Dialogue model adjustment assistance device, dialogue model adjustment assistance method, dialogue model adjustment assistance program, and recording medium

Abstract

Provided is a dialogue model adjustment assistance device capable of adjusting a dialogue model so as to provide dialogue individually adapted to an interlocutor.　A dialogue model adjustment assistance device according to the present disclosure includes a dialogue data acquisition unit, a dialogue type estimation unit, a response evaluation unit, an adjustment data recording unit, and a dialogue model adjustment unit. The dialogue data acquisition unit acquires dialogue data between an interlocutor and a dialogue model. The dialogue type estimation unit estimates a dialogue type of the interlocutor on the basis of the dialogue data, the dialogue type being information which defines a feature of an interlocutor who has dialogue with the dialogue model. The response evaluation unit evaluates response data in the dialogue data by using an evaluation standard and a weight for the evaluation standard corresponding to the dialogue type, and calculates an evaluation value. The adjustment data recording unit records the evaluation value and the dialogue data as dialogue model adjustment data for the interlocutor. The dialogue model adjustment unit adjusts, by using the dialogue model adjustment data, the dialogue model so as to perform dialogue adapted to the interlocutor.

Description

Dialogue model adjustment support apparatus, dialogue model adjustment support method, dialogue model adjustment support program, and recording medium

[0001] The present disclosure relates to a dialogue model adjustment support apparatus, a dialogue model adjustment support method, a dialogue model adjustment support program, and a recording medium.

[0002] In recent years, research has been conducted to construct dialogue systems such as chatbots using machine-learned models. For example, Patent Document 1 discloses that using an acquisition unit that acquires an inquiry sentence, a positive example learning response sentence for a learning inquiry sentence, and a negative example learning response sentence sampled according to the frequency distribution of the learning response sentences, similarity between the inquiry sentence for learning and the positive example learning response sentence is increased, and learning processing is performed such that similarity between the learning inquiry sentence and the negative example learning response sentence is decreased. A question answering apparatus is described, comprising: a calculation unit that calculates similarity between the inquiry sentence and a plurality of answer candidate sentences using a learned model generated by executing the learning processing; and an extraction unit that extracts one or more answer candidate sentences from the plurality of answer candidate sentences based on the similarity.

[0003] Japanese Unexamined Patent Publication No. 2021-124824

[0004] A interlocutor who conducts a dialogue with a dialogue system has an expected response style for the dialogue system that serves as the dialogue partner, but the invention as described in Patent Document 1 has a problem in that there is no mechanism for realizing dialogue individually adapted for each interlocutor.

[0005] Accordingly, an object of the present disclosure is to provide a dialogue model adjustment support apparatus, a dialogue model adjustment support method, a dialogue model adjustment support program, and a recording medium that enable adjustment of a dialogue model that realizes dialogue individually adapted to an interlocutor.

[0006] To achieve the above objective, the dialogue model adjustment support device of the present disclosure includes a dialogue data acquisition unit, a dialogue type estimation unit, a response evaluation unit, an adjustment data recording unit, and a dialogue model adjustment unit, wherein the dialogue data acquisition unit acquires dialogue data between a person and a dialogue model, the dialogue data includes at least one input data input by the person to the dialogue model and at least one response data output by the dialogue model based on the input data, the dialogue type estimation unit estimates the dialogue type of the person based on the dialogue data, the dialogue type is information defining the characteristics of the person who interacts with the dialogue model, the response evaluation unit evaluates the response data in the dialogue data using an evaluation criterion and the weight of the evaluation criterion corresponding to the dialogue type and calculates an evaluation value, the adjustment data recording unit records the evaluation value and the dialogue data as dialogue model adjustment data for the person, and the dialogue model adjustment unit adjusts the dialogue model using the dialogue model adjustment data to perform a dialogue adapted to the person.

[0007] The dialogue model adjustment support method of this disclosure includes a dialogue data acquisition step, a dialogue type estimation step, a response evaluation step, an adjustment data recording step, and a dialogue model adjustment step, wherein each step is performed by a computer. The dialogue data acquisition step acquires dialogue data between a person and a dialogue model, the dialogue data includes at least one input data input by the person to the dialogue model and at least one response data output by the dialogue model based on the input data, the dialogue type estimation step estimates the dialogue type of the person based on the dialogue data, the dialogue type is information that defines the characteristics of the person who interacts with the dialogue model, the response evaluation step evaluates the response data in the dialogue data using an evaluation criterion and the weight of the evaluation criterion corresponding to the dialogue type to calculate an evaluation value, the adjustment data recording step records the evaluation value and the dialogue data as dialogue model adjustment data for the person, and the dialogue model adjustment step adjusts the dialogue model using the dialogue model adjustment data to perform a dialogue adapted to the person.

[0008] The dialogue model adjustment support program of this disclosure is a program that causes a computer to execute each of the following procedures: a dialogue data acquisition procedure, a dialogue type estimation procedure, a response evaluation procedure, an adjustment data recording procedure, and a dialogue model adjustment procedure, wherein the dialogue data acquisition procedure acquires dialogue data between a person and a dialogue model, the dialogue data includes at least one input data input by the person to the dialogue model and at least one response data output by the dialogue model based on the input data, the dialogue type estimation procedure estimates the dialogue type of the person based on the dialogue data, the dialogue type is information that defines the characteristics of the person who interacts with the dialogue model, the response evaluation procedure evaluates the response data in the dialogue data using evaluation criteria and weights of evaluation criteria corresponding to the dialogue type to calculate an evaluation value, the adjustment data recording procedure records the evaluation value and the dialogue data as dialogue model adjustment data for the person, and the dialogue model adjustment procedure adjusts the dialogue model using the dialogue model adjustment data to perform a dialogue adapted to the person.

[0009] The recording medium of this disclosure is a computer-readable recording medium that records a dialogue model adjustment support program for causing a computer to execute each of the following procedures: a dialogue data acquisition procedure, a dialogue type estimation procedure, a response evaluation procedure, an adjustment data recording procedure, and a dialogue model adjustment procedure, wherein the dialogue data acquisition procedure acquires dialogue data between a person and a dialogue model, the dialogue data includes at least one input data input by the person to the dialogue model and at least one response data output by the dialogue model based on the input data, the dialogue type estimation procedure estimates the dialogue type of the person based on the dialogue data, the dialogue type is information defining the characteristics of a person who interacts with the dialogue model, the response evaluation procedure evaluates the response data in the dialogue data using evaluation criteria and weights of evaluation criteria corresponding to the dialogue type to calculate an evaluation value, the adjustment data recording procedure records the evaluation value and the dialogue data as dialogue model adjustment data for the person, and the dialogue model adjustment procedure adjusts the dialogue model using the dialogue model adjustment data to perform a dialogue adapted to the person.

[0010] According to this disclosure, it is possible to adjust the dialogue model to realize a dialogue that is individually adapted to the person speaking to it.

[0011] Figure 1 is a block diagram showing the configuration of an example of the dialogue model adjustment support device of this disclosure. Figure 2 is a block diagram showing an example of the hardware configuration of the dialogue model adjustment support device of this disclosure. Figure 3 is a flowchart showing an example of processing in the dialogue model adjustment support device of this disclosure.

