Dialog device, dialog method, and dialog program

Abstract

This dialog device comprises an input unit, an action selection unit, an utterance generation unit, and an output unit. The input unit receives an input of an utterance of a user. On the basis of the input utterance of the user, profile information about the user, and the context of a dialog with the user, the action selection unit selects an action for the user by calling functions by means of a natural language processing model and using a result of calling, from among predefined functions, a function necessary for an action for maintaining the context of the dialog with the user. Thereafter, the utterance generation unit uses the natural language processing model to generate and output an utterance necessary for the action on the basis of the action selected by the action selection unit. The output unit outputs a log of the dialog with the user, which includes an utterance serving as a response from the user.

Description

Dialogue device, dialogue method, and dialogue program

[0001] The present invention relates to a dialogue device, a dialogue method, and a dialogue program.

[0002] Conventionally, methods for analyzing a user's personality from text data or the like include, for example, a questionnaire-based method and a method using social media data.

[0003] However, questionnaire-based methods are prone to bias in user responses, and methods using social media data have the problem of being dependent on the quality and quantity of data, limiting the accuracy of analysis (see Non-Patent Document 1).

[0004] Furthermore, personality analysis through dialogue with experts is highly accurate but has the drawback of being expensive, and interviews with experts do not result in natural dialogue, and users have a high psychological hurdle when undergoing a medical interview.

[0005] For this reason, medical diagnosis using expert systems, which are programs that present questions to experts, and task-oriented dialogue, which generates dialogue by filling in slots according to the purpose, have been studied (see Non-Patent Document 2).

[0006] Sanja Stajner et al., A Survey of Automatic Personality Detection from Texts, [online], [Retrieved August 19, 2024], Internet <URL: https: / / aclanthology.org / 2020.coling-main.553 / > Libo Qin et al., End-to-end Task-oriented Dialogue: A Survey of Tasks, Methods, and Future Directions, [online], [Retrieved August 19, 2024], Internet <URL: https: / / arxiv.org / abs / 2311.09008>

[0007] However, when analyzing a user's personality through dialogue with the user, there are many points to consider, such as how to ask questions that will keep the user engaged in the dialogue and extract information from the user that is free from self-perception bias. Therefore, in addition to conventional systems, it is necessary to dynamically adjust the dialogue strategy according to the user's situation during the dialogue (e.g., the user's emotions and reactions) and effectively collect user information.

[0008] However, as described above, there has been no conventional technology for dynamically adjusting a dialogue strategy and effectively collecting user information while taking into account the user's situation during dialogue.

[0009] Therefore, an object of the present invention is to solve the above-mentioned problems, to dynamically adjust a dialogue strategy while taking into account the user's situation during dialogue, and to effectively collect user information.

[0010] In order to solve the above-mentioned problems, the present invention is characterized by comprising an input unit that accepts input of a user's utterances; an action selection unit that calls a function from a natural language processing model based on the user's utterances, the user's profile information, and the context of the dialogue with the user, and selects an action for the user using the result of calling the function; an utterance generation unit that uses a natural language processing model to generate an utterance necessary for the action based on the selected action, and outputs the utterance to the user; and an output unit that outputs a dialogue log including the utterances from the user and the utterances output to the user.

[0011] According to the present invention, it is possible to dynamically adjust a dialogue strategy while taking into consideration the user's situation during dialogue, and to effectively collect user information.

[0012] Fig. 1 is a diagram for explaining an overview of the dialogue device. Fig. 2 is a diagram showing an example of the configuration of the dialogue device. Fig. 3 is a diagram showing an example of user profile information stored in a user information DB. Fig. 4 is a diagram showing an example of a schema of a function used by the dialogue device. Fig. 5 is a flowchart showing an example of a processing procedure executed by the dialogue device. Fig. 6 is a diagram showing an example of a computer that executes a dialogue program.

[0013] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a description will be given of an embodiment of the present invention with reference to the drawings, but the present invention is not limited to the embodiment.

[0014] [Overview] An overview of the dialogue device 10 of this embodiment will be described. The dialogue device 10 collects and outputs information about a user through natural dialogue with the user. During the dialogue with the user, the dialogue device 10 dynamically analyzes the user's emotions and the context of the dialogue with the user, and generates and outputs a response that encourages the user to continue the dialogue.

