trainer terminal 100. In addition, the mental state may include, for example, states such as interest, excitement, joy, surprise, distress, anxiety, anger, disgust, contempt, fear, shame, and guilt. The estimation of the mental state may be, for example, a numerical value indicating the probability or reliability of the trainer being in which mental state.

[0024] In addition, the model information 33 stores, for example, information regarding a machine learning model for estimating the risk of an event such as special fraud, such as the risk of the trainer being deceived by special fraud, model parameters for constructing the machine learning model, training data, and the like. The machine learning model may be generated by machine learning using, for example, the mental state of a person as a feature amount and the risk of the person being deceived by special fraud or the risk of an event such as special fraud of the person as a correct label. Also, for example, further, as feature amounts for the machine learning, information such as the basic data of the person such as the person's age and gender, the emotional pattern, and the psychological characteristics may be used to train the machine learning model. The emotional pattern may be, for example, a pattern of changes in the mental state of the person within a preset time. Also, the psychological characteristics may be, for example, data indicating the personality of the person, such as a personality that is easily deceived.

[0025] The digital twin information 34 stores information regarding the digital twin, such as a virtual agent that deals with real-world fraudsters. Further, the digital twin information 34 may store, for example, the damage situation in the real world and the utterance information of the virtual agent generated by a simulation on the digital twin using the virtual agent or the like. Note that the digital twin is a technology for representing an object existing in the physical space of the real world in the virtual space. For example, in the digital twin, various information occurring in the real world is collected using IoT (Internet of Things) or the like, and events in the real world are simulated in the virtual space, and the result is utilized for estimation in the real world.

[0026] Note that the above information stored in the storage unit 30 is merely an example, and the storage unit 30 can store various information other than the above information.

[0027] The control unit 40 is a processing unit that controls the entire dialogue support apparatus 10, and is, for example, a processor or the like. The control unit 40 includes an acquisition unit 41, a generation unit 42, an output unit 43, and an evaluation unit 44. Note that each processing unit is an example of an electronic circuit included in the processor or an example of a process executed by the processor.

[0028] The acquisition unit 41 acquires, for example, damage information of special fraud from a predetermined database, and classifies the acquired damage information for each similar case using an existing clustering algorithm. FIG. 3 is a diagram showing an example of the overall flow of scenario generation according to the present embodiment. As shown on the left side of FIG. 3, for example, since the damage information of special fraud is extracted in advance from the public security information database and stored in the event occurrence information 31, the acquisition unit 41 acquires the damage information from the event occurrence information 31. Note that the damage information extracted from the public security information database may be, for example, damage information for an arbitrary period in the target area. Further, as shown in FIG. 3, there are also a large number of fragmentary information among the damage information.

[0029] In addition, the acquisition unit 41 acquires the speech information of the trainer, for example, by analyzing the speech data of the telephone used by the trainer.

[0030] The generation unit 42 generates a scenario related to an event such as special fraud, for example, using the occurrence information of the event such as the damage information of special fraud acquired by the acquisition unit 41. The generation of the scenario may be created, for example, using a generation AI (model) for scenario generation. However, since the underlying damage information is also fragmentary information, an unnatural fraud scenario may be generated.

[0031] FIG. 4 is a diagram showing an example of a NG example of scenario generation. In FIG. 4, the elements shown as "purpose", "means to achieve the purpose", and "background of the purpose" are examples of the respective elements constituting special fraud from which damage information is extracted. Each element extracted from the damage information is not limited to these. For example, as shown in the center of FIG. 4, it may be various elements constituting special fraud such as "title of the person who made the call" and "person who called".

[0032] However, since the selection conditions for which elements to use for scenario generation are not specified from the extracted elements, for example, as shown in FIG. 4, an unnatural scenario without a connection between "means to achieve the purpose" and "background of the purpose" and lacking consistency is generated. Therefore, in the present embodiment, by specifying the selection conditions for each extracted element and selecting the elements to be used for scenario generation based on the selection conditions, a scenario with higher consistency and accuracy is generated.

[0033] FIG. 5 is a diagram showing an example of an OK example of scenario generation. For example, by specifying the selection conditions of the elements as shown in the upper right of FIG. 5, which satisfy special fraud, as a prompt for the generation AI (model), elements that meet the conditions are selected, and a scenario with consistency can be generated.

[0034] Therefore, for example, based on a predetermined selection condition, the generation unit 42 selects a first element from a plurality of elements constituting the special fraud in the damage information of the special fraud. The first element selected here is, for example, "purpose", "means to achieve the purpose", "background of the purpose", etc. as shown in FIG. 5, each of which is connected. Then, for example, the generation unit 42 generates a scenario using the selected first element.

[0035] Regarding the selection of the first element, more specifically, for example, the generation unit 42 analyzes the damage information of the special fraud by using existing natural language processing or the like, and extracts a plurality of character strings constituting the special fraud from the damage information as a plurality of elements constituting the special fraud. Then, for example, the generation unit 42 selects a first element from the plurality of extracted elements based on a predetermined selection condition for establishing the special fraud. Note that the first element may be, for example, a character string related to at least one of the person who made the call for the special fraud and their title, the person who received the call, the purpose of the call, the conditions for achieving the purpose, and the background of the purpose.

