Proxy action support device, proxy action support method, proxy action support program, and recording medium

Abstract

Provided is a proxy action support device based on a proxy existence model for improving the effects of proxy actions based on the proxy existence model.　A proxy action support device based on a proxy existence model according to the present invention includes: an input information acquisition unit that acquires input information of a user of the proxy existence model, input to the proxy existence model; a proxy action acquisition unit that acquires proxy action information output by the proxy existence model on the basis of the input information; and a call response output unit that outputs call response information generated by a call response model on the basis of the proxy action information.

Description

Agency Act Support Device, Agency Act Support Method, Agency Act Support Program, and Recording Medium

[0001] The present disclosure relates to an agency act support device, an agency act support method, an agency act support program, and a recording medium.

[0002] In recent years, research has been conducted on constructing dialogue systems such as chatbots using machine-learned models. For example, in Patent Document 1, an acquisition unit that acquires an inquiry sentence, and a learned answer sentence for a positive example for a learning inquiry sentence and a learned answer sentence for a negative example sampled according to the frequency distribution of the learned answer sentence are used. A calculation unit that calculates the similarity between the learning inquiry sentence and the learned answer sentence for the positive example, and a similarity between the learning inquiry sentence and the learned answer sentence for the negative example. An extraction unit that extracts one or a plurality of answer candidate sentences from the plurality of answer candidate sentences based on the similarity is described.

[0003] Japanese Patent Application Laid-Open No. 2021-124824

[0004] In the invention as described in Patent Document 1, learning is performed by giving information such as a specialized book so that answers to questions can be generated in advance. However, since the information input for learning contains the thoughts of the information producer (author) and objective knowledge (formal knowledge), the information learned by the model may be in an uncertain state. Here, if only formal knowledge is extracted in advance from information such as a specialized book and learned, the author's thoughts are not reflected, and the handling of formal knowledge in the learned model becomes opaque. Therefore, the inventors of the present disclosure devised an agent presence model that reflects the thoughts of the information producer and can output by imitating the behavior of the information producer. However, whether the user correctly understands what the agency act of the agent presence model intends is left to the user himself / herself. For this reason, there is a problem that when the intention of the agency act is not correctly conveyed to the user, the agent presence model may not be effectively used as intended by the principal (the information producer).

[0005] Therefore, this disclosure aims to provide an agency support device, an agency support method, an agency support program, and a recording medium that improve the effectiveness of agency acts based on the agency existence model.

[0006] To achieve the above objective, the proxy action support device for the proxy existence model of the present disclosure includes an input information acquisition unit, a proxy action acquisition unit, and a response output unit, wherein the input information acquisition unit acquires input information of the user of the proxy existence model that has been input to the proxy existence model, the proxy existence model is a trained model constructed to mimic a proxy target, the proxy action acquisition unit acquires proxy action information output by the proxy existence model based on the input information, the response output unit outputs response information generated by the response model based on the proxy action information, and the response model is a model trained to learn examples of dialogues of a proxy target and examples of responses of a proxy target corresponding to the dialogues, and to generate response information for the proxy action information.

[0007] The proxy action support method of the proxy existence model of this disclosure includes an input information acquisition step, a proxy action acquisition step, and a response output step, wherein each step is performed by a computer.

[0008] The proxy action support program for the proxy existence model of this disclosure includes an input information acquisition procedure, a proxy action acquisition procedure, and a response output procedure, wherein the input information acquisition procedure acquires input information of the user of the proxy existence model that has been input to the proxy existence model, the proxy existence model is a trained model constructed to mimic a proxy target, the proxy action acquisition procedure acquires proxy action information output by the proxy existence model based on the input information, the response output procedure outputs response information generated by the response model based on the proxy action information, and the response model is a model trained to learn examples of dialogues of a proxy target and examples of responses of a proxy target corresponding to the dialogues, and to generate response information for the proxy action information, and is a program to cause a computer to execute each of these procedures.

[0009] The recording medium of this disclosure includes an input information acquisition procedure, a proxy action acquisition procedure, and a response output procedure, wherein the input information acquisition procedure acquires input information of a user of the proxy existence model that has been input into the proxy existence model, the proxy existence model is a trained model constructed to mimic a proxy target, the proxy action acquisition procedure acquires proxy action information output by the proxy existence model based on the input information, the response output procedure outputs response information generated by the response model based on the proxy action information, and the response model is a model trained to learn examples of dialogues of a proxy target and examples of responses of the proxy target corresponding to the dialogues, and to generate response information for the proxy action information, and is a computer-readable recording medium that records a proxy action support program for causing a computer to execute each of these procedures.

[0010] According to this disclosure, the effectiveness of proxy actions using the proxy existence model can be improved.

[0011] Figure 1 is a block diagram showing the configuration of an example of a proxy action support device for the proxy existence model of this disclosure. Figure 2 is a block diagram showing an example of a hardware configuration of the proxy action support device for the proxy existence model of this disclosure. Figure 3 is a flowchart showing an example of processing in the proxy action support device for the proxy existence model of this disclosure. Figure 4 is a conceptual diagram illustrating an example of use of the proxy action support device for this disclosure. Figure 5 is a block diagram showing the configuration of an example of a proxy action support device for the proxy existence model of this disclosure. Figure 6 is a flowchart showing an example of processing in the proxy action support device for the proxy existence model of this disclosure. Figure 7 is a block diagram showing the configuration of an example of a proxy existence model construction device for this disclosure. Figure 8 is a block diagram showing an example of a hardware configuration of the proxy existence model construction device for this disclosure. Figure 9 is a flowchart showing an example of processing in the proxy existence model construction device for this disclosure.