[0012] Next, embodiments of the present disclosure will be described with reference to the drawings. The present disclosure is not limited to the following embodiments. In the following drawings, the same parts are denoted by the same reference numerals. Furthermore, unless otherwise specified, the descriptions of each embodiment can be used interchangeably with those of the others, and unless otherwise specified, the configurations of each embodiment can be combined.

[0013] [Embodiment 1] The dialogue model adjustment support device of this embodiment will be described with reference to Figure 1. Figure 1 is a block diagram showing the configuration of an example of the dialogue model adjustment support device 10 of this embodiment. As shown in Figure 1, the dialogue model adjustment support device 10 (hereinafter also referred to as "this device 10") includes a dialogue data acquisition unit 11, a dialogue type estimation unit 12, a response evaluation unit 13, an adjustment data recording unit 14, and a dialogue model adjustment unit 15. In addition, although not shown, this device 10 may also include, for example, an input unit, an output unit, a display unit and / or a storage unit.

[0014] The device 10 may be, for example, a single device including the aforementioned parts, or it may be a device in which the aforementioned parts can be connected via a communication network. Furthermore, the device 10 can be connected to an external device described later via a communication network. The communication network is not particularly limited and a known network can be used, for example, it may be wired or wireless. Examples of communication networks include the Internet, WWW (World Wide Web), telephone lines, LAN (Local Area Network), SAN (Storage Area Network), DTN (Delay Tolerant Networking), LPWA (Low Power Wide Area), L5G (Local 5G), etc. Examples of wireless communication include Wi-Fi®, Bluetooth®, Local 5G, LPWA, etc. The aforementioned wireless communication may be in the form of direct communication between devices (Ad Hoc communication), infrastructure communication, or indirect communication via an access point. The device 10 may, for example, be incorporated into a server as a system. The device 10 may also be, for example, a personal computer (PC, e.g., desktop or notebook), smartphone, tablet terminal, etc., on which the program disclosed herein is installed. Furthermore, the device 10 may be in the form of cloud computing or edge computing, for example, in which at least one of the aforementioned parts is on a server and the other parts are on a terminal.

[0015] Figure 2 illustrates a block diagram of the hardware configuration of the device 10. The device 10 includes, for example, a central processing unit 101, memory 102, bus 103, storage device 104, input device 105, output device 106, communication device (communication unit) 107, etc. Each part of the device 10 is interconnected via the bus 103 through its respective interface (I / F).

[0016] The central processing unit 101 operates in coordination with other components via controllers (system controller, I / O controller, etc.) and is responsible for the overall control of the device 10. In the device 10, the central processing unit 101 executes, for example, the program disclosed herein or other programs, and also reads and writes various types of information. Specifically, for example, the central processing unit 101 functions as a dialogue data acquisition unit 11, a dialogue type estimation unit 12, a response evaluation unit 13, an adjustment data recording unit 14, and a dialogue model adjustment unit 15. The device 10 may also include other arithmetic units such as a CPU, GPU (Graphics Processing Unit), APU (Accelerated Processing Unit), or a combination thereof.

[0017] Bus 103 can also be connected to external devices, for example. Examples of such external devices include external storage devices (external databases, etc.), electrocardiographs, printers, external input devices, external display devices, audio output devices such as speakers, external imaging devices such as cameras, and various sensors such as acceleration sensors, geomagnetic sensors, and direction sensors. The device 10 can be connected to an external network (the aforementioned communication network) by a communication device 107 connected to bus 103, for example, and can also be connected to other devices via the external network.

[0018] Memory 102 may be, for example, main memory. When the central processing unit 101 performs processing, memory 102 reads various operational programs, such as the program of this disclosure, stored in the storage device 104 (described later), and the central processing unit 101 receives data from memory 102 and executes the program. The main memory may be, for example, RAM (random access memory). Alternatively, memory 102 may be, for example, ROM (read-only memory).

[0019] The storage device 104 is also called an auxiliary storage device, for example, in relation to the main memory (primary memory). As described above, the storage device 104 stores an operating program including the program of this disclosure. The storage device 104 may be, for example, a combination of a recording medium and a drive for reading and writing to the recording medium. The recording medium is not particularly limited and may be internal or external, for example, an HD (hard disk), CD-ROM, CD-R, CD-RW, MO, DVD, flash memory, memory card, etc. The storage device 104 may be, for example, a hard disk drive (HDD) in which the recording medium and the drive are integrated, or a solid state drive (SSD). If the device 10 includes, for example, the storage device 104 functions as the storage unit.

[0020] In this device 10, the memory 102 and storage device 104 can also store various types of information, such as log information, information obtained from an external database (not shown) or external devices, information generated by this device 10, and information used by this device 10 when executing processing. At least some of the information may be stored on an external server other than the memory 102 and storage device 104, or it may be stored in a distributed manner across multiple terminals using blockchain technology or the like.

[0021] The device 10 further includes, for example, an input device 105 and an output device 106. The input device 105 may include, for example, a pointing device such as a touch panel, trackpad, or mouse; a keyboard; imaging means such as a camera or scanner; a card reader such as an IC card reader or magnetic card reader; an audio input means such as a microphone; and so on. The output device 106 may include, for example, a display device such as an LED display or liquid crystal display; an audio output device such as a speaker; a printer; and so on. In this embodiment 1, the input device 105 and the output device 106 are configured separately, but the input device 105 and the output device 106 may be configured as an integrated unit, such as a touch panel display.

[0022] Next, an example of the dialogue model adjustment support method of this embodiment will be described based on the flowchart in Figure 3. The dialogue model adjustment support method of this embodiment can be implemented as follows, for example, using the dialogue model adjustment support device 10 shown in Figures 1 and 2. Note that the dialogue model adjustment support method of this embodiment is not limited to the use of the dialogue model adjustment support device 10 shown in Figures 1 and 2.

[0023] First, prior to processing by the device 10, an interaction is performed between the dialogue model and its responder (dialogue participant). The content of the interaction is not particularly limited as long as it is an exchange between the dialogue model and the responder, and may be, for example, a text exchange such as a text chat, or a voice exchange such as a phone call. The dialogue model may be a model that has been fine-tuned using predetermined training data for a large-scale learning model. The large-scale learning model may be, for example, a machine learning model that has been trained using predetermined big data. The large-scale learning model may be, for example, a model that has been trained on big data of natural language (large-scale language model), a model that has been trained on big data of speech (large-scale speech model), or a model that has been trained on big data of images (large-scale image model).

[0024] In this disclosure, the dialogue model is centered on a large-scale learning model, which is commonly referred to as a "foundation model" in recent times. The foundation model is a machine learning model pre-trained on predetermined big data, and is not limited to large-scale language models (LLMs) that have learned natural language, but may also include large-scale speech models, large-scale image models, and multimodal models (such as visual language models) that handle language, images, speech, and video across different systems. Furthermore, a configuration may be adopted in which a small-scale language model (SLM) is placed on the terminal side to cooperate with the large-scale model on the cloud side, a configuration that includes search extension generation (RAG) using an external knowledge source, tool execution / function calls, agent-oriented control logic, etc. The provider of the large-scale language model is not particularly limited, and examples include, but is not limited to, various LLMs / multimodal models provided by OpenAI, Anthropique, Alphabet (Google), META, Microsoft, Cohere, Mistral, xAI, NEC Corporation, NTT, etc.