[0036] In addition, for example, the generation unit 42 uses a machine learning model trained with performance information of past events such as information on the modus operandi extracted from the damage information of the special fraud, and generates the speech information of the virtual agent for the speech information of the trainee acquired by the acquisition unit 41.

[0037] More specifically, for example, the generation unit 42 inputs the selected first element into a natural language processing model trained with an existing scenario related to the special fraud to generate a scenario. The natural language processing model may be, for example, a machine learning model generated by machine learning using each element selected based on a predetermined selection condition for establishing the special fraud from the damage information of the special fraud as a feature amount and a scenario related to the special fraud as a correct label.

[0038] Further, the generation unit 42 generates and arranges, for example, a plurality of virtual agents on the virtual space associated with a scenario related to an event on the digital twin using the event occurrence information. More specifically, the generation unit 42 generates and arranges, for example, a plurality of swindlers on the virtual space associated with a scenario related to special fraud on the digital twin using the real-world damage situation regarding special fraud. Note that the reason for generating a plurality of swindlers is that, for example, there are various types of frauds in special fraud, such as impersonation fraud, savings fraud, false charge claim fraud, refund fraud, etc. In order to cope with such various frauds, swindlers for each type of fraud are generated on the digital twin. Also, for example, depending on the region such as the residence area of the trainer, the way of speaking of the swindler (standard language or dialect), the scenario of special fraud, etc. may be different. Therefore, a plurality of swindlers and scenarios are set on the digital twin so that a simulation adapted to the trainer can be performed.

[0039] The generation unit 42 arranges, for example, a first virtual agent corresponding to a swindler in the first area on the digital twin using the damage situation that occurred in the first area. In this case, the first virtual agent is, for example, an agent that reproduces a swindler who commits fraud in the first area. Also, the generation unit 42 arranges, for example, a second virtual agent corresponding to a swindler in the second area on the digital twin using the damage situation that occurred in the second area. In this case, the second virtual agent is, for example, an agent that reproduces a swindler who commits fraud in the second area.

[0040] Then, the generation unit 42, for example, captures information in the real world, performs a simulation of a special fraud execution on the digital twin, and generates speech information of the fraudster. Here, the information in the real world captured on the digital twin may be, for example, the speech information and mental state of the trainer. Therefore, an agent corresponding to the trainer may be further generated and arranged on the digital twin. Also, the simulation may be a simulation for each type of fraud. For example, when the trainer selects the type of fraud and the fraudster for which training is desired, the corresponding simulation is executed.

[0041] The output unit 43, for example, causes the terminal used by the trainer to output the scenario generated by the generation unit 42 as the speech information of the virtual agent. This may be, for example, outputting the voice data of the speech information of the virtual agent generated by the generation unit 42 to the telephone used by the trainer.

[0042] Also, the output unit 43, for example, identifies a first virtual agent selected by the trainer from among a plurality of virtual agents on the virtual space arranged on the digital twin, and outputs the voice data of the speech information of the first virtual agent generated using at least the first virtual agent and the scenario to the terminal used by the trainer. More specifically, the output unit 43, for example, identifies a first fraudster selected by the trainer from among a plurality of fraudsters on the virtual space arranged on the digital twin, and outputs the voice data of the speech information of the first fraudster generated using at least the first fraudster and the scenario to the terminal used by the trainer.

[0043] The evaluation unit 44 analyzes the mental state of the trainer, for example, using the physiological reaction of the trainer during a predetermined period in which a conversation is being executed between the virtual agent and the trainer. FIG. 6 is a diagram showing an example of the training tool according to the present embodiment. As shown in FIG. 6, for example, the evaluation unit 44 acquires the vital information of the trainer as the physiological reaction of the trainer from a sensing device such as a millimeter wave sensor that is communicably connected to the dialogue support device 10, the trainer terminal 100, etc., and estimates the mental state by AI.

[0044] Also, the visualization screen and the feedback screen shown in FIG. 6 are examples of the screens of the training tool displayed via the trainer terminal 100. On the visualization screen, for example, as shown in FIG. 6, the conversation content between the virtual agent and the trainer, the estimated mental state of the trainer, the acquired vital information, etc. may be displayed. Also, on the feedback screen, for example, as shown in FIG. 6, risks for events such as special fraud, such as the risk of the trainer being deceived by special fraud estimated from the mental state of the trainer, and explanations of scenarios related to events such as special fraud used in the training are displayed.

[0045] Also, the estimation of the mental state by AI may be executed, for example, by using a machine learning model generated by machine learning with the vital information of a person as a feature amount and the mental state of the person as a correct label, and inputting the vital information of the trainer into the machine learning model.

[0046] Also, the evaluation unit 44 evaluates risks for events such as special fraud, such as the risk of the trainer being deceived by special fraud, for example, using the analyzed mental state. The risk evaluation may be executed by inputting the trainer information into a machine learning model generated by machine learning with features such as the mental state of a person and the risk for events such as special fraud of the person, such as the risk of the person being deceived by special fraud, as the correct label.