[0012] In this disclosure, a “proxy existence model” is a machine learning model trained to behave, for example, like a specific proxy. The method for manufacturing a proxy existence model is not particularly limited and can be manufactured by any method, but it can be manufactured, for example, by the method for manufacturing a proxy existence model described in this disclosure, as described later.

[0013] 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.

[0014] [Embodiment 1] The proxy action support device for the proxy existence model 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 proxy action support device 10 for the proxy existence model of this embodiment. As shown in Figure 1, the proxy action support device 10 for the proxy existence model (hereinafter also referred to as "this device 10") includes an input information acquisition unit 11, a proxy action acquisition unit 12, and a response output unit 13. 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.

[0015] 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.

[0016] 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).

[0017] 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 an input information acquisition unit 11, a proxy action acquisition unit 12, and a response output unit 13. The device 10 may also include other computing devices such as a CPU, GPU (Graphics Processing Unit), APU (Accelerated Processing Unit), or a combination thereof as a computing device.

[0018] 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.

[0019] 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).

[0020] 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.

[0021] 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.

[0022] 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.

[0023] Next, an example of a proxy action support method for the proxy existence model of this embodiment will be described based on the flowchart in Figure 3. The proxy action support method for the proxy existence model of this embodiment can be implemented, for example, using the proxy action support device 10 for the proxy existence model shown in Figures 1 and 2, as follows. Note that the proxy action support method for the proxy existence model of this embodiment is not limited to the use of the proxy action support device 10 for the proxy existence model shown in Figures 1 and 2.

[0024] Prior to processing by the device 10, an interaction is performed between the surrogate existence model and its user. The user who interacts with the surrogate existence model may be the target person who was trained during the creation of the surrogate existence model, or a person other than the target person. The content of the interaction is not particularly limited as long as it is an exchange between the surrogate existence model and the user, and may be, for example, a text exchange such as a text chat, or a voice exchange such as a phone call.

[0025] The input information acquisition unit 11 of this device 10 acquires input information from the user of the proxy existence model that has been input to the proxy existence model (S1, input information acquisition step). The proxy existence model is a trained model constructed to mimic a proxy target, as described above. The input information may be, for example, character information (text information), image information, voice information, or a combination thereof. The input information may be, for example, a question from the user to the proxy target mimicked by the proxy existence model. The input information acquisition unit 11 may acquire the input information from, for example, an interface that allows the user and the proxy existence model to interact, or from log information, or it may acquire the input information from the proxy existence model.

[0026] Furthermore, the proxy action acquisition unit 12 acquires proxy action information output by the proxy existence model based on the input information (S2, proxy action acquisition step). The proxy action information may be, for example, character information (text information), image information, voice information, or a combination thereof. The proxy action information is not particularly limited as long as it is information output by the proxy existence model, and may be any kind of information. For example, if the input information is a question to the proxy target, the proxy action information is the answer to the question generated and output by the proxy existence model. The proxy action acquisition unit 12 may acquire the proxy action information from, for example, an interface that allows the user and the proxy existence model to interact, or from log information, etc., or it may acquire the proxy action information from the proxy existence model, or it may acquire the proxy action information from another device that the proxy existence model has outputted.

[0027] Note that S1 and S2 may be executed simultaneously or sequentially, and in the latter case, the order is arbitrary.

[0028] The response output unit 13 outputs response information generated by the response model based on the proxy behavior information (S3, response output step). The response information is, for example, information of the ideal dialoguer's response in response to the proxy behavior information, as assumed by the proxy subject that served as the model for the proxy existence model. The response output unit 13 can generate the response information by, for example, inputting the proxy behavior information into the response model. The response model is a model that has been trained to generate response information for the proxy behavior information by learning dialogue examples of the proxy subject and response examples of the proxy subject corresponding to the dialogue examples. Specifically, the response model may be, for example, a model in which a large-scale learning model is given the dialogue examples and response examples, as well as instruction information that instructs it to behave as the ideal dialoguer of the proxy subject. The response model is also called a self-extending model (self-extending AI) because it behaves as the ideal dialoguer that exists inside the proxy subject, for example. The large-scale learning model is, for example, a machine learning model trained using predetermined big data. The large-scale learning model may be, for example, a model trained on big data of natural language (large-scale language model), a model trained on big data of speech (large-scale speech model), or a model trained on big data of images (large-scale image model). The instruction information is, for example, instruction information (prompt) for generating an ideal response (correspondence response information) for the surrogate subject based on an example of a dialogue of the surrogate subject and an example of the surrogate subject's response corresponding to the dialogue example. The instruction information may be, but is not limited to, a document such as, "When generating text, follow the rules below: Refer to the example of the surrogate subject's dialogue and its example response; Output the response based on the intention of the surrogate action."

[0029] The response model may be configured, for example, as a multilayer neural network with a large number of parameters (e.g., 100,000 or more, and in some cases, several million or more). Specifically, the response model is trained using a training dataset that includes examples of dialogues between surrogate subjects and examples of responses from the surrogate subject corresponding to the dialogues. The training dataset may consist of multiple sets of training data that show the correspondence between surrogate behavior information output by the surrogate existence model and the response information from the ideal dialoguer that the surrogate subject expects. In training the response model, each training data is tokenized and converted into an embedding vector sequence, and a forward propagation calculation is performed using the embedding vector sequence as input to calculate a loss function such as the cross-entropy loss between the model's output and the target response information. Then, based on the value of the loss function, the model parameters are iteratively updated in mini-batch units using gradient descent or a modified algorithm therefor, thereby training the response model to output the ideal response information that the surrogate subject expects for the surrogate behavior information. In this process, the training process requires repeatedly performing a large amount of numerical calculations, including generating high-dimensional embedding vectors for each training data point, matrix multiplication, applying nonlinear activation functions, and calculating gradient backpropagation. Such processing is assumed to be performed automatically by an electronic computer equipped with processing units such as a CPU or GPU, and is impossible for a human to perform in a realistic amount of time, either mentally or with paper and pencil. Therefore, the generation process of response information by a response model relies on numerical processing specific to computer technology.