[0025] The dialogue model may be, for example, a model that is adjusted to provide responses according to the dialogue type of the interlocutor. The dialogue model may be, for example, a model adjusted by the device 10, or a model adjusted manually. If the dialogue model is a model adjusted by the device 10, the dialogue model adjustment unit 15 of the device 10 may, for example, set at least one dialogue type of the interlocutor in which the dialogue model is expected to interact prior to the dialogue between the interlocutor and the dialogue model, set at least one response type that defines the type of response the dialogue model will give for each of the set dialogue types, set at least one evaluation criterion for evaluating the dialogue based on the response type, provide the dialogue type, the response type, and the evaluation criterion to the dialogue model as dialogue instruction information, and adjust the dialogue model to execute a dialogue based on the response type that is expected to correspond to the dialogue type of the interlocutor. The device 10 may, for example, record the set dialogue type, response type, and evaluation criterion in the storage unit.

[0026] The dialogue model adjustment unit 15 can set at least one selected from the group consisting of the dialogue type, the response type, and the evaluation criteria by, for example, providing definition generation instruction information to the large language model, which instructs it to define at least one selected from the group consisting of the dialogue type, the response type, and the evaluation criteria. The large language model may be, for example, the same as the dialogue model, or it may be different from the dialogue model. The definition generation instruction information may use prompts from the following specific examples, for example, but this disclosure is not limited to these specific examples. - Example prompt for a large-scale language model to define dialogue types: "Please define three user types with as many distinct characteristics as possible. Give each user type a name and describe the characteristics of each type." - Example prompt for a large-scale language model to define response types: "Based on the characteristics of each user type, please define an appropriate LLM personality type that would be a good match for that user type." - Example prompt for a large-scale language model to define evaluation criteria: "Assuming that each LLM personality type's response to a hypothetical user statement will be scored, please create five scoring criteria. For each scoring criterion, create criteria for high score, medium score, and low score."

[0027] The aforementioned dialogue type is information that defines the characteristics of the dialoguer who engages in dialogue with the dialogue model. The aforementioned dialogue type may be defined in natural language, in terms of features, or in combination thereof.

[0028] When the aforementioned dialogue type is defined in natural language, the dialogue type is not particularly limited, and specific examples include, for example, those listed below, but this disclosure is not limited in any way to these examples: • Dialogue Type: Beginner Characteristics: Unfamiliar with technology, seeks detailed explanations, prefers a slow pace. In the stage of learning something new, needs basic support. Frequently asks questions about usage and basic knowledge. Often lacks confidence and seeks encouragement and reassurance. • Dialogue Type: Efficiency-Oriented Professional Characteristics: Seeks quick answers, understands technical terms, wants to complete time-sensitive tasks quickly, efficiency-oriented dialoguer. Does not require detailed explanations and seeks direct and quick answers. Has a deep understanding of the service or tool. • Dialogue Type: Emotionally Supported User Characteristics: Seeks empathetic responses, has anxieties or concerns, needs encouragement, is in a situation requiring emotional support and empathy. Being listened to and sharing emotions is important. Seeks reassurance and comfort.

[0029] If the dialogue type is defined by a feature, the feature may be a vector whose elements include one or more numerical values ​​(continuous or discrete values) calculated based on dialogue data between the interlocutor and the dialogue model. The feature may include at least one selected from the group consisting of, for example, the average sentence length, average number of words, input frequency, proportion of question sentences, proportion of imperative sentences, proportion of negative expressions, frequency of occurrence of symbols such as exclamation marks in the interlocutor's input data, an index representing the level of explanation requested by the interlocutor (e.g., frequency of occurrence of predetermined words such as "why," "reason," "basis," "specific example"), an index representing the interlocutor's response preferences (e.g., whether or not bullet points are requested, whether or not conclusions are given first, whether or not conciseness is requested, whether or not politeness is requested), frequency of occurrence of expressions suggesting the interlocutor's emotional state (e.g., anxious words, reassuring words, encouraging words, etc.), tendency to use vocabulary suggesting the interlocutor's expertise or proficiency (frequency of occurrence of technical terms, frequency of occurrence of proper nouns, etc.), and an index representing the interlocutor's dialogue pace (input interval, waiting time until response, etc.). Furthermore, the aforementioned features may include statistical measures (mean vector, variance, change in the time series direction, etc.) of embedding vectors (e.g., real-valued vectors of several hundred to several thousand dimensions) obtained by dividing the dialogue data into unit sentences or unit utterances. The types, dimensions, and calculation methods of the aforementioned features are not limited to the examples given above.

[0030] The response type is, for example, information defining a response type suitable for each dialogue type. The "response type" may include, but is not limited to, the content to be prioritized in the response, the format of the response (e.g., whether or not bullet points are used), and the degree of politeness of the response. The response type is not particularly limited, and specific examples such as those shown below can be given, but this disclosure is not limited to these examples. When the dialogue type is a beginner, for example, the response type is a guide type. The guide type is a response type that provides responses with the following characteristics, for example: Personality type: Kind and patient guide. Detailed explanation: For beginners, basic concepts and usage are explained carefully. Encouraging: Praises even small progress and offers words of encouragement. Responding to repetition: Responds patiently to the same questions and doubts and explains them as many times as needed. When the dialogue type is an efficiency-oriented professional, for example, the response type is a concierge type. The aforementioned concierge type is a response type that provides responses with the following characteristics, for example: Personality type: Clear and prompt concierge Short and to-the-point answers: Provides simple and direct answers, omitting unnecessary explanations and background information. Prompt response: Responds quickly to inquiries, minimizing waiting times. Suggestions for efficiency improvements: Suggests advanced functions, shortcuts, and efficient workflows. When the interlocutor type is an emotionally supported user, a response type such as the friend type can be cited. The aforementioned friend type is a response type that provides responses with the following characteristics, for example: Personality type: Empathetic and reassuring friend Empathetic responses: Shows empathy and understanding for the interlocutor's emotions and feelings. Comfort and support: Provides appropriate encouragement and reassurance, creating a sense of security. Deep dialogue: Does not merely provide responses, but listens to the interlocutor and continues the dialogue.

[0031] The aforementioned evaluation criteria are information for evaluating the content of responses from a dialogue model that has interacted with an interlocutor. The evaluation criteria include, for example, at least one scoring category, and for each scoring category, information that links a description of the content of the response with the corresponding evaluation score. The following are specific examples of evaluation criteria, but this disclosure is not limited to these examples. 1. Empathy and Understanding 1 point: No understanding or empathy is felt for the interlocutor's feelings or situation. 3 points: Basic empathy and understanding are shown, but it is somewhat formal. 5 points: Deep empathy is shown for the interlocutor's feelings, and a more specific understanding is conveyed. 2. Specificity and Practicality 1 point: The response is abstract and lacks specific solutions or action plans. 3 points: There is basic specificity, but more detailed actions or suggestions are desired. 5 points: It is very specific and includes action plans and suggestions that can be put into practice. 3. Adaptation to the Interlocutor's Needs 1 point: It does not match the interlocutor's specific situation or needs at all. 3 points: It addresses the interlocutor's needs to some extent, but not completely. 5 points: The response is perfectly tailored to the needs of the interlocutor. 4. Tone and Attitude 1 point: The tone is inappropriate and fails to encourage the interlocutor. 3 points: The tone is appropriate, but somewhat formal and lacking in friendliness. 5 points: The tone is very friendly and reassuring to the interlocutor. 5. Solution Orientation 1 point: There is no sign of a willingness to solve the problem. 3 points: There is a willingness to solve the problem, but more action is needed. 5 points: Strong support and concrete suggestions are provided to solve the problem.