[0047] Also, the risk for events such as special fraud may be a numerical value or classification indicating the magnitude of the risk, such as 75% (high), as shown on the feedback screen of FIG. 6, for example.

[0048] [Flow of processing] Next, with reference to FIG. 7, the flow of the scenario generation process according to the present embodiment will be described. FIG. 7 is a flowchart showing an example of the flow of the scenario generation process according to the present embodiment. The scenario generation shown in FIG. 7 may be executed at an arbitrary timing, such as triggered by newly extracting damage information of special fraud from a database of public security information, for example.

[0049] First, as shown in FIG. 7, the dialogue support device 10 acquires, for example, damage information of special fraud from the event occurrence information 31 (step S101). The dialogue support device 10 acquires, for example, occurrence information of a predetermined event. More specifically, the dialogue support device 10 acquires, for example, damage information in one modus operandi of special fraud. Here, one modus operandi of special fraud is, for example, ole ole type special fraud, fictitious charge claim fraud, financing guarantee fraud, refund fraud, etc. Also, in this case, the dialogue support device 10 may acquire damage information in a plurality of modi operandi of special fraud, for example.

[0050] Next, the dialogue support device 10 classifies the damage information acquired in step S101, for example, into similar cases using an existing clustering algorithm (step S102). The dialogue support device 10 classifies the occurrence information of a predetermined event into similar cases using an existing clustering algorithm, for example. More specifically, for example, when acquiring damage information regarding one modus operandi of special fraud in step S101, the dialogue support device 10 classifies the damage information regarding one modus operandi of special fraud into similar cases using an existing clustering algorithm. Also, for example, when acquiring damage information regarding a plurality of modi operandi of special fraud in step S101, the dialogue support device 10 classifies the damage information regarding the plurality of modi operandi of special fraud into similar cases using an existing clustering algorithm that utilizes job titles of persons or the like. Note that the subsequent processing may be performed for each classified similar case.

[0051] Next, the dialogue support device 10 extracts elements constituting special fraud from each piece of damage information classified in step S102, for example (step S103). The dialogue support device 10 extracts elements constituting a predetermined event from each piece of predetermined event occurrence information classified in step S102, for example. More specifically, the dialogue support device 10 analyzes each piece of damage information classified in step S102 by using existing natural language processing or the like, and extracts a plurality of character strings constituting the special fraud from each piece of damage information as a plurality of elements constituting the special fraud.

[0052] Next, the dialogue support device 10 selects a first element from the plurality of elements extracted in step S103, for example (step S104). The dialogue support device 10 selects a first element from the plurality of elements extracted in step S103 based on a predetermined selection condition for establishing the purpose of the event, for example. More specifically, the dialogue support device 10 selects a first element from the plurality of elements extracted in step S103 based on a predetermined selection condition for establishing special fraud, for example.

[0053] Next, the dialogue support device 10 generates a scenario related to special fraud (step S105) by inputting, for example, the first element selected in step S104 into the generation AI (model) for scenario generation. The dialogue support device 10 generates a scenario related to a predetermined event by inputting, for example, the first element selected in step S104 into the generation AI (model) for scenario generation. The generated scenario may be stored in, for example, the scenario information 32. After the execution of step S105, the scenario generation process shown in FIG. 7 ends.

[0054] Next, with reference to FIG. 8, the flow of the risk assessment process according to the present embodiment will be described. FIG. 8 is a flowchart showing an example of the flow of the risk assessment process according to the present embodiment. The risk assessment process shown in FIG. 8 may be executed, for example, triggered by a trainer starting a risk assessment of being deceived by special fraud by using the training tool as shown in FIG. 6 using the trainer terminal 100.

[0055] First, as shown in FIG. 8, the dialogue support device 10 acquires, for example, the speech information of the trainer (step S201). More specifically, the dialogue support device 10 acquires the speech information of the trainer by analyzing, for example, the voice data of the trainer via the trainer terminal 100 such as a telephone used by the trainer.

[0056] Next, the dialogue support device 10 generates, for example, the speech information of the virtual agent that dialogues with the trainer via the trainer terminal 100 (step S202). More specifically, the dialogue support device 10 incorporates information in the real world, such as the speech information of the trainer acquired in step S201, and performs a simulation of special fraud execution on the digital twin to generate the speech information of the fraudster, which is a virtual agent.

[0057] Next, the dialogue support device 10 outputs, for example, the speech information of the virtual agent generated in step S202 via the trainer terminal 100 (step S203). Note that through steps S201 to S203, a dialogue between the trainer and the virtual agent will be conducted. However, the interaction between the trainer and the virtual agent may be repeated any number of times, and steps S201 to S203 may also be repeated accordingly.

[0058] Next, the dialogue support device 10 analyzes the mental state of the trainer using, for example, the physiological reaction of the trainer (step S204). More specifically, the dialogue support device 10 acquires, as the physiological reaction of the trainer, the vital information of the trainer in a predetermined section where a dialogue is being executed between the virtual agent and the trainer using, for example, a sensing device or the like, and estimates and analyzes the mental state by AI. Note that the estimation of the mental state by AI may be executed by inputting the vital information of the trainer into a machine learning model generated by machine learning using, for example, the vital information of a person as a feature amount and the mental state of the person as a correct label.