[0030] The response output unit 13 may, for example, output one response information for one proxy action information, or it may output two or more response information. In the latter case, for example, one type of response model may generate and output multiple response information, or two or more types of response models may generate and output multiple response information.

[0031] Furthermore, the response output unit 13 may, for example, estimate the user's state and output the response information according to the estimated user's state. Specifically, for example, the response information output unit 13 may acquire the elapsed time since the proxy action information from the proxy existence model was output to the user, and if the elapsed time exceeds a threshold, it may estimate that the user is in a state where response information is needed (for example, a state where the user is stuck for an answer, a state where the proxy action information is not understood, etc.) and output the response information. This makes it possible to provide appropriate support, for example, when the proxy actions of the proxy existence model reach a state where the proxy existence model alone cannot adequately handle the user's situation.

[0032] The response output unit 13 may modify the response information generated by the response model based on the input information. In this case, the response output unit 13 may, for example, estimate the difference between the response information generated based on the proxy action information and the input information corresponding to the proxy action information, and modify the response information to be generated next based on the difference. Specifically, the response output unit 13 may, for example, convert the proxy action information and the input information corresponding to the proxy action information into vectorized embedding vector sequences for each unit sentence or unit utterance, and quantify the degree of semantic difference between the two by calculating the cosine similarity or Euclidean distance between the corresponding embedding vectors. If the degree of difference exceeds a predetermined threshold, the response output unit 13 may determine that the proxy action information is insufficient as an explanation of the user's intent regarding the input information, and may add a prompt instructing the response model to explicitly explain the intent and preconditions behind the proxy action information in the response information to be generated next. The response output unit 13 can, for example, control the response information to be limited to only a concise expression of empathy when the degree of difference is below a threshold, in order to avoid increasing the burden on the user.

[0033] An example of using the device 10 will be explained using Figure 4. Figure 4 is a conceptual diagram showing the concept of a text chat being conducted by a proxy existence model (proxy AI) and its user, and its content, image arrangement, and information display method are not limited in any way to this disclosure.

[0034] As shown in Figure 4, first, user A inputs the following question as input information to a proxy existence model (proxy AI) that mimics their mentor: "I'm worried about my career. I can't find fulfillment in my work." The proxy AI acts as user A's mentor and, in response to the question, outputs the following question as proxy action information: "It must be tough when you don't find fulfillment in your work. What does work mean to you?" The device 10 acquires the input information ("I'm worried about my career. I can't find fulfillment in my work.") and the proxy action information ("It must be tough when you don't find fulfillment in your work. What does work mean to you?"), and uses a response model (self-augmenting AI) to output a response information that corresponds to the proxy action information, which is a statement made by the mentor, the proxy target: "For me, work is about making someone's tomorrow exciting." This allows user A, who received proxy action information and was unsure of the intent behind the statement (e.g., "Is it to get paid? What does it want to know?"), to realize that the proxy AI's intent in this case was ("I want you to think about how you perceive the essential meaning of work"). Furthermore, it becomes possible to input the following new information into the proxy AI: "For me, work is about making tomorrow more convenient than today."

[0035] According to this disclosure, the input information acquisition unit acquires input information from the user of the proxy existence model that has been input into the proxy existence model, the proxy action acquisition unit acquires proxy action information output by the proxy existence model based on the input information, and the response output unit outputs response information generated by the response model based on the proxy action information. As a result, for example, if the proxy existence model is unable to realize a dialogue with the user as expected by the proxy target, it can generate and output response information that is in line with the intentions of the proxy target to the user, making it easier for the user to read the intentions of the proxy target hidden in the proxy actions of the proxy existence model. Therefore, according to this disclosure, the effectiveness of proxy actions by the proxy existence model (e.g., dialogue with the user) can be improved. Furthermore, according to this disclosure, for example, the reliability of proxy actions by the proxy existence model can be improved for the proxy target. Accordingly, according to this disclosure, in dialogue between the proxy existence model and the user, the breakdown of dialogue and unnecessary re-inquiries caused by misunderstandings of replies or intentions can be reduced. In other words, by combining a surrogate existence model and a response model, the response consistency of the entire dialogue system and the usability of the user interface can be improved, and the quality of dialogue processing can be technically improved compared to general-purpose chatbots, even with the same hardware configuration.

[0036] [Embodiment 2] Embodiment 2 is another example of a proxy action support device.

[0037] The proxy action support device of this embodiment is the same as the proxy action support device 10 of Embodiment 1, except that it includes an evaluation information acquisition unit 14, an adjustment parameter determination unit 15, and a model adjustment unit 16 in addition to the configuration of the proxy action support device 10 of Embodiment 1, and the description thereof can be applied accordingly.