[0015] The dialogue device 10 takes the following into consideration when generating a response to the user: (1) Estimation of the user's situation: The dialogue device 10 estimates the user's emotional and psychological states from the content of the user's statements, the language used, the speed of the response, etc. (2) Utilization of already collected user profile information: The dialogue device 10 generates the next question and response to the user based on the user's profile information obtained in the course of the dialogue with the user. (3) Consideration of dialogue context: The dialogue device 10 analyzes the flow of the dialogue with the user and the user's past utterances, and generates the next question and response to the user so as to elicit information about the user's personality while maintaining the dialogue context.

[0016] By taking the above points into consideration, the dialogue device 10 can naturally and efficiently collect information about the user through dialogue with the user.

[0017] For example, when the dialogue device 10 receives input of multimodal information such as speech, images, and audio when a user speaks, it analyzes the flow of the dialogue with the user and the user's past utterances and selects an action for the user that maintains the context of the dialogue (action selection). Furthermore, if the dialogue device 10 needs to update the user's profile information based on the content of the user's utterance, it updates the user's profile information in a user information database (DB) (data editing). Then, the dialogue device 10 generates an utterance required for the selected action and outputs it to the user (utterance generation).

[0018] The interactive device 10 repeats the above process until the interaction with the user ends, and when the interaction ends, outputs information about the user (for example, the user's profile information and interaction log).

[0019] [Configuration Example] Next, a configuration example of the dialogue device 10 will be described with reference to Fig. 2. The dialogue device 10 includes, for example, an input / output unit 11, a storage unit 12, and a control unit 13. Note that, in the following, a case will be described in which the natural language processing model used by the dialogue device 10 is LLM (Large Language Models), but is not limited to this.

[0020] The input / output unit 11 is an interface that controls the input and output of various data. The input / output unit 11 receives, for example, input of a user's utterance (user's utterance information). The input / output unit 11 also outputs a dialogue log obtained through a dialogue with the user, user profile information, and the like.

[0021] The storage unit 12 stores data, programs, etc. referenced when the control unit 14 executes various processes. The storage unit 12 is realized by a semiconductor memory element such as a random access memory (RAM) or a flash memory, or a storage device such as a hard disk or an optical disk. For example, the storage unit 12 stores a function schema used by the LLM for function calls (details will be described later). The storage unit 12 also includes a user information DB that stores user information (user profile information).

[0022] [User Information DB] The user information DB is a DB that stores user profile information collected through dialogue with the user. The control unit 13 reads out the user profile information from the user information DB as needed and uses it to select an action to be taken by the user (action selection).

[0023] As shown in FIG. 3, the user profile information includes, for example, the user name, information indicating the user's attributes, and other information (e.g., notes). Information indicating the user's attributes is, for example, information indicating the user's age, occupation, gender, region, etc. The data format of the user information DB can be any format, such as a relational database, a JSON file format, or a text file. The user profile information in the user information DB is added and updated as appropriate by the control unit 13.

[0024] Returning to the description of Fig. 2, the control unit 13 is responsible for overall control of the dialogue device 10. The functions of the control unit 13 are realized, for example, by a CPU (Central Processing Unit) executing a program stored in the storage unit 12.

[0025] The control unit 13 includes an utterance input unit 131, an action selection unit 132, a data editing unit 133, an utterance generation unit 134, and an output unit 135. The reinforcement learning unit 136 may or may not be provided, and the cases where it is provided will be described later.

[0026] [Utterance Input Unit] The utterance input unit 131 accepts input of user utterance information. For example, the utterance input unit 131 accepts input of multimodal information such as text, images, videos, and audio of the user's utterance as the user's utterance information.

[0027] [Action Selection Unit] The action selection unit 132 selects an action appropriate for the dialogue with the user by calling a function using the LLM, based on the user's speech information output from the speech input unit 131, the user's profile information read from the user information DB, the context of the dialogue with the user, etc.

[0028] For example, the LLM used by the behavior selection unit 132 calls a function from predefined functions required for behavior to maintain the context of the dialogue with the user, and uses the result of the function call to select a behavior appropriate for the dialogue with the user.The behavior selection unit 132 then outputs the selected behavior to the utterance generation unit 134.