[0032] The weights of the evaluation criteria corresponding to the aforementioned dialogue types are, for example, coefficients for the scoring categories that are important for each dialogue type, and any values ​​can be set. As specific examples, the weights for the scoring categories emphasized for each dialogue type are shown below, but this disclosure is not limited to these specific examples. 1. Dialogue Type: Beginner Empathy and Understanding: Weight 1.1 Specificity and Practicality: Weight 1.1 Adaptation to the Interlocutor's Needs: Weight 1.0 Tone and Attitude: Weight 1.1 Solution-Oriented: Weight 0.7 2. Dialogue Type: High-Efficiency Professional Empathy and Understanding: Weight 0.8 Specificity and Practicality: Weight 1.2 Adaptation to the Interlocutor's Needs: Weight 1.1 Tone and Attitude: Weight 0.8 Solution-Oriented: Weight 1.1 3. Dialogue Type: Interlocutor Seeking Mental Support Empathy and Understanding: Weight 1.2 Specificity and Practicality: Weight 0.8 Adaptation to the Interlocutor's Needs: Weight 1.1 Tone and Attitude: Weight 1.2 Solution-Oriented: Weight 0.7

[0033] Furthermore, in the aforementioned evaluation criteria, information on scoring categories emphasized for each dialogue type may be linked, for example, in place of or in addition to the aforementioned weights. Specific examples of scoring categories emphasized for each dialogue type are shown below, but this disclosure is not limited to these examples. 1. Dialogue Type: Beginner Empathy and understanding Specificity and practicality Tone and attitude 2. Dialogue Type: Professional seeking high efficiency Specificity and practicality Adaptation to the needs of the dialogue participant Solution-oriented 3. Dialogue Type: Dialogue participant seeking mental support Empathy and understanding Adaptation to the needs of the dialogue participant Tone and attitude

[0034] The dialogue data acquisition unit 11 acquires dialogue data between the interlocutor and the dialogue model (S1, dialogue data acquisition step). The dialogue data includes at least one input data input by the interlocutor to the dialogue model and at least one response data output by the dialogue model based on the input data. The response data is, for example, data generated by the dialogue model based on the input data of the given interlocutor. The response data may be, for example, data generated by one dialogue model or data jointly generated by multiple dialogue models. The input data is not particularly limited and may be, for example, input data when the interlocutor of the dialogue model interacts with the dialogue model. The dialogue data acquisition unit 11 may, for example, acquire dialogue data from the dialogue model, or acquire various data from an external recording medium that records various data. The dialogue data acquisition unit 11 may, for example, store the acquired various data in the storage unit of the device 10.

[0035] If the dialogue model is a model that is adjusted to provide responses of a response type corresponding to the dialogue type of the interlocutor, the dialogue data may include, for example, type identification information that identifies the response type adopted by the dialogue model.

[0036] The dialogue type estimation unit 12 estimates the dialogue type of the interlocutor based on the dialogue data (S2, dialogue type estimation step). The dialogue type estimation unit 12 can estimate the dialogue type of the interlocutor who interacted with the dialogue model by, for example, analyzing the content of the input data in response to the response data output by the dialogue model in the dialogue data, and analyzing the interlocutor's response to the response (e.g., additional questions, words of thanks, expressions of frustration, etc.). Alternatively, the dialogue type estimation unit 12 may estimate (obtain) the dialogue type by, for example, providing a large-scale language model with instruction information (prompt) that instructs it to estimate the dialogue type based on the dialogue data, and the dialogue data, thereby causing the model to estimate the dialogue type of the interlocutor. In this case, the estimation result may be information that expresses the dialogue type in natural language, or it may be a feature quantity (e.g., a feature quantity vector) that represents the dialogue type. Furthermore, the dialogue type estimation unit 12 may, for example, divide the input data and response data included in the dialogue data into unit sentences or unit utterances, and convert each unit sentence or unit utterance into an embedding vector. The dialogue type estimation unit 12 may extract statistical quantities (mean vector, variance, change in the time series direction, etc.) from the generated embedding vector sequence as features, and use the extracted features as input to estimate the dialogue type of the interlocutor using a classifier (e.g., a multilayer neural network, support vector machine, decision tree, etc.). Since the embedding vector may contain real-valued elements of several hundred to several thousand dimensions, the feature extraction and classification process consists of numerical calculations involving numerous matrix operations. For this reason, it is practically impossible for a human to perform the feature extraction and classification process in their mind or with paper and pencil, and relies on automated processing by an electronic computer equipped with a processor and memory.

[0037] The response evaluation unit 13 evaluates the response data in the dialogue data and calculates an evaluation value using the evaluation criteria and the weights of the evaluation criteria corresponding to the dialogue type (S3, response evaluation step). Specifically, the response evaluation unit 13 can calculate the evaluation value of the response data by, for example, calculating an evaluation value for the response data in the dialogue data according to the five scoring categories in the aforementioned evaluation criteria, and further multiplying it by the weights corresponding to the dialogue type of the interlocutor estimated in S2. The response evaluation unit 13 may, for example, generate an evaluation score vector with the evaluation score as an element, and calculate the overall evaluation value of the response data by multiplying the evaluation score vector by the weights of the evaluation criteria (weight vector) set corresponding to the dialogue type and adding it. For example, the overall evaluation value may be calculated as the dot product (linear combination) of the evaluation score vector and the weight vector. Alternatively, the response evaluation unit 13 may, for example, identify the response type of the dialogue model based on the response data and calculate the evaluation value as the evaluation value of the response type. The response type may be identified, for example, based on the type identification information contained in the dialogue data, or it may be identified as the response type corresponding to the dialogue type estimated in S2, or the response type may be identified by analyzing the contents of the response data. The response evaluation unit 13 may, for example, when the response type is identified, calculate an evaluation value suitable for the response type by switching the weight vector in accordance with the response type, or by updating only some dimensions of the weight vector. Furthermore, the response evaluation unit 13 may normalize the calculated overall evaluation value to a predetermined range (for example, 0 to 1) so that the calculated overall evaluation value can be used as weight information for the dialogue sample in the dialogue model adjustment data.