[0038] Figure 5 is a block diagram showing an example configuration of the proxy action support device 10A of this embodiment. As shown in Figure 5, the proxy action support device 10A includes an evaluation information acquisition unit 14, an adjustment parameter determination unit 15, and a model adjustment unit 16 in addition to the configuration of the proxy action support device 10 of Embodiment 1. The hardware configuration of the proxy action support device 10A is the same as that of the proxy action support device 10 in Figure 2, except that the central processing unit 101 has the configuration of the proxy action support device 10A in Figure 5 instead of the configuration of the proxy action support device 10 in Figure 1.

[0039] The processing of the evaluation information acquisition unit 14, the adjustment parameter determination unit 15, and the model adjustment unit 16 will be explained below with reference to Figure 6. Figure 6 is a flowchart showing an example of processing by this device 10A. The processing of the evaluation information acquisition unit 14, the adjustment parameter determination unit 15, and the model adjustment unit 16 may be appropriately inserted at any position in the flowchart of Figure 3 described in the embodiment 1 above.

[0040] First, prior to processing by the device 10A, the parameter setting criteria information is adjusted. The parameter setting criteria information is, for example, information in which the adjustment weights of the proxy existence model are set for each evaluation item of the model evaluation information described later. The adjustment weights are parameters for adjusting the weights between nodes of the trained model, and any value can be set. Specific examples of the adjustment weights include, for example, the weights of the delegation item and the reliability item in the model evaluation information (questionnaire) described later. The device 10A may, for example, record the parameter setting criteria information in the storage unit.

[0041] Furthermore, an interaction is conducted between the proxy existence model and its responder (user). The responder who interacts with the proxy existence model may be the proxy target person who was trained during the creation of the proxy existence model, or a person other than the proxy target person. The content of the interaction is not particularly limited as long as it is an exchange between the proxy existence model and the responder, and may be, for example, a text exchange such as a text chat, or an audio exchange such as a phone call.

[0042] The evaluation information acquisition unit 14 acquires model evaluation information (S11, evaluation information acquisition step). The model evaluation information includes evaluation information for the call response model. The model evaluation information is, for example, regarding the call response information output by the call response model in the interaction result with the proxy presence model, and is a questionnaire regarding the proxy-likeness of the proxy subjects that were the learning sources of the proxy presence model and the responses thereto. The model evaluation information may be, for example, a selection-type questionnaire in which scores are selected for each evaluation item, a descriptive-type questionnaire in which responses are described for each evaluation item, or a composite form of these questionnaires. Specific examples of the items include, for example, items such as delegability and reliability. The delegability is, for example, an item for confirming whether it is acceptable to entrust one's own work to the call response model, and questions such as "Can this call response model, together with one's own proxy presence model, be entrusted with part of one's own work?" can be cited. The reliability is, for example, an item for confirming the likeness of the statements intended by the proxy subject with respect to the call response model, and questions such as "Is the statement of this call response model an ideal statement for oneself?" can be cited. The evaluation information acquisition unit 14 may, for example, conduct the questionnaire on the proxy subject who has confirmed the interaction result and directly acquire the response from the proxy subject as model evaluation information, or may acquire the model evaluation information from a recording medium that has recorded the response results of the questionnaire.

[0043] The adjustment parameter determination unit 15 determines the adjustment parameters for the proxy presence model based on the model evaluation information and the parameter setting reference information (S12, adjustment parameter determination step). The adjustment parameter determination unit 15, for example, converts the questionnaire results into numerical data, multiplies the changed numerical data by the adjustment weights, and determines the adjustment parameters for the proxy presence model.

[0044] The model adjustment unit 16 adjusts the proxy presence model based on the adjustment parameters (S13, model adjustment step). The processing by the model adjustment unit 16 can be implemented, for example, in the same manner as known parameter update methods for learned models.

[0045] As described above, the call-and-response model is a model that learns, for each agent target for which the proxy presence model has been created, the dialogue example of the agent target and the answer example of the agent target corresponding to the dialogue example. On the other hand, the amount of information in the dialogue example of the agent target and the answer example of the agent target corresponding to the dialogue example is limited, and there is a problem that additional learning is difficult. According to the present disclosure, the evaluation information acquisition unit acquires model evaluation information including evaluation information for the call-and-response model, and the adjustment parameter determination unit determines the adjustment parameter for the proxy presence model based on the model evaluation information and the parameter setting reference information. The model adjustment unit can adjust the call-and-response model based on the adjustment parameter. Therefore, according to the present disclosure, additional learning of the call-and-response model with limited learning data becomes possible.

[0046] [Embodiment 3] The proxy behavior support program of the proxy presence model of the present embodiment is a program for causing a computer to execute each step of the proxy behavior support method of the proxy presence model described above. Specifically, the proxy behavior support program of the proxy presence model of the present embodiment is a program for causing a computer to execute an input information acquisition procedure, a proxy action acquisition procedure, and a call-and-response output procedure.

[0047] The input information acquisition procedure acquires the input information of the user of the proxy presence model input to the proxy presence model. The proxy presence model is a learned model constructed by imitating an agent target. The proxy action acquisition procedure acquires the proxy action information output by the proxy presence model based on the input information. The call-and-response output procedure outputs the call-and-response information generated by the call-and-response model based on the proxy action information. The call-and-response model is a model that learns the dialogue example of the agent target and the answer example of the agent target corresponding to the dialogue example, and is learned to generate call-and-response information for the proxy action information.

[0048] Further, the proxy behavior support program of the proxy presence model of the present embodiment can also be said to be a program that causes a computer to function as an input information acquisition procedure, a proxy action acquisition procedure, and a call-and-response output procedure.