[0029] In this embodiment, the selection of an action in the action selection unit 132 is implemented as a function call by an LLM, thereby enabling the definition of the function to enhance the controllability of the dialogue with the user. The content of the action includes, for example, writing, reading, and editing the user's profile information, utterances to draw the user's interest, utterances to ask the user questions, etc. The content of the action can be expanded depending on the use case.

[0030] An example of the schema of the above function is shown in Figure 4. The schema shown in Figure 4 shows an example of the schema for the function "read_user_profile" for reading user profile information.

[0031] The LLM used by the behavior selection unit 132 repeatedly selects a behavior while calling an appropriate function from among predefined functions according to the context of the user's dialogue. For example, the LLM calls a predefined prompt or an external API (Application Programming Interface) based on the called function, and repeatedly selects a behavior based on the results of the call.

[0032] Furthermore, the behavior selection unit 132 determines whether or not the user's profile information needs to be updated based on the input user utterances using the LLM. For example, if the behavior selection unit 132 determines that the user's profile information needs to be edited (e.g., read or write the user's profile information) based on the input user utterance information, the data editing unit 133 updates the user's profile information based on the user's utterance information.

[0033] For example, consider a case where the behavior selection unit 132 receives a user's utterance information such as "I recently had a stomachache and went to the hospital..." and determines that the profile information of the user needs to be updated. In this case, the data editing unit 133 writes a note such as "The user recently had a stomachache and went to the hospital" in the profile information of the user in the user information DB (see FIG. 3).

[0034] In this way, the dialogue device 10 continues the dialogue with the user and updates the user's profile information based on the information obtained through the dialogue. Thereafter, the dialogue device 10 refers to the updated user's profile information, for example, in Retrieval-Augmented Generation (RAG) when selecting an action for the user or generating utterances for the user. This allows the dialogue device 10 to select an appropriate action for the user and generate utterances.

[0035] Furthermore, the behavior selection unit 132 may select a behavior for the user by using the user's utterance information received from the utterance input unit 131, the user's profile information read from the user information DB, the context of the dialogue with the user, and the user's psychological state estimated based on the user's utterance information. The estimation of the user's psychological state based on the user's utterance information may be performed, for example, by the LLM used by the behavior selection unit 132, or may be performed by a tool other than the LLM. In this way, the dialogue device 10 can select a more appropriate behavior and generate utterances for the user.

[0036] [Utterance Generation Unit] The utterance generation unit 134 uses the LLM to generate and output an appropriate response to the user (utterance to the user) based on the action selected by the action selection unit 132. For example, the utterance generation unit 134 inputs the action selected by the action selection unit 132 and an instruction to generate an utterance required for the action to the LLM, and outputs the utterance output from the LLM.

[0037] For example, if the behavior selection unit 132 selects the behavior of chatting with the user, the utterance generation unit 134 inputs an instruction to the LLM to generate the utterance necessary for the behavior of chatting with the user, and outputs the utterance output from the LLM.

[0038] [Output Unit] The output unit 135 outputs a dialogue log with a user and updated profile information of the user. The dialogue log with a user is a log showing pairs of utterances by the user and utterances output by the utterance generation unit 134 in response to the utterances. Note that the above-mentioned utterances by the user include utterances made by the user in response to the utterances output by the utterance generation unit 134.

[0039] According to such a dialogue device 10, for example, it is possible to efficiently acquire information about the user (profile information about the user, a dialogue log including utterances by the user) to be used for analyzing the user's personality.

[0040] [Example of Processing Procedure] Next, an example of processing procedure executed by the dialogue device 10 will be described with reference to Fig. 5. First, the utterance input unit 131 of the dialogue device 10 accepts input of a user's utterance (utterance information) (S11). Next, the action selection unit 132 estimates the user's emotion using the user's utterance input in S11 (S12). Thereafter, the action selection unit 132 selects an appropriate action for dialogue with the user by function calling using the LLM based on the user's utterance information input in S1, the user's profile information read from the user information DB, the context of the dialogue with the user, etc. (S13: Selection of action for user).