[0038] The adjustment data recording unit 14 records the evaluation value and the dialogue data as dialogue model adjustment data for the dialogue user (S4, adjustment data recording step). Specifically, the adjustment data recording unit 14 determines, for example, whether the evaluation value exceeds a threshold, and if the evaluation value exceeds the threshold, it records the evaluation value and the dialogue data as dialogue model adjustment data. The threshold is not particularly limited and any value can be set. The threshold may be recorded, for example, in the storage unit of the device 10, or on a recording medium outside the device 10. In this case, the adjustment data recording unit 14 may record, for example, the evaluation value, the response type, and the dialogue data as dialogue model adjustment data for the dialogue user. Furthermore, if the evaluation value is less than the threshold, the adjustment data recording unit 14 may further record data generated from the dialogue data as dialogue model adjustment data. The adjustment data recording unit 14 can, for example, estimate the type of response (response type) that the interlocutor is likely to consider important based on the weights (weight vectors) of evaluation criteria corresponding to the interlocutor's dialogue type, generate response data corresponding to the estimated type, and record the generated data in place of or in addition to the response data as the model adjustment data. Furthermore, the adjustment data recording unit 14 may record a pair of response data (first response) whose evaluation value is below a threshold and response data (second response) generated according to the estimated response type, and may add comparison information (ranking information) indicating that the second response is preferable to the first response. This allows the dialogue model adjustment unit 15 to use the dialogue model adjustment data not only as training data for supervised learning but also for learning based on preferred comparisons (e.g., ranking learning). The adjustment data recording unit 14 may also record the evaluation value as weight information normalized or clipped to a predetermined range (e.g., 0 to 1).

[0039] The dialogue model adjustment unit 15 uses the dialogue model adjustment data to adjust the dialogue model so that it can perform a dialogue adapted to the dialogue user (S5, dialogue model adjustment step). The dialogue model adjustment unit 15 may, for example, perform the dialogue model adjustment process each time the dialogue model adjustment data is accumulated in S4, or it may perform the dialogue model adjustment process when the accumulated amount of dialogue model adjustment data exceeds a certain level. Furthermore, the dialogue model adjustment unit 15 may detect changes in the dialogue user's preferences and control the timing of the adjustment process, for example, by executing the adjustment process when the distribution of the evaluation values ​​included in the dialogue model adjustment data (mean value, variance, percentage above a predetermined threshold, etc.) satisfies predetermined conditions. Examples of the dialogue model adjustment process include, but are not limited to, a fine-tuning process using the dialogue model adjustment data, an adjustment using evolutionary model merging, and a process of writing the dialogue model adjustment data to the system prompt of the dialogue model.

[0040] The fine-tuning process is not particularly limited and may, for example, involve retraining the weights of all layers of the dialogue model, or it may involve retraining the weights of a specific layer. The fine-tuning process may, for example, involve adding parameters such as an adapter layer for additional learning or a low-rank adaptation (LoRA), and updating only the added parameters. Specifically, the dialogue model adjustment unit 15 may use the input data and response data included in the dialogue model adjustment data as training samples and calculate the loss (e.g., cross-entropy loss) between the output distribution obtained when the input data is input to the dialogue model and the response data. The dialogue model adjustment unit 15 may use the evaluation value (e.g., normalized or clipped value) calculated in step S3 as the weight of the training sample and define a weighted loss obtained by multiplying the loss by the weight. The dialogue model adjustment unit 15 may then fine-tune the dialogue model to perform a dialogue adapted to the dialogue user by iteratively updating a number of parameters of the dialogue model in mini-batch units using gradient descent or a modified algorithm therefor (e.g., Adam) so that the weighted loss is reduced. Furthermore, if, as dialogue model adjustment data, a response data with a low evaluation value (first response) and a response data generated according to the estimated response type (second response) are recorded as a pair, and comparative information (ranking information) indicating that the second response is preferable to the first response is provided, the dialogue model adjustment unit 15 may adjust the dialogue model by ranking learning based on the comparative information. Note that since the fine-tuning process consists of numerical calculations that repeatedly perform a large number of matrix operations and backpropagation calculations, it is practically impossible for a human to perform it in their head or with paper and pencil, and therefore relies on automatic processing by an electronic computer equipped with a processor and memory.

[0041] The evolutionary model merging described above is a method that utilizes an evolutionary algorithm when merging (merging) multiple dialogue models. For evolutionary model merging, for example, the method described in Reference 1 below may be used. In this case, the dialogue model adjustment unit 15 first prepares multiple dialogue models as parent models. The parent models may or may not include dialogue models that have interacted with the dialogue user. The parent models are not particularly limited and may include, for example, models that are adjusted to provide responses according to the dialogue type of the dialogue user. Next, the dialogue model adjustment unit 15 selects multiple arbitrary parent models and generates a new child model by combining the parameters of the selected parent models. At this time, the dialogue model adjustment unit 15 may, for example, perform mutation or crossover in the way the parameters are combined. Next, the dialogue model adjustment unit 15 evaluates the generated child model using the dialogue model adjustment data. As a specific example, the dialogue model adjustment unit 15, for instance, provides the input data of the dialoguer to the generated child model based on the dialogue model adjustment data, causing the dialogue model to generate a response based on the input data. The dialogue model adjustment unit 15 then calculates an evaluation value for the response data generated by the child model based on the evaluation criteria and weights used in step S3. The dialogue model adjustment unit 15 then retains, for example, the child model with the highest calculated evaluation value as the next-generation parent model. By repeating this process, the dialogue model adjustment unit 15 can evolve the model to be optimized for the dialoguer. Note that the generation of numerous candidate models, inference execution, and evaluation value calculation in the evolutionary model merging require a large amount of computing resources, so the process relies on automated processing by a computer.Reference 1: Takuya Akiba, Makoto Shing, Yujin Tang, Qi Sun, David Ha, “Evolutionary Optimization of Model Merging Recipes”, [online], March 19, 2024, arXiv, [searched on March 23, 2017], Internet <URL: https: / / arxiv.org / pdf / 2403.13187>.

[0042] When writing dialogue model adjustment data to the dialogue model system prompt, the dialogue model adjustment unit 15 may, for example, instruct the system prompt to prioritize evaluation criteria that the dialoguer considers important. In this case, the dialogue model adjustment unit 15 writes to the dialogue model system prompt so that it generates a response that prioritizes scoring categories for which the weights of the evaluation criteria corresponding to the dialogue type of the dialoguer estimated in S2 are set high. In this case, the dialogue model adjustment unit 15 may, for example, extract response rules corresponding to scoring categories with high weights based on the weights (weight vectors) of the evaluation criteria corresponding to the dialogue type of the dialoguer estimated in S2, and add or update the system prompt with these response rules. For example, the top N scoring categories with high weights may be selected and written to the system prompt as instruction information that clearly indicates the priority of response types (conclusion first, bullet points, evidence presentation, empathy expression, etc.) corresponding to the selected scoring categories. For example, if the interlocutor is a beginner, prompts such as "Please respond in a way that maximizes the evaluation criteria for empathy and understanding, specificity and practicality, and tone and attitude" or "Please respond in a way that maximizes the evaluation criteria for empathy and understanding, specificity and practicality, and tone and attitude. Specifically, please respond in a way that: shows deep empathy for the interlocutor's feelings and conveys concrete understanding; includes very specific and actionable action plans or suggestions; and has a very friendly and reassuring tone" may be added. Furthermore, the dialogue model adjustment unit 15 may update the content of the system prompts to follow the interlocutor's preferences by prioritizing the retention of instruction information corresponding to dialogues in which the evaluation value exceeds a predetermined threshold, and by deleting or modifying instruction information corresponding to dialogues in which the evaluation value falls below the predetermined threshold.