[0049] The proxy action support program for the proxy existence model of this embodiment can be based on the descriptions in the proxy action support device and proxy action support method for the proxy existence model of the present disclosure. In each of the above steps, for example, "step" can be read as "process". The program of this embodiment may also be recorded on a computer-readable recording medium, for example. The recording medium is not particularly limited and includes, for example, random access memory (RAM), read-only memory (ROM), hard disk (HD), flash memory (e.g., SSD (Solid State Drive), USB flash memory, SD / SDHC card, etc.), optical disc (e.g., CD-R / CD-RW, DVD-R / DVD-RW, BD-R / BD-RE, etc.), magneto-optical disk (MO), floppy disk (FD), etc. The proxy action support program for the proxy existence model of this embodiment (for example, also referred to as a programming product or a proxy action support program product for the proxy existence model) may also be distributed from an external computer, for example. The aforementioned "distribution" may be, for example, distribution via a communication network, or distribution via a wired connected device. The proxy action support program for the proxy existence model of this embodiment may be installed and executed on the distributed device, or it may be executed without being installed.

[0050] [Embodiment 4] The proxy existence model manufacturing apparatus of this embodiment will be described with reference to Figure 7. Figure 7 is a block diagram showing the configuration of an example of the proxy existence model manufacturing apparatus 20 of this embodiment. As shown in Figure 7, the proxy existence model manufacturing apparatus 20 (hereinafter also referred to as "this apparatus 20") includes a knowledge information acquisition unit 21, a construction information extraction unit 22, and a model construction unit 23. In addition, although not shown, this apparatus 20 may also include, for example, an input unit, an output unit, a display unit and / or a storage unit.

[0051] The device 20 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 20 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 20 may, for example, be incorporated into a server as a system. Alternatively, the device 20 may be, for example, a personal computer (PC, e.g., desktop or notebook), smartphone, or tablet terminal on which the program disclosed herein is installed. Furthermore, the device 20 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.

[0052] Figure 8 illustrates a block diagram of the hardware configuration of the device 20. The device 20 includes, for example, a central processing unit 201, memory 202, bus 203, storage device 204, input device 205, output device 206, communication device (communication unit) 207, etc. Each part of the device 20 is interconnected via the bus 203 through its respective interface (I / F).

[0053] The central processing unit 201 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 20, the central processing unit 201 executes, for example, the program of this disclosure and other programs, and also reads and writes various types of information. Specifically, for example, the central processing unit 201 functions as a knowledge information acquisition unit 21, a construction information extraction unit 22, and a model construction unit 23. The device 20 may also include other computing devices such as a CPU, GPU (Graphics Processing Unit), APU (Accelerated Processing Unit), or a combination thereof as computing devices.

[0054] Bus 203 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 20 can be connected to an external network (the aforementioned communication network) by a communication device 207 connected to bus 203, for example, and can also be connected to other devices via the external network.

[0055] Memory 202 may be, for example, main memory. When the central processing unit 201 performs processing, memory 202 reads various operational programs, such as the program of this disclosure, stored in the storage device 204 (described later), and the central processing unit 201 receives data from memory 202 and executes the program. The main memory may be, for example, RAM (random access memory). Alternatively, memory 202 may be, for example, ROM (read-only memory).

[0056] The storage device 204 is also called an auxiliary storage device, for example, in relation to the main memory (primary memory). As described above, the storage device 204 stores an operating program including the program of this disclosure. The storage device 204 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 20 includes, for example, the storage device 204 functions as the storage unit.

[0057] In this device 20, the memory 202 and storage device 204 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 20, and information used by this device 20 when executing processing. At least some of the information may be stored, for example, on an external server other than the memory 202 and storage device 204, or distributed and stored across multiple terminals using blockchain technology or the like.

[0058] The device 20 further includes, for example, an input device 205 and an output device 206. The input device 205 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 206 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 3, the input device 205 and the output device 206 are configured separately, but the input device 205 and the output device 206 may be configured as an integrated unit, such as a touch panel display.

[0059] Next, an example of the method for manufacturing a surrogate existence model according to this embodiment will be described based on the flowchart in Figure 9. The method for manufacturing a surrogate existence model according to this embodiment can be carried out as follows, for example, using the surrogate existence model manufacturing apparatus 20 shown in Figures 7 and 8. Note that the method for manufacturing a surrogate existence model according to this embodiment is not limited to the use of the surrogate existence model manufacturing apparatus 20 shown in Figures 7 and 8.

[0060] First, the knowledge information acquisition unit 21 acquires the subject's knowledge information (S21, knowledge information acquisition step). The format of the knowledge information is not particularly limited; for example, it may be text information, image information, audio information, or a combination thereof. The knowledge information is, for example, information linked to predetermined information and subject identification information that identifies the creator of the information (the subject). The predetermined information is, for example, information that includes the subject's personality information and explicit knowledge information. The personality information is, for example, information that represents the subject's thoughts from the knowledge information. The personality information is also called, for example, a partial stance. The explicit knowledge information is, for example, the part of the knowledge information excluding the personality information, and includes objective knowledge. If the knowledge information is a book, paper, etc., the explicit knowledge information may, for example, be information such as technical terms and experimental results, but is not limited to these. The subject identification information may, for example, be a name, address, telephone number, email address, identification number (for example, My Number (individual number), etc.). Specific examples of the aforementioned knowledge information include, but are not limited to, books and papers written by the subject, video data of lectures given by the subject, audio data of lectures given by the subject, and image data created by the subject. The knowledge information acquisition unit 21 may, for example, acquire knowledge information recorded in the storage unit of the device 20, or it may acquire the aforementioned knowledge information from outside the device 20 via the input device 205. The knowledge information acquisition unit 21 may, for example, acquire one type of knowledge information of the subject, or it may acquire two or more types. The knowledge information acquisition unit 21 may, for example, record the acquired knowledge information in the storage unit of the device 20.