[0059] Next, the dialogue support device 10 evaluates the risk of the trainer being deceived by special fraud using, for example, the mental state analyzed in step S204 (step S205). The risk evaluation may be executed by inputting trainer information into a machine learning model generated by machine learning using, for example, the mental state of a person as a feature amount and the risk of the person being deceived by special fraud or the risk of an event such as special fraud of the person as a correct label. After the execution of step S205, the risk evaluation process shown in FIG. 8 ends.

[0060] Next, with reference to FIG. 9, the flow of the virtual agent selection process according to the present embodiment will be described. FIG. 9 is a flowchart showing an example of the flow of the virtual agent selection process according to the present embodiment.

[0061] First, the dialogue support device 10 generates, for example, a plurality of virtual agents on the digital twin (step S301). The dialogue support device 10 generates, for example, a plurality of fraudsters in the virtual space associated with the scenario on the digital twin.

[0062] Next, the dialogue support device 10 displays, for example, the profile information of each of the plurality of virtual agents on the trainer terminal 100 (step S302). The dialogue support device 10 displays, for example, the modus operandi of each of the plurality of fraudsters.

[0063] Next, the dialogue support device 10 identifies, for example, one virtual agent selected by the trainer from the trainer terminal 100 (step S303). The dialogue support device 10 identifies, for example, one first fraudster selected by the trainer.

[0064] Next, the dialogue support device 10 sets, for example, the scenario of the selected one virtual agent (step S304). The dialogue support device 10 sets, for example, the scenario of one first fraudster.

[0065] Next, the dialogue support device 10 generates, for example, the utterance information of one virtual agent based on the scenario (step S305). More specifically, the dialogue support device 10 generates, for example, the utterance information of the selected one virtual agent as a response according to the trainer's utterance content so as to achieve the purpose of the event based on the set scenario. The dialogue support device 10 generates, for example, the utterance information of one first fraudster as a response according to the trainer's utterance content so that the trainer pays money based on the scenario.

[0066] [Effect] The dialogue support device 10 executes a control program executed by a dialogue support device that supports a dialogue between a virtual agent indicating a special fraud perpetrator and a trainer for special fraud. By using the damage information of special fraud, a scenario related to special fraud is generated, and the generated scenario is output to the trainer terminal 100 used by the trainer as the speech information of the virtual agent.

[0067] In this way, the dialogue support device 10 outputs the scenario generated using the damage information of special fraud to the trainer terminal 100 as the speech information of the virtual agent, and by causing the trainer and the virtual agent to have a dialogue, the trainer can be effectively trained. In addition, training can be performed according to the latest scenario of special fraud. In special fraud, when countermeasures against the modus operandi of special fraud are taken, fraudsters will try to deceive the elderly and devise new modus operandi of special fraud one after another. In this case, the dialogue support device 10 can efficiently train the trainer based on the latest modus operandi of special fraud.

[0068] The dialogue support device 10 analyzes the voice data of the telephone used by the trainer to obtain the speech information of the trainer, and uses a machine learning model trained with information on the modus operandi extracted from the damage information to generate the speech information of the virtual agent for the obtained speech information of the trainer, and outputs the voice data of the generated speech information of the virtual agent to the telephone.

[0069] Thereby, the dialogue support device 10 can effectively train the trainer. In addition, training can be performed according to the latest scenario of special fraud.

[0070] In addition, the dialogue support device 10 analyzes the mental state of the trainer using the physiological reaction of the trainer in a predetermined section where the dialogue between the virtual agent and the trainer is being executed, and evaluates the risk of the trainer being deceived by special fraud using the analyzed mental state.

[0071] As a result, the dialogue support device 10 can analyze mental states such as overconfidence and panic, and evaluate the risk of being caught in a special fraud for the events of the trainee. In addition, training can be conducted in accordance with the latest scenarios of special fraud. Further, the dialogue support device 10 can make the trainee recognize the risk of being deceived by special fraud.

[0072] In addition, the process of generating a scenario executed by the dialogue support device 10 includes a process of selecting a first element from a plurality of elements constituting special fraud in the damage information based on a predetermined selection condition, and generating a scenario using the first element.

[0073] As a result, the dialogue support device 10 can conduct training in accordance with the latest scenarios of special fraud.

[0074] In addition, the process of selecting the first element executed by the dialogue support device 10 includes a process of analyzing the damage information, extracting a plurality of character strings constituting special fraud from the damage information as a plurality of elements, and selecting the first element from the plurality of elements based on a predetermined selection condition for establishing special fraud.

[0075] As a result, the dialogue support device 10 can conduct training in accordance with the latest scenarios of special fraud.

[0076] In addition, the process of generating a scenario using the first element executed by the dialogue support device 10 includes a process of generating a scenario by inputting the first element into a natural language processing model trained with an existing scenario related to special fraud.

[0077] As a result, the dialogue support device 10 can conduct training in accordance with the latest scenarios of special fraud.