[0043] According to the present disclosure, a dialogue data acquisition unit acquires dialogue data between a dialogue participant and a dialogue model; a dialogue type estimation unit estimates the dialogue type of the dialogue participant based on the dialogue data; a response evaluation unit evaluates the response data in the dialogue data and calculates an evaluation value by using an evaluation criterion and a weight of the evaluation criterion corresponding to the dialogue type; an adjustment data recording unit records the evaluation value and the dialogue data as dialogue model adjustment data for the dialogue participant; and a dialogue model adjustment unit can adjust the dialogue model by using the dialogue model adjustment data to implement dialogue adapted to the dialogue participant. Therefore, according to the present disclosure, an evaluation value according to each dialogue participant's preference can be reflected as a learning signal in the adjustment processing of the dialogue model, and the response quality of a dialogue system can be continuously improved.

[0044] The response type of the dialogue model adjusted according to the present disclosure is optimized in accordance with the dialogue type of the dialogue participant, thereby enabling implementation of dialogue individually adapted to the dialogue participant. Therefore, according to the present disclosure, a dialogue model individually adapted to a dialogue participant can be adjusted. Further, according to the present disclosure, dialogue model adjustment data is configured using an evaluation value obtained by switching the weight (weight vector) of an evaluation criterion in accordance with a dialogue type, and the evaluation value can be used as a learning weight, so that dialogue samples matching the dialogue participant's preference can be relatively strongly reflected in learning, and the influence of dialogue samples not matching the preference can be suppressed. Therefore, according to the present disclosure, for example, it becomes possible to provide various personalized services to a dialogue participant who is a user by using a dialogue model. Further, according to the present disclosure, compared to a case where the same dialogue model is uniformly learned or adjusted, learning signals can be efficiently extracted even from limited dialogue logs, and while suppressing unnecessary re-learning and inappropriate adjustment, the speed of adjustment for generating a response adapted to a dialogue participant can be increased. In addition, according to the configuration in which the evaluation criterion is switched or updated according to the dialogue type and the response type, the consistency and stability of dialogue can be improved, and the response quality of the entire dialogue system can be improved.

[0045] While the present disclosure has been described above with reference to embodiments, the present disclosure is not limited to the embodiments described above. Various modifications to the structure and details of the present disclosure are possible, as can be understood by those skilled in the art within the scope of the present disclosure.

[0046] This application claims priority based on Japanese Patent Application No. 2025-049894, filed on 25 March 2025, and incorporates all of its disclosures herein.

[0047] <Note> Some or all of the above embodiments may be described as follows, but are not limited to the following. (Note 1) A dialogue model adjustment support device comprising: a dialogue data acquisition unit, a dialogue type estimation unit, a response evaluation unit, an adjustment data recording unit, and a dialogue model adjustment unit, wherein the dialogue data acquisition unit acquires dialogue data between a person and a dialogue model, the dialogue data includes at least one input data input by the person to the dialogue model and at least one response data output by the dialogue model based on the input data, the dialogue type estimation unit estimates the dialogue type of the person based on the dialogue data, the dialogue type is information defining the characteristics of the person who engages in dialogue with the dialogue model, the response evaluation unit evaluates the response data in the dialogue data using an evaluation criterion and the weight of the evaluation criterion corresponding to the dialogue type to calculate an evaluation value, the adjustment data recording unit records the evaluation value and the dialogue data as dialogue model adjustment data for the person, and the dialogue model adjustment unit adjusts the dialogue model using the dialogue model adjustment data to perform dialogue adapted to the person. (Note 2) The dialogue model adjustment support device according to Note 1, wherein the response evaluation unit identifies the response type of the dialogue model based on the response data, calculates the evaluation value of the response type, the response type is information defining a response type suitable for each dialogue type, and the adjustment data recording unit records the evaluation value, the response type, and the dialogue data as dialogue model adjustment data for the dialogue user.(Note 3) The dialogue model adjustment support device according to Note 1 or 2, wherein, prior to the dialogue between the interlocutor and the dialogue model, the dialogue model adjustment unit sets at least one dialogue type of the interlocutor with whom the dialogue model is expected to engage in dialogue, sets at least one response type that defines the type of response the dialogue model will give for each of the set dialogue types, sets at least one evaluation criterion for evaluating the dialogue based on the response type, provides the dialogue type, the response type, and the evaluation criterion to the dialogue model as dialogue instruction information, and adjusts the dialogue model to perform a dialogue based on the response type that is expected to correspond to the dialogue type of the interlocutor. (Note 4) The dialogue model adjustment support device according to Note 3, wherein the dialogue model adjustment unit sets at least one selected from the group consisting of the dialogue type, the response type, and the evaluation criterion by providing definition generation instruction information that instructs the large-scale language model to define at least one selected from the group consisting of the dialogue type, the response type, and the evaluation criterion. (Note 5) A dialogue model adjustment support method comprising a dialogue data acquisition step, a dialogue type estimation step, a response evaluation step, an adjustment data recording step, and a dialogue model adjustment step, wherein each step is performed by a computer. The dialogue data acquisition step acquires dialogue data between a person and a dialogue model, the dialogue data includes at least one input data input by the person to the dialogue model and at least one response data output by the dialogue model based on the input data, the dialogue type estimation step estimates the dialogue type of the person based on the dialogue data, the dialogue type is information defining the characteristics of the person who interacts with the dialogue model, the response evaluation step evaluates the response data in the dialogue data using an evaluation criterion and the weight of the evaluation criterion corresponding to the dialogue type to calculate an evaluation value, the adjustment data recording step records the evaluation value and the dialogue data as dialogue model adjustment data for the person, and the dialogue model adjustment step adjusts the dialogue model using the dialogue model adjustment data to perform a dialogue adapted to the person.