[0061] The information extraction unit 22 for construction extracts personality information of the subject from the knowledge information (S22, information extraction step for construction). The information extraction unit 22 for construction may also, for example, further extract explicit knowledge information from the knowledge information of the subject. The information extraction unit 22 for construction may, for example, use known natural language processing techniques to extract at least one of the personality information and explicit knowledge information of the subject, or use a large-scale language model to extract at least one of the personality information and explicit knowledge information of the subject. If the knowledge information is textual information such as a book or a paper, the information extraction unit 22 for construction can extract the personality information or explicit knowledge information based, for example, on the end of sentences in a document or the chapter structure of a book. The information extraction unit 22 for construction may, for example, record at least one of the extracted personality information and explicit knowledge information in the storage unit of the device 20. In this case, the information extraction unit 22 for construction can, for example, record the personality information and explicit knowledge information linked to the knowledge information from which they were extracted and the creator identification information of the knowledge information.

[0062] The information extraction unit 22 for construction can, for example, analyze the knowledge information and extract sentences whose sentence ends with a word that expresses the author's thoughts as the personality information. Examples of words that express the author's thoughts include, but are not limited to, words such as "I want to," "I think," "I believe," and "I want." The information extraction unit 22 for construction can also analyze the knowledge information and extract sentences whose sentence ends with a word that indicates explicit knowledge as the explicit knowledge information. Examples of words that indicate explicit knowledge include, but are not limited to, words such as "It is," "It was," and "As a result."

[0063] The information extraction unit 22 for construction may, for example, analyze the knowledge information and extract sentences contained in chapters that describe the author's ideas as personality information. Examples of chapters that describe the author's ideas include the "preface," "introduction," and "afterword."

[0064] The information acquisition unit 22 for construction may, for example, use AI to extract personality information and explicit knowledge information from knowledge information. In this disclosure, AI may refer to, for example, a large-scale learning model called a "foundation model." The foundation model is a machine learning model pre-trained on predetermined big data and is not limited to a large-scale language model (LLM) that has learned natural language, but may also include a large-scale model for speech, a large-scale model for images, and a multimodal model (such as a visual language model) that handles language, images, speech, and video across the board. Furthermore, a configuration may be adopted in which a small-scale language model (SLM) is placed on the terminal side and cooperates 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 call, agent-oriented control logic, etc. The providers of large-scale language models are not particularly limited. Examples include, but are not limited to, various LLM / multimodal models provided by companies such as OpenAI, Anthropique, Alphabet (Google), META, Microsoft, Cohere, Mistral, xAI, NEC Corporation, and NTT.

[0065] When the information acquisition unit 22 for construction uses AI to extract personality information and explicit knowledge information from knowledge information, the information acquisition unit 22 can cause the AI ​​to extract the user's personality information and explicit knowledge information by inputting extraction instruction information, which instructs the AI ​​to extract personality information and explicit knowledge information, along with the knowledge information. The extraction instruction information may be recorded in the storage unit of the device 10, stored externally, or input by the user each time. Specific examples of the extraction instruction information include, for example, "classifying information into the following two categories: sentences and paragraphs that show the author's thoughts and opinions, and sentences and paragraphs that show knowledge, such as scientific verification results and historical facts." This disclosure is not limited to the above examples.

[0066] Furthermore, the AI ​​may be, for example, a model finely tuned to extract personality information and explicit knowledge information from knowledge information. In this case, for example, fine tuning for extracting personality information and explicit knowledge information from knowledge information can be performed by training the AI ​​with knowledge information and a set of personality information and explicit knowledge information previously extracted from the knowledge information. In this case, the AI ​​may be, for example, a multilayer neural network model having a large number of parameters (e.g., 100,000 or more, and in some cases, several million or more). Such fine tuning processing requires the rapid and repeated execution of a huge amount of numerical calculations, making it practically impossible for a human to perform it in their head or with paper and pencil, and thus relies on automated processing by electronic computers such as processors and GPUs. During the fine tuning, in order to suppress overfitting of the training data, known regularization methods such as L2 regularization, dropout, and early stopping may be combined and applied. This makes it possible to obtain a model that can extract personality information and explicit knowledge information with high generalization performance even from unknown knowledge information, and improves the accuracy and robustness of information extraction compared to simple threshold judgment or rule-based processing.

[0067] Furthermore, the information extraction unit 22 for construction may extract the personality information of the subject by, for example, providing a large-scale language model with the subject's knowledge information and instruction information that instructs the model to extract the subject's personality information from the knowledge information based on the subject's knowledge information, thereby causing the model to extract the personality information from the knowledge information. The large-scale language model is not particularly limited and includes, but is not limited to, OpenAI®'s GPT-3®, GPT-4®, Alphabet Inc. (Google®)'s BERT, LaMDA, PaLM2, META®'s LlaMA, NEC Corporation's LLM, NTT®'s LLM, etc. The instruction information that instructs the model to extract the subject's personality information from the knowledge information based on the subject's knowledge information is not particularly limited as long as it is a document that instructs the model to divide the knowledge information into parts that contain thoughts and parts that contain knowledge. Specific examples of instruction information that instructs the extraction of personality information of the subject from the knowledge information of the subject include, but are not limited to, documents such as, "Classify the information into the following two categories: - Sentences and paragraphs that show the author's thoughts and opinions, such as the author's ideas; - Sentences and paragraphs that show knowledge, such as scientific verification results and historical facts."