[0041] Furthermore, the behavior selection unit 132 determines whether or not the user's profile information needs to be updated based on the user's utterance information input by the LLM (S14). If the behavior selection unit 132 determines that the user's profile information needs to be updated (Yes in S14), the data editing unit 133 updates the user's profile information in the user information DB (S15). Then, the process proceeds to S16. On the other hand, if the behavior selection unit 132 determines that the user's profile information does not need to be updated (No in S14), the process skips S15 and proceeds to S16.

[0042] The utterance generation unit 134 generates and outputs an appropriate response to the user's input (utterance to the user) using the LLM based on the action selected in S13 (S16: Generate and output utterance). Thereafter, when the dialogue device 10 determines that the dialogue with the user has ended (Yes in S17), the output unit 135 outputs the profile information of the user and the dialogue log (S18). On the other hand, when the dialogue device 10 determines that the dialogue with the user has not yet ended (No in S17), the process returns to S11.

[0043] By executing the above-described process, the dialogue device 10 can efficiently acquire information (profile information and dialogue log of the user) used for analyzing the user's personality, for example.

[0044] [Other Embodiments] The LLM used in the behavior selection unit 132 may be an LLM that has undergone imitation reinforcement learning using a dialogue log of a human interacting with another person, or self-reinforcement learning using a unique reward function.

[0045] For example, the LLM used in the behavior selection unit 132 may be an LLM that has undergone reinforcement learning to select behavior that corresponds to the context of the conversation with the user, using a conversation log that associates the user's past conversation information with at least one of the user's biometric information and emotions at the time of speaking as a learning dataset.

[0046] In this case, the dialogue device 10 may further include a reinforcement learning unit 136 (see FIG. 2) that performs reinforcement learning of the LLM. For example, a dialogue log annotated with information such as biometric information, emotions, and enjoyment level of the user at the time of utterance is prepared as a training dataset as a reward function for the reinforcement learning. The reinforcement learning unit 136 then performs reinforcement learning of the LLM using the training dataset.

[0047] In this way, the reinforcement learning 136 can cause the LLM to learn function call patterns for selecting actions suitable for the user, based on information such as the context of the dialogue with the user, the selected action, and the enjoyment of the user at that time. As a result, the action selection unit 132 can select actions that will continue the dialogue while entertaining the user, and will enable more efficient collection of information about the user.

[0048] In addition, like the behavior selection unit 132, the utterance generation unit 134 can also fine-tune the LLM used in the utterance generation unit 134 using a training dataset to generate utterances that are in line with the behavior selected by the user.

[0049] According to the dialogue device 10 described above, it is possible to efficiently acquire information (user profile information, dialogue log) used for analyzing the user's personality, etc., through natural interaction with the user.

[0050] [System Configuration, etc.] The components of each unit shown in the figure are conceptual functional units and do not necessarily have to be physically configured as shown. In other words, the specific form of distribution and integration of each device is not limited to that shown, and all or part of them can be functionally or physically distributed and integrated in any unit depending on various loads, usage conditions, etc. Furthermore, all or any part of the processing functions performed by each device can be realized by a CPU and a program executed by the CPU, or can be realized as hardware using wired logic.

[0051] Furthermore, among the processes described in the above embodiments, all or part of the processes described as being performed automatically can be performed manually, or all or part of the processes described as being performed manually can be performed automatically using a known method.In addition, the information including the processing procedures, control procedures, specific names, various data and parameters shown in the above documents and drawings can be changed as desired unless otherwise specified.

[0052] [Program] The above-described interactive device 10 can be implemented by installing a program (interactive program) as package software or online software on a desired computer. For example, by executing the above-described program on an information processing device, the information processing device can function as the interactive device 10. The information processing device referred to here includes mobile communication terminals such as smartphones, mobile phones, and PHS (Personal Handyphone Systems), as well as terminals such as PDAs (Personal Digital Assistants).

[0053] 6 is a diagram showing an example of a computer that executes an interactive program. The computer 1000 includes, for example, a memory 1010 and a CPU 1020. The computer 1000 also includes a hard disk drive interface 1030, a disk drive interface 1040, a serial port interface 1050, a video adapter 1060, and a network interface 1070. These components are connected by a bus 1080.