(Note 6) The dialogue model adjustment support method according to Note 5, wherein the response evaluation step includes identifying the response type of the dialogue model based on the response data, calculating the evaluation value of the response type, the response type being information defining a suitable response type for each dialogue type, and the adjustment data recording step includes recording the evaluation value, the response type, and the dialogue data as dialogue model adjustment data for the interlocutor. (Note 7) The dialogue model adjustment step includes, prior to the dialogue between the interlocutor and the dialogue model, setting at least one dialogue type of the interlocutor with whom the dialogue model is expected to engage in dialogue, setting at least one response type defining a response type to which the dialogue model will respond for each of the set dialogue types, setting at least one evaluation criterion for evaluating the dialogue by the response type, providing the dialogue type, the response type, and the evaluation criterion to the dialogue model as dialogue instruction information, and adjusting the dialogue model to perform a dialogue by the response type that assumes the dialogue type of the interlocutor. (Note 8) The dialogue model adjustment support method according to Note 7, wherein the dialogue model adjustment step sets at least one selected from the group consisting of the dialogue type, the response type, and the evaluation criteria by providing definition generation instruction information that instructs the large-scale language model to define at least one selected from the group consisting of the dialogue type, the response type, and the evaluation criteria.(Note 9) A dialogue model adjustment support program for causing a computer to execute each of the following procedures: a dialogue data acquisition procedure, a dialogue type estimation procedure, a response evaluation procedure, an adjustment data recording procedure, and a dialogue model adjustment procedure, wherein the dialogue data acquisition procedure acquires dialogue data between a person and a dialogue model, the dialogue data includes at least one input data input by the person to the dialogue model and at least one response data output by the dialogue model based on the input data, the dialogue type estimation procedure estimates the dialogue type of the person based on the dialogue data, the dialogue type is information defining the characteristics of a person who interacts with the dialogue model, the response evaluation procedure evaluates the response data in the dialogue data using an evaluation criterion and the weight of the evaluation criterion corresponding to the dialogue type to calculate an evaluation value, the adjustment data recording procedure records the evaluation value and the dialogue data as dialogue model adjustment data for the person, and the dialogue model adjustment procedure adjusts the dialogue model using the dialogue model adjustment data to perform a dialogue adapted to the person. (Note 10) The dialogue model adjustment support program according to Note 9, wherein the response evaluation procedure identifies the response type of the dialogue model based on the response data, calculates the evaluation value of the response type, the response type is information defining a suitable response type for each dialogue type, and the adjustment data recording procedure records the evaluation value, the response type, and the dialogue data as dialogue model adjustment data for the interlocutor. (Note 11) The dialogue model adjustment procedure, prior to the dialogue between the interlocutor and the dialogue model, sets at least one dialogue type of the interlocutor with whom the dialogue model is expected to engage in dialogue, sets at least one response type defining a response type to which the dialogue model will respond for each set dialogue type, sets at least one evaluation criterion for evaluating the dialogue by the response type, provides the dialogue type, the response type, and the evaluation criterion to the dialogue model as dialogue instruction information, and adjusts the dialogue model to perform a dialogue by the response type that assumes the dialogue type of the interlocutor.(Note 12) The dialogue model adjustment procedure is a dialogue model adjustment support program as described in Note 11, which sets at least one selected from the group consisting of the dialogue type, the response type, and the evaluation criteria by providing definition generation instruction information that instructs the large-scale language model to define at least one selected from the group consisting of the dialogue type, the response type, and the evaluation criteria. (Note 13) A computer-readable recording medium recording a dialogue model adjustment support program for causing a computer to execute each of the following procedures: a dialogue data acquisition procedure, a dialogue type estimation procedure, a response evaluation procedure, an adjustment data recording procedure, and a dialogue model adjustment procedure, wherein the dialogue data acquisition procedure acquires dialogue data between a person and a dialogue model, the dialogue data includes at least one input data input by the person to the dialogue model and at least one response data output by the dialogue model based on the input data, the dialogue type estimation procedure estimates the dialogue type of the person based on the dialogue data, the dialogue type is information defining the characteristics of a person who interacts with the dialogue model, the response evaluation procedure evaluates the response data in the dialogue data using an evaluation criterion and the weight of the evaluation criterion corresponding to the dialogue type to calculate an evaluation value, the adjustment data recording procedure records the evaluation value and the dialogue data as dialogue model adjustment data for the person, and the dialogue model adjustment procedure adjusts the dialogue model using the dialogue model adjustment data to perform a dialogue adapted to the person. (Note 14) The recording medium described in Note 13, wherein the response evaluation procedure identifies the response type of the dialogue model based on the response data, calculates the evaluation value of the response type, the response type is information defining a suitable response type for each dialogue type, and the adjustment data recording procedure records the evaluation value, the response type, and the dialogue data as dialogue model adjustment data for the dialogue user.(Supplementary Note 15) The dialogue model adjustment procedure, prior to a dialogue between said interlocutor and said dialogue model, sets at least one dialogue type for an interlocutor with which said dialogue model is expected to converse, sets at least one response type defining a response pattern for said dialogue model for each set dialogue type, sets at least one evaluation criterion for evaluating dialogue based on said response type, provides said dialogue type, said response type, and said evaluation criterion as dialogue instruction information to said dialogue model, and adjusts said dialogue model to cause it to execute dialogue according to said response type assuming the dialogue type of said interlocutor. The recording medium according to Supplementary Note 13 or 14. (Supplementary Note 16) In the dialogue model adjustment procedure, definition generation instruction information that instructs a large language model to define at least one selected from the group consisting of said dialogue type, said response type, and said evaluation criterion is provided, thereby setting at least one selected from the group consisting of said dialogue type, said response type, and said evaluation criterion. The recording medium according to Supplementary Note 15.