[0068] The information extraction unit 22 for construction may, for example, preprocess the knowledge information and extract personality information and explicit knowledge information. Specifically, for example, the information extraction unit 22 for construction may convert the knowledge information into a vectorized embedding vector sequence for each unit text, and then classify or group the knowledge information into personality information and explicit knowledge information based on the embedding vector sequence, thereby extracting personality information and explicit knowledge information based on the knowledge information. The embedding vectors may, for example, contain hundreds to thousands of real-valued elements for each unit text. For this reason, for example, a large number of embedding vectors are generated for the entire knowledge information, and the classification or grouping process consists of a large number of numerical operations, mainly matrix operations, which is a process that is practically impossible for a human to perform in their head or with paper and pencil.

[0069] The model building unit 23 constructs a surrogate existence model that mimics the target person based on the personality information (S23, model building step). The model building unit 23 can construct the surrogate existence model by, for example, providing a large-scale learning model with the personality information and instruction information that instructs the large-scale learning model to construct a surrogate existence model that mimics the target person based on the personality information. The large-scale learning model is, 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). The model building unit 23 can construct the surrogate existence model by, for example, providing a large-scale language model, as the large-scale learning model, with the personality information and instruction information that instructs the large-scale learning model to construct a surrogate existence model that mimics the target person based on the personality information. The instruction information is, for example, instruction information (prompt) for generating the behavior of a person handling knowledge. Examples of the aforementioned instructional information include, but are not limited to, documents such as, "When generating text, follow the rules below: - Describe examples that reflect personal information. - Do not reflect personal information in terms of knowledge. - Structure the text such that, for example, you state a fact that is correct as knowledge, and then express an opinion on that fact as personal information."

[0070] Furthermore, the model building unit 23 may generate multiple sets of training data showing the correspondence between the subject's input information and responses based on the extracted personality information. The model building unit 23 may input the training data in mini-batch units into a large-scale learning model, calculate a loss function based on the error between the model's output for the training data and the response, and train a personality model capable of outputting behavior that mimics the subject by iteratively updating a large number of model parameters using gradient descent or a modified algorithm thereof.

[0071] Furthermore, the model building unit 23 may, for example, construct an explicit knowledge model by further training a large-scale language model with the explicit knowledge information. The model building unit 23 can construct the explicit knowledge model by, for example, providing the explicit knowledge information to the large-scale language model and fine-tuning it. The explicit knowledge model is also called, for example, a large-scale language model with domain knowledge.

[0072] The aforementioned surrogate existence model is, for example, a model that has learned the personality information of the subject from the information that constitutes the knowledge information. Therefore, the surrogate existence model has learned, for example, the subject's way of thinking and understanding when handling knowledge, values ​​such as thoughts and beliefs, and how they interact with others (personality). For this reason, the surrogate existence model manufacturing method of this disclosure makes it possible to easily manufacture a model that reflects the thoughts of the information creator. Furthermore, the surrogate existence model manufactured by the surrogate existence model manufacturing method of this disclosure is capable of outputting products that are characteristic of the subject. In other words, according to this disclosure, for example, it becomes possible to construct a model that suppresses personal hallucination. Personal hallucination refers to, for example, a hallucination (illusion) of how to handle knowledge and output policies that the creator of the knowledge information would not say. Therefore, according to the surrogate existence model of this disclosure, for example, it becomes possible to more accurately extract and output the knowledge that knowledge information (e.g., a book) explicitly or implicitly contains. The output that is characteristic of the target person is not particularly limited and may be, for example, text output, voice output, or instructions to a designated machine.

[0073] The model building unit 23 may, for example, use extracted personality information to train a personality model capable of outputting behavior that mimics the subject by iteratively updating the parameters of a large-scale learning model having many parameters based on gradient descent to reproduce the correspondence between the subject's past input information and responses. Similarly, the model building unit 23 may, for example, use extracted explicit knowledge information to train an explicit knowledge model that outputs explicit knowledge information contained in the knowledge information. In this case, the model building unit 23 can, for example, generate proxy behavior information consistent with the subject's personality and output the proxy behavior information by coordinating the trained personality model and the explicit knowledge model. By configuring the personality model and the explicit knowledge model separately in this way, and coordinating the output of explicit knowledge based on the knowledge information with the output based on the personality information, it is possible to reduce the inclusion of personal expressions not included in the knowledge information and improve the consistency and reliability of the generated responses, compared to, for example, simply providing prompts to a general-purpose large-scale language model.

[0074] According to this disclosure, it is possible to create models that reflect the thoughts of the information creators. Therefore, according to this disclosure, for example, it is possible to construct surrogate models for individuals with limited human resources (e.g., busy researchers, supervisors, managers, teachers, etc.), thereby reducing the burden on those individuals.

[0075] 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.

[0076] This application claims priority based on Japanese Patent Application No. 2024-229569, filed on 26 December 2024, and incorporates all of its disclosures herein.