[0078] In addition, the dialogue support device 10 acquires damage information from a predetermined database, and classifies the acquired damage information for each similar case using a clustering algorithm.

[0079] As a result, the dialogue support device 10 can perform training in accordance with the latest scenario of special fraud.

[0080] In addition, the process of selecting the first element, which is executed by the dialogue support device 10, is a process of selecting, as the first element, a character string related to at least one of the person who made the phone call for special fraud and their title, the person who received the phone call, the purpose of the phone call, the conditions for achieving the purpose, and the background of the purpose, from a plurality of elements that are a plurality of character strings constituting special fraud, based on a predetermined selection condition for establishing special fraud.

[0081] As a result, the dialogue support device 10 can perform training in accordance with the latest scenario of special fraud.

[0082] In addition, the dialogue support device 10 uses the actual damage situation of the real world against special fraud to generate a plurality of fraudsters in the virtual space associated with the scenario on the digital twin, displays the profile information of each of the plurality of fraudsters, identifies one first fraudster selected by the trainer from among the plurality of fraudsters in the virtual space, and outputs the voice data of the speech information of the first fraudster generated using at least the first fraudster and the scenario to the terminal used by the trainer.

[0083] As a result, the dialogue support device 10 can perform training in accordance with the latest scenario of special fraud.

[0084] In addition, the dialogue support device 10 is a control program executed by a dialogue support device that supports dialogue between a virtual agent indicating the executor of an event and a trainer for the event. Using the occurrence information of the event, it generates a scenario related to the event and outputs the generated scenario to the terminal used by the trainer as the speech information of the virtual agent.

[0085] As a result, the dialogue support device 10 can provide effective training to the trainer. In addition, it can conduct training in accordance with the latest scenarios of special fraud. In events such as special fraud, when countermeasures against the modus operandi of the event are taken, new modus operandi of subsequent events will emerge one after another. In this case, the dialogue support device 10 can provide efficient training to the trainer based on the latest modus operandi of the event.

[0086] In addition, the dialogue support device 10 analyzes the voice data of the telephone used by the trainer to obtain the speech information of the trainer, and uses a machine learning model trained with the performance information of past events to generate the speech information of the virtual agent for the obtained speech information of the trainer, and outputs the voice data of the generated speech information of the virtual agent to the telephone.

[0087] As a result, the dialogue support device 10 can provide effective training to the trainer. In addition, it can conduct training in accordance with the latest scenarios of special fraud.

[0088] In addition, the dialogue support device 10 analyzes the mental state of the trainer using the physiological reaction of the trainer in a predetermined section where the dialogue executed between the virtual agent and the trainer is taking place, and evaluates the risk of the trainer against the event using the analyzed mental state.

[0089] As a result, the dialogue support device 10 can analyze mental states such as overconfidence and panic, and evaluate the risk of the trainer being caught in an event such as special fraud against the event of the trainer. In addition, it can conduct training in accordance with the latest scenarios of special fraud. In addition, the dialogue support device 10 can make the trainer recognize the risk of being caught in an event such as special fraud.

[0090] [Other events] The trainer can have a conversation with a virtual agent representing the event executor via the trainer terminal 100 and can virtually experience events such as special fraud. In this case, the event can be applied to various events other than special fraud. For example, it can be applied to role-playing in call center operations. The dialogue support device 10 supports the dialogue between a virtual agent representing a customer with complaints or claims and the trainer for that customer. For example, the dialogue support device 10 generates a scenario regarding complaints or claims using past customer complaint and claim information. Then, the dialogue support device 10 causes the generated scenario to be output to the terminal used by the trainer as the speech information of the virtual agent. Also, the dialogue support device 10 analyzes the mental state of the trainer using the physiological reaction of the trainer in a predetermined section where the dialogue being executed between the virtual agent and the trainer is taking place. And the dialogue support device 10 evaluates the risk of the trainer's customer response using the analyzed mental state. Thereby, by improving issues and problems, the accuracy of actual customer response can be enhanced.

[0091] [System] The processing procedures, control procedures, specific names, and information including various data and parameters shown in the above document and drawings may be arbitrarily changed unless otherwise specified. Also, the specific examples, distributions, numerical values, etc. described in the embodiments are merely examples and may be arbitrarily changed.

[0092] Also, the specific forms of the dispersion and integration of the components of each device are not limited to those shown. That is, all or part of the components may be functionally or physically dispersed or integrated in any unit according to various loads, usage situations, etc. Furthermore, each processing function of each device may be realized in whole or in any part by a CPU (Central Processing Unit) and a program analyzed and executed by the CPU, or may be realized as hardware by wired logic.

[0093] [Hardware] FIG. 10 is a diagram for explaining a hardware configuration example of the dialogue support device 10. As shown in FIG. 10, the dialogue support device 10 includes a communication interface 10a, a HDD (Hard Disk Drive) 10b, a memory 10c, and a processor 10d. Further, each part shown in FIG. 10 is interconnected by a bus or the like.

[0094] The communication interface 10a is a network interface card or the like and communicates with other servers. The HDD 10b stores programs and data for operating the functions shown in FIG. 2.