[0048] Therefore, according to the present disclosure, for example, a dialogue model can be adjusted so as to enable dialogue individually adapted to an interlocutor. Accordingly, the present disclosure is useful in a wide variety of industries that utilize fine-tuned large-scale learning models, for example.

[0049] 10 Dialogue model adjustment support device 11 Dialogue data acquisition unit 12 Dialogue type estimation unit 13 Response evaluation unit 14 Adjustment data recording unit 15 Dialogue model adjustment unit 101 Central processing unit 102 Memory 103 Bus 104 Storage device 105 Input device 106 Output device 107 Communication device

Claims

It includes a dialogue data acquisition unit, a dialogue type estimation unit, a response evaluation unit, an adjustment data recording unit, and a dialogue model adjustment unit. The aforementioned dialogue data acquisition unit acquires dialogue data between the dialoguer and the dialogue model. The dialogue data includes at least one input data entered by the dialoguer to the dialogue model, and at least one response data output by the dialogue model based on the input data. The dialogue type estimation unit estimates the dialogue type of the dialoguers based on the dialogue data. The aforementioned dialogue type is information that defines the characteristics of the dialoguer who engages in dialogue with the aforementioned dialogue model. The response evaluation unit evaluates the response data in the dialogue data using the evaluation criteria and the weights of the evaluation criteria corresponding to the dialogue type, and calculates an evaluation value. The adjustment data recording unit records the evaluation value and the dialogue data as dialogue model adjustment data for the dialogue participant. The dialogue model adjustment unit is a dialogue model adjustment support device that uses the dialogue model adjustment data to adjust the dialogue model so that it can perform a dialogue that is adapted to the person speaking. The response evaluation unit, Based on the response data, the response type of the dialogue model is identified. The evaluation value of the response type is calculated, The response type is information that defines a suitable response type for each dialogue type. The dialogue model adjustment support device according to claim 1, wherein the adjustment data recording unit records the evaluation value, the response type, and the dialogue data as dialogue model adjustment data for the dialogue participant. Prior to the dialogue between the dialoguer and the dialogue model, the dialogue model adjustment unit performs the following actions: The dialogue model is expected to engage with at least one type of dialogue between the dialoguers, For each of the configured dialogue types, at least one response type is defined that specifies the type of response the dialogue model will use. At least one evaluation criterion is established for evaluating the dialogue based on the aforementioned response type, The dialogue model adjustment support device according to claim 1 or 2, wherein the dialogue type, the response type, and the evaluation criteria are provided to the dialogue model as dialogue instruction information, and the dialogue model is adjusted to perform a dialogue using the response type that assumes the dialogue type of the interlocutor. The dialogue model adjustment support device according to claim 3, wherein the dialogue model adjustment unit sets at least one selected from the group consisting of the dialogue type, the response type, and the evaluation criteria by providing definition generation instruction information that instructs the large-scale language model to define at least one selected from the group consisting of the dialogue type, the response type, and the evaluation criteria. This process includes a dialogue data acquisition step, a dialogue type estimation step, a response evaluation step, a data recording step for adjustment, and a dialogue model adjustment step. The dialogue data acquisition process acquires dialogue data between the dialoguer and the dialogue model. The dialogue data includes at least one input data entered by the dialoguer to the dialogue model, and at least one response data output by the dialogue model based on the input data. The dialogue type estimation step estimates the dialogue type of the dialoguers based on the dialogue data, The aforementioned dialogue type is information that defines the characteristics of the dialoguer who engages in dialogue with the aforementioned dialogue model. The response evaluation step evaluates the response data in the dialogue data using the evaluation criteria and the weights of the evaluation criteria corresponding to the dialogue type, and calculates an evaluation value. The adjustment data recording step records the evaluation value and the dialogue data as dialogue model adjustment data for the dialogue participant. The dialogue model adjustment step involves adjusting the dialogue model using the dialogue model adjustment data to perform a dialogue that is adapted to the person speaking. A method for assisting the adjustment of a dialogue model, in which each step is performed by a computer. The response evaluation step is, Based on the response data, the response type of the dialogue model is identified. The evaluation value of the response type is calculated, The response type is information that defines a suitable response type for each dialogue type. The dialogue model adjustment support method according to claim 5, wherein the adjustment data recording step records the evaluation value, the response type, and the dialogue data as dialogue model adjustment data for the dialogue participant. The dialogue model adjustment process is performed prior to the dialogue between the dialoguer and the dialogue model, The dialogue model is expected to engage with at least one type of dialogue between the dialoguers, For each of the configured dialogue types, at least one response type is defined that specifies the type of response the dialogue model will use. At least one evaluation criterion is established for evaluating the dialogue based on the aforementioned response type, A dialogue model adjustment support method according to claim 5 or 6, comprising providing the dialogue type, the response type, and the evaluation criteria to the dialogue model as dialogue instruction information, and adjusting the dialogue model to perform a dialogue using the response type that assumes the dialogue type of the interlocutor. The dialogue model adjustment support method according to claim 7, wherein the dialogue model adjustment step sets at least one selected from the group consisting of the dialogue type, the response type, and the evaluation criteria by providing definition generation instruction information that instructs the large-scale language model to define at least one selected from the group consisting of the dialogue type, the response type, and the evaluation criteria. This includes procedures for acquiring dialogue data, estimating dialogue type, evaluating responses, recording adjustment data, and adjusting the dialogue model. The aforementioned dialogue data acquisition procedure acquires dialogue data between the dialoguer and the dialogue model, The dialogue data includes at least one input data entered by the dialoguer to the dialogue model, and at least one response data output by the dialogue model based on the input data. The dialogue type estimation procedure estimates the dialogue type of the dialoguers based on the dialogue data, The aforementioned dialogue type is information that defines the characteristics of the dialoguer who engages in dialogue with the aforementioned dialogue model. The response evaluation procedure evaluates the response data in the dialogue data using the evaluation criteria and the weights of the evaluation criteria corresponding to the dialogue type, and calculates an evaluation value. The adjustment data recording procedure records the evaluation value and the dialogue data as dialogue model adjustment data for the dialogue participant. The dialogue model adjustment procedure involves adjusting the dialogue model using the dialogue model adjustment data to perform a dialogue adapted to the dialogue with the dialogue subject. A dialogue model adjustment support program for instructing a computer to perform each step. The response evaluation procedure described above is: Based on the response data, the response type of the dialogue model is identified. The evaluation value of the response type is calculated, The response type is information that defines a suitable response type for each dialogue type. The dialogue model adjustment support program according to claim 9, wherein the adjustment data recording procedure records the evaluation value, the response type, and the dialogue data as dialogue model adjustment data for the dialogue participant. The aforementioned dialogue model adjustment procedure is performed prior to the dialogue between the interlocutor and the dialogue model, The dialogue model is expected to engage with at least one type of dialogue between the dialoguers, For each of the configured dialogue types, at least one response type is defined that specifies the type of response the dialogue model will use. At least one evaluation criterion is established for evaluating the dialogue based on the aforementioned response type, A dialogue model adjustment support program according to claim 9 or 10, which provides the dialogue type, the response type, and the evaluation criteria to the dialogue model as dialogue instruction information, and adjusts the dialogue model to perform a dialogue using the response type that assumes the dialogue type of the interlocutor. The dialogue model adjustment procedure sets at least one selected from the group consisting of the dialogue type, the response type, and the evaluation criteria by providing definition generation instruction information to a large language model that instructs it to define at least one selected from the group consisting of the dialogue type, the response type, and the evaluation criteria, according to claim 11. This includes procedures for acquiring dialogue data, estimating dialogue type, evaluating responses, recording adjustment data, and adjusting the dialogue model. The aforementioned dialogue data acquisition procedure acquires dialogue data between the dialoguer and the dialogue model, The dialogue data includes at least one input data entered by the dialoguer to the dialogue model, and at least one response data output by the dialogue model based on the input data. The dialogue type estimation procedure estimates the dialogue type of the dialoguers based on the dialogue data, The aforementioned dialogue type is information that defines the characteristics of the dialoguer who engages in dialogue with the aforementioned dialogue model. The response evaluation procedure evaluates the response data in the dialogue data using the evaluation criteria and the weights of the evaluation criteria corresponding to the dialogue type, and calculates an evaluation value. The adjustment data recording procedure records the evaluation value and the dialogue data as dialogue model adjustment data for the dialogue participant. The dialogue model adjustment procedure involves adjusting the dialogue model using the dialogue model adjustment data to perform a dialogue adapted to the dialogue with the dialogue subject. A computer-readable recording medium containing a program that assists in adjusting an interactive model to allow a computer to perform each step. The response evaluation procedure described above is: Based on the response data, the response type of the dialogue model is identified. The evaluation value of the response type is calculated, The response type is information that defines a suitable response type for each dialogue type. The recording medium according to claim 13, wherein the adjustment data recording procedure records the evaluation value, the response type, and the dialogue data as dialogue model adjustment data for the dialogue participant. The aforementioned dialogue model adjustment procedure is performed prior to the dialogue between the interlocutor and the dialogue model, The dialogue model is expected to engage with at least one type of dialogue between the dialoguers, For each of the configured dialogue types, at least one response type is defined that specifies the type of response the dialogue model will use. At least one evaluation criterion is established for evaluating the dialogue based on the aforementioned response type, The recording medium according to claim 13 or 14, wherein the dialogue type, the response type, and the evaluation criteria are provided to the dialogue model as dialogue instruction information, and the dialogue model is adjusted to perform a dialogue using the response type that assumes the dialogue type of the interlocutor. The recording medium according to claim 15, wherein the dialogue model adjustment procedure sets at least one selected from the group consisting of the dialogue type, the response type, and the evaluation criteria by providing definition generation instruction information that instructs the large-scale language model to define at least one selected from the group consisting of the dialogue type, the response type, and the evaluation criteria.

Images