[0077] <Note> Some or all of the above embodiments may be described as follows, but are not limited to the following. (Note 1) A proxy action support device for a proxy model, comprising an input information acquisition unit, a proxy action acquisition unit, and a response output unit, wherein the input information acquisition unit acquires input information of the user of the proxy model that has been input to the proxy existence model, the proxy existence model is a trained model constructed to mimic a proxy target, the proxy action acquisition unit acquires proxy action information output by the proxy existence model based on the input information, the response output unit outputs response information generated by the response model based on the proxy action information, and the response model is a model trained to learn examples of dialogues of a proxy target and examples of responses of a proxy target corresponding to the dialogues, and to generate response information for the proxy action information. (Note 2) The proxy action support device according to Note 1, wherein the response model is a model that has been given the dialogue examples and response examples, as well as instruction information that instructs the large-scale learning model to behave as the ideal dialoguer of the surrogate target. (Note 3) The proxy action support device according to Note 1 or 2, wherein the response output unit modifies the response information generated by the response model based on the input information. (Note 4) The proxy action support device according to Note 3, wherein the response output unit estimates the difference between the response information generated based on the surrogate action information and the input information corresponding to the surrogate action information, and modifies the next response information to be generated based on the difference. (Note 5) The proxy action support device according to any one of Notes 1 to 4, further comprising an evaluation information acquisition unit, an adjustment parameter determination unit, and a model adjustment unit, wherein the evaluation information acquisition unit acquires model evaluation information, the model evaluation information includes evaluation information for a response model, the adjustment parameter determination unit determines adjustment parameters for the response model based on the model evaluation information and parameter setting criterion information, and the model adjustment unit adjusts the response model based on the adjustment parameters.(Note 6) A method for supporting the proxy actions of a proxy model, comprising an input information acquisition step, a proxy action acquisition step, and a response output step, wherein the input information acquisition step acquires input information of the user of the proxy model that has been input into the proxy existence model, the proxy existence model is a trained model constructed to mimic a proxy target, the proxy action acquisition step acquires proxy action information output by the proxy existence model based on the input information, the response output step outputs response information generated by the response model based on the proxy action information, and the response model is a model trained to learn dialogue examples of the proxy target and response examples of the proxy target corresponding to the dialogue examples, and to generate response information for the proxy action information, each step of which is performed by a computer. (Note 7) The method for supporting the proxy actions of a proxy model according to Note 6, wherein the response model is a model that has been given the dialogue examples and response examples, as well as instruction information that instructs the large-scale learning model to behave as the ideal dialoguer of the proxy target. (Note 8) The proxy action support method according to Note 6 or 7, wherein the response output step modifies the response information to be generated by the response model based on the input information. (Note 9) The proxy action support method according to Note 8, wherein the response output step estimates the difference between the response information generated based on the proxy action information and the input information corresponding to the proxy action information, and modifies the response information to be generated next based on the difference. (Note 10) The proxy action support method according to any one of Notes 6 to 9, further comprising an evaluation information acquisition step, an adjustment parameter determination step, and a model adjustment step, wherein the evaluation information acquisition step acquires model evaluation information, the model evaluation information includes evaluation information for the response model, the adjustment parameter determination step determines adjustment parameters for the response model based on the model evaluation information and parameter setting criterion information, and the model adjustment step adjusts the response model based on the adjustment parameters.(Note 11) A proxy action support program for a proxy model that causes a computer to execute each of the following procedures: (Note 12) The proxy action support program for a proxy model, which includes an input information acquisition procedure, a proxy action acquisition procedure, and a response output procedure, wherein the input information acquisition procedure acquires input information of the user of the proxy model that has been input into the proxy model, the proxy model is a trained model constructed to mimic a proxy target, the proxy action acquisition procedure acquires proxy action information output by the proxy model based on the input information, the response output procedure outputs response information generated by the response model based on the proxy action information, and the response model is a model trained to learn dialogue examples of the proxy target and response examples of the proxy target corresponding to the dialogue examples, and to generate response information for the proxy action information. (Note 13) The proxy action support program according to Note 11 or 12, wherein the response output procedure causes the response model to modify the response information to be generated based on the input information. (Note 14) The proxy action support program according to Note 13, wherein the response output procedure estimates the difference between the response information generated based on the proxy action information and the input information corresponding to the proxy action information, and modifies the response information to be generated next based on the difference. (Note 15) The proxy action support program according to any one of Notes 11 to 14, further comprising an evaluation information acquisition procedure, an adjustment parameter determination procedure, and a model adjustment procedure, wherein the evaluation information acquisition procedure acquires model evaluation information, the model evaluation information includes evaluation information for the response model, the adjustment parameter determination procedure determines adjustment parameters for the response model based on the model evaluation information and parameter setting criterion information, and the model adjustment procedure adjusts the response model based on the adjustment parameters.(Note 16) A computer-readable recording medium recording a proxy action support program for a proxy existence model, each procedure of which is executed by a computer. (Note 17) The recording medium according to Note 16, wherein the response-response model is a large-scale learning model that has been given input information from a user of the proxy existence model, and the proxy existence model is a trained model constructed to mimic a proxy target, the proxy action acquisition procedure acquires proxy action information output by the proxy existence model based on the input information, and the response-response output procedure outputs response-response information generated by the response-response model based on the proxy action information, and the response-response model is a model that has been trained to learn examples of dialogues of a proxy target and examples of responses from the proxy target corresponding to the dialogues, and to generate response-response information for the proxy action information. (Note 18) The recording medium according to Note 16 or 17, wherein the response output procedure causes the response model to modify the response information to be generated based on the input information. (Note 19) The recording medium according to Note 18, wherein the response output procedure estimates the difference between the response information generated based on the proxy action information and the input information corresponding to the proxy action information, and modifies the response information to be generated next based on the difference. (Note 20) The recording medium according to any one of Notes 16 to 19, further comprising an evaluation information acquisition procedure, an adjustment parameter determination procedure, and a model adjustment procedure, wherein the evaluation information acquisition procedure acquires model evaluation information, the model evaluation information