[0095] The processor 10d is a hardware circuit that operates a process for executing each function described in FIG. 2 and the like by reading a program that executes the same processing as each processing unit shown in FIG. 2 from the HDD 10b or the like and expanding it in the memory 10c. That is, this process executes the same functions as each processing unit of the dialogue support device 10. Specifically, the processor 10d reads a program having the same functions as the acquisition unit 41, the generation unit 42, the output unit 43, and the evaluation unit 44 from the HDD 10b or the like. Then, the processor 10d executes a process that executes the same processing as the acquisition unit 41, the generation unit 42, the output unit 43, and the evaluation unit 44.

[0096] In this way, the dialogue support device 10 operates as an information processing device that executes an operation control process by reading and executing a program that executes the same processing as each processing unit shown in FIG. 2. Further, the dialogue support device 10 can also realize the same functions as the above-described embodiments by reading a program from a recording medium by a medium reading device and executing the read program. Note that the program in this other embodiment is not limited to being executed by the dialogue support device 10. For example, the present embodiment may be similarly applied when another information processing device executes the program, or when the dialogue support device 10 and another information processing device cooperate to execute the program.

[0097] In addition, a program that executes the same processing as each processing unit shown in FIG. 2 can be distributed via a network such as the Internet. Further, this program is recorded on a computer-readable recording medium such as a hard disk, a flexible disk (FD), a CD-ROM, a MO (Magneto-Optical disk), a DVD (Digital Versatile Disc), and can be executed by being read from the recording medium by a computer.

[0098] Regarding the embodiments including the above embodiments, the following additional notes are further disclosed.

[0099] (Supplementary Note 1) A control program executed by a dialogue support device that supports dialogue between a virtual agent indicating a special fraud perpetrator and a trainer for the special fraud, generating a scenario related to the special fraud using the damage information of the special fraud, causing the generated scenario to be output to a terminal used by the trainer as speech information of the virtual agent. A control program characterized by causing the dialogue support device to execute the processing.

[0100] (Supplementary Note 2) By analyzing the voice data of a telephone used by the trainer, obtaining the speech information of the trainer, generating speech information of the virtual agent for the obtained speech information of the trainer using a machine learning model trained with information on the modus operandi extracted from the damage information, causing the voice data of the generated speech information of the virtual agent to be output to the telephone. The control program according to Supplementary Note 1, characterized by causing the dialogue support device to execute the processing.

[0101] (Supplementary Note 3) Analyzing the mental state of the trainer using the physiological reaction of the trainer in a predetermined section where dialogue is being executed between the virtual agent and the trainer, Using the analyzed mental state, evaluate the risk of the trainer being deceived by special fraud A control program according to appended claim 1, characterized in that the dialogue support device is caused to execute the processing

[0102] (Appended claim 4) The process of generating the scenario is Based on predetermined selection conditions, select a first element from a plurality of elements constituting the special fraud in the damage information Generate the scenario using the first element A control program according to appended claim 1, characterized in that the process is included

[0103] (Appended claim 5) The process of selecting the first element is By analyzing the damage information, extract a plurality of character strings constituting the special fraud from the damage information as the plurality of elements Based on the predetermined selection conditions for establishing the special fraud, select the first element from the plurality of elements A control program according to appended claim 4, characterized in that the process is included

[0104] (Appended claim 6) The process of generating the scenario using the first element is Generate the scenario by inputting the first element into a natural language processing model trained with an existing scenario related to the special fraud A control program according to appended claim 5, characterized in that the process is included

[0105] (Appended claim 7) Obtain the damage information from a predetermined database Classify the obtained damage information for each similar case using a clustering algorithm A control program according to appended claim 5 or 6, characterized in that the dialogue support device is caused to execute the processing

[0106] (Appended claim 8) The process of selecting the first element is Based on the predetermined selection conditions for establishing the special fraud, from the plurality of elements that are the plurality of character strings constituting the special fraud, a string related to at least one of the person who made the phone call for the special fraud and their title, the person to whom the phone call was made, the purpose of the phone call, the conditions for achieving the purpose, and the background of the purpose is selected as the first element. The control program according to appendix 5 or 6, characterized by including the process.

[0107] (Appendix 9) Using the real-world damage situation for the special fraud, generate a plurality of fraudsters in the virtual space associated with the scenario on the digital twin, Display the profile information of each of the plurality of fraudsters, Identify one first fraudster selected by the trainer from the plurality of fraudsters in the virtual space, Output the voice data of the speech information of the first fraudster generated using at least the first fraudster and the scenario to the terminal used by the trainer. The control program according to appendix 1, characterized by causing the dialogue support device to execute the process.

[0108] (Appendix 10) A control program executed by a dialogue support device that supports the dialogue between a virtual agent indicating the executor of an event and a trainer for the event, Generate a scenario related to the event using the occurrence information of the event, Output the generated scenario to the terminal used by the trainer as the speech information of the virtual agent. A control program characterized by causing the dialogue support device to execute the process.

[0109] (Appendix 11) By analyzing the voice data of the telephone used by the trainer, obtain the speech information of the trainer, Generate the speech information of the virtual agent for the obtained speech information of the trainer using a machine learning model trained with the performance information of the event that occurred in the past. Cause the telephone to output the voice data of the speech information of the generated virtual agent. A control program according to appended note 10, characterized in that the processing is caused to be executed by the dialogue support device.