[0054] The memory 1010 includes a read-only memory (ROM) 1011 and a random access memory (RAM) 1012. The ROM 1011 stores a boot program such as a basic input / output system (BIOS). The hard disk drive interface 1030 is connected to a hard disk drive 1090. The disk drive interface 1040 is connected to a disk drive 1100. A removable storage medium such as a magnetic disk or optical disk is inserted into the disk drive 1100. The serial port interface 1050 is connected to a mouse 1110 and a keyboard 1120, for example. The video adapter 1060 is connected to a display 1130, for example.

[0055] The hard disk drive 1090 stores, for example, an OS 1091, an application program 1092, a program module 1093, and program data 1094. That is, the programs that define the processes executed by the above-mentioned interactive device 10 are implemented as program modules 1093 in which computer-executable code is written. The program modules 1093 are stored, for example, in the hard disk drive 1090. For example, the program modules 1093 for executing processes similar to those of the functional configuration of the interactive device 10 are stored in the hard disk drive 1090. Note that the hard disk drive 1090 may be replaced by an SSD (Solid State Drive).

[0056] Data used in the processing of the above-described embodiment is stored as program data 1094, for example, in the memory 1010 or the hard disk drive 1090. The CPU 1020 then reads the program module 1093 or the program data 1094 stored in the memory 1010 or the hard disk drive 1090 into the RAM 1012 as necessary and executes them.

[0057] The program module 1093 and program data 1094 may not necessarily be stored in the hard disk drive 1090, but may also be stored in a removable storage medium and read by the CPU 1020 via the disk drive 1100 or the like. Alternatively, the program module 1093 and program data 1094 may be stored in another computer connected via a network (such as a local area network (LAN) or a wide area network (WAN)). The program module 1093 and program data 1094 may then be read by the CPU 1020 from the other computer via the network interface 1070.

[0058] REFERENCE SIGNS LIST 10 Dialogue device 11 Input / output unit 12 Memory unit 13 Control unit 131 Utterance input unit 132 Action selection unit 133 Data editing unit 134 Utterance generation unit 135 Output unit 136 Reinforcement learning unit

Claims

1. A dialogue device comprising: an input unit that accepts input of a user's utterance; an action selection unit that calls a function from a natural language processing model based on the user's utterance, the user's profile information, and the context of the dialogue with the user, by calling a function from among predefined functions that is necessary for an action to maintain the context of the dialogue with the user, and uses the result of calling the function to select an action for the user; an utterance generation unit that uses a natural language processing model based on the selected action to generate an utterance necessary for the action and output it to the user; and an output unit that outputs a dialogue log with the user that includes utterances that are responses from the user.

2. The interaction device described in claim 1, characterized in that the behavior selection unit further determines, using a natural language processing model, whether or not the user's profile information needs to be updated based on the user's input utterance, and the interaction device further includes a data editing unit that updates the user's profile information based on the user's utterance when the behavior selection unit determines that the user's profile information needs to be updated, and the output unit further outputs the updated user's profile information.

3. The dialogue device according to claim 1, characterized in that the natural language processing model used in the action selection unit is a natural language processing model that has undergone reinforcement learning to select actions that correspond to the context of the dialogue with the user, using a dialogue log that associates past dialogue information of the user with at least one of biometric information and emotions at the time of the user's speech as a learning dataset.

4. The dialogue device according to claim 1, characterized in that the action selection unit further selects an action for the user using the user's psychological state estimated based on the user's utterance.

5. A dialogue method executed by a dialogue device, comprising the steps of: accepting input of a user's utterance; calling a function from a natural language processing model based on the user's utterance, the user's profile information, and the context of the dialogue with the user, from among predefined functions, a function required for an action to maintain the context of the dialogue with the user, and using the result of calling the function to select an action for the user; generating an utterance required for the action using a natural language processing model based on the selected action, and outputting the utterance to the user; and outputting a dialogue log with the user including the utterance that is a response from the user.

6. A dialogue program for causing a computer to execute the following steps: accepting input of a user's utterance; calling a function from a natural language processing model based on the user's utterance, the user's profile information, and the context of the dialogue with the user, from among predefined functions, a function required for an action to maintain the context of the dialogue with the user, and using the result of calling the function to select an action for the user; generating an utterance required for the action using a natural language processing model based on the selected action, and outputting the utterance to the user; and outputting a dialogue log with the user including utterances that are responses from the user.

Images