[0110] (Appended note 12) Analyze the mental state of the trainer using the physiological reaction of the trainer in a predetermined section where a dialogue is being executed between the virtual agent and the trainer, Evaluate the risk of the trainer against the event using the analyzed mental state. A control program according to appended note 10 or 11, characterized in that the processing is caused to be executed by the dialogue support device.

[0111] (Appended note 13) A control method executed by a dialogue support device that supports a dialogue between a virtual agent indicating a special fraud perpetrator and a trainer against the special fraud, Generate a scenario related to the special fraud using the damage information of the special fraud, Cause the generated scenario to be output to the terminal used by the trainer as the speech information of the virtual agent. A control method characterized by executing the processing.

[0112] (Appended note 14) Obtain the speech information of the trainer by analyzing the voice data of the telephone used by the trainer, Generate the speech information of the virtual agent for the obtained speech information of the trainer using a machine learning model trained with information on the modus operandi extracted from the damage information, Cause the telephone to output the voice data of the generated speech information of the virtual agent. A control method according to appended note 13, characterized in that the processing is executed by the dialogue support device.

[0113] (Appended note 15) Analyze the mental state of the trainer using the physiological reaction of the trainer in a predetermined section where a dialogue is being executed between the virtual agent and the trainer, Using the analyzed mental state, evaluate the risk of the trainer being deceived by special fraud The control method according to appended note 13, characterized in that the dialogue support device executes the process

[0114] (Appended note 16) The process of generating the scenario is Based on predetermined selection conditions, select a first element from a plurality of elements constituting the special fraud in the damage information Generate the scenario using the first element The control method according to appended note 13, characterized by including the process

[0115] (Appended note 17) The process of selecting the first element is By analyzing the damage information, extract a plurality of character strings constituting the special fraud from the damage information as the plurality of elements Select the first element from the plurality of elements based on the predetermined selection conditions for establishing the special fraud The control method according to appended note 16, characterized by including the process

[0116] (Appended note 18) The process of generating the scenario using the first element is Input the first element into a natural language processing model trained with an existing scenario related to the special fraud to generate the scenario The control method according to appended note 17, characterized by including the process

[0117] (Appended note 19) Obtain the damage information from a predetermined database Classify the obtained damage information for each similar case using a clustering algorithm The control method according to appended note 17 or 18, characterized in that the dialogue support device executes the process

[0118] (Appended note 20) The process of selecting the first element is Based on the predetermined selection conditions for establishing the special fraud, from the plurality of elements that are a plurality of character strings constituting the special fraud, a string related to at least one of the person who made the phone call for the special fraud, their title, the person who received the phone call, the purpose of the phone call, the conditions for achieving the purpose, and the background of the purpose is selected as the first element. The control method according to appendix 17 or 18, characterized by including the process.

[0119] (Appendix 21) Using the real-world damage situation for the special fraud, on the digital twin, a plurality of fraudsters in the virtual space associated with the scenario are generated. The profile information of each of the plurality of fraudsters is displayed. One first fraudster selected by the trainer is identified from among the plurality of fraudsters in the virtual space. The voice data of the speech information of the first fraudster generated using at least the first fraudster and the scenario is output to the terminal used by the trainer. The control method according to appendix 13, characterized in that the process is executed by the dialogue support device.

[0120] (Appendix 22) A control method executed by a dialogue support device that supports the dialogue between a virtual agent indicating the executor of an event and a trainer for the event, Using the occurrence information of the event, a scenario related to the event is generated. The generated scenario is output to the terminal used by the trainer as the speech information of the virtual agent. The control method characterized in that the process is executed by the dialogue support device.

[0121] (Appendix 23) By analyzing the voice data of the telephone used by the trainer, the speech information of the trainer is obtained. Using a machine learning model trained with the performance information of the events that occurred in the past, the speech information of the virtual agent for the obtained speech information of the trainer is generated. Causing the telephone set to output voice data of the speech information of the generated virtual agent The control method according to supplementary note 22, wherein the dialogue support device executes the process

[0122] (Supplementary note 24) Analyzing the mental state of the trainer using the physiological reaction of the trainer in a predetermined section where a dialogue is being executed between the virtual agent and the trainer Evaluating the risk of the trainer with respect to the event using the analyzed mental state The control method according to supplementary note 22 or 23, wherein the dialogue support device executes the process

[0123] (Supplementary note 25) A dialogue support device for supporting a dialogue between a virtual agent indicating a perpetrator of special fraud and a trainer for the special fraud, comprising a processor and A memory operably connected to the processor The processor generates a scenario related to the special fraud using the damage information of the special fraud and Causes the generated scenario to be output to a terminal used by the trainer as speech information of the virtual agent And executes a process A dialogue support device characterized by

[0124] (Supplementary note 26) A dialogue support device for supporting a dialogue between a virtual agent indicating a perpetrator of an event and a trainer for the event, comprising a processor and A memory operably connected to the processor The processor generates a scenario related to the event using the occurrence information of the event and Causes the generated scenario to be output to a terminal used by the trainer as speech information of the virtual agent And executes a process A dialogue support device characterized by

Explanation of reference numerals

[0125] 1 Information Processing System 10 Dialogue Support Device 10a Communication Interface 10b HDD 10c Memory 10d Processor 20 Communication Unit 30 Storage Unit 31 Event Generation Information 32 Scenario Information 33 Model Information 34 Digital Twin Information 40 Control Unit 41 Acquisition Unit 42 Generation Unit 43 Output Unit 44 Evaluation Unit 50 Network 100 Trainer Terminal

Claims

1. A control program executed by a dialogue support device that supports a dialogue between a virtual agent representing a perpetrator of a specialized fraud and a trainer for the specialized fraud, comprising: Using the damage information of the specialized fraud, a scenario regarding the specialized fraud is generated; The generated scenario is output to a terminal used by the trainee as utterance information of the virtual agent. A control program for causing the dialogue support device to execute a process.

2. acquiring speech information of the trainee by analyzing voice data of the telephone used by the trainee; generating utterance information of the virtual agent in response to the acquired utterance information of the trainee using a machine learning model trained with information on the modus operandi extracted from the damage information; The generated voice data of the speech information of the virtual agent is output to the telephone.

2. The control program according to claim 1, wherein the control program causes the dialogue support device to execute a process.

3. analyzing a psychological state of the trainee using a physiological response of the trainee during a predetermined section during which a dialogue is being executed between the virtual agent and the trainee; Using the analyzed psychological state, the risk of the trainee being deceived by a specialized fraud is evaluated.

2. The control program according to claim 1, wherein the control program causes the dialogue support device to execute a process.

4. The process of generating a scenario includes: Selecting a first element from a plurality of elements constituting the specialized fraud in the victim information based on a predetermined selection condition; The scenario is generated using the first element.

2. The control program according to claim 1, further comprising a process for:

5. The process of selecting the first element includes: By analyzing the damage information, a plurality of character strings constituting the special fraud are extracted as the plurality of elements from the damage information; Selecting the first element from the plurality of elements based on the predetermined selection condition for establishing the specialized fraud.

5. The control program according to claim 4, further comprising a process for:

6. The process of generating the scenario using the first element includes: The scenario is generated by inputting the first element into a natural language processing model trained with an existing scenario related to the specialized fraud.

6. The control program according to claim 5, further comprising a process for:

7. The damage information is acquired from a predetermined database; The acquired damage information is classified into similar cases using a clustering algorithm.

7. The control program according to claim 5, wherein the control program causes the dialogue support device to execute a process.

8. The process of selecting the first element includes: Based on the predetermined selection condition for establishing the specialized fraud, a character string related to at least one of a person who made a call for the specialized fraud and his / her title, a person who received the call, a purpose of the call, a condition for achieving the purpose, and a background of the purpose is selected as the first element from the plurality of elements which are a plurality of character strings constituting the specialized fraud.

7. The control program according to claim 5, further comprising a process for:

9. Using the real-world damage situation for the specialized fraud, a plurality of fraudsters in a virtual space are generated in the digital twin, which are associated with the scenario; displaying profile information for each of said plurality of fraudsters; Identifying a first impostor selected by a trainer from among a plurality of impostors in the virtual space; outputting voice data of speech information of the first fraudster, which is generated using at least the first fraudster and the scenario, to a terminal used by the trainee; 2. The control program according to claim 1, wherein the control program causes the dialogue support device to execute a process.

10. A control program executed by a dialogue support device that supports a dialogue between a virtual agent representing an event performer and a trainee for the event, generating a scenario relating to the event using occurrence information of the event; The generated scenario is output to a terminal used by the trainee as utterance information of the virtual agent. A control program for causing the dialogue support device to execute a process.

11. acquiring speech information of the trainee by analyzing voice data of the telephone used by the trainee; generating utterance information for the virtual agent in response to the acquired utterance information of the trainee using a machine learning model trained on performance information of the event that has occurred in the past; The generated voice data of the speech information of the virtual agent is output to the telephone.

11. The control program according to claim 10, which causes the dialogue support device to execute a process.

12. analyzing a psychological state of the trainee using a physiological response of the trainee during a predetermined section during which a dialogue is being executed between the virtual agent and the trainee; Using the analyzed psychological state to assess the trainee's risk for the event.

12. The control program according to claim 10, wherein the control program causes the dialogue support device to execute a process.

13. A control method executed by a dialogue support device that supports a dialogue between a virtual agent representing a perpetrator of a specialized fraud and a trainer for the specialized fraud, comprising: Using the damage information of the specialized fraud, a scenario regarding the specialized fraud is generated; The generated scenario is output to a terminal used by the trainee as utterance information of the virtual agent. A control method comprising: executing a process.

14. A control method executed by a dialogue support device that supports a dialogue between a virtual agent representing an actor of an event and a trainee for the event, comprising: generating a scenario relating to the event using occurrence information of the event; The generated scenario is output to a terminal used by the trainee as utterance information of the virtual agent. A control method comprising: executing a process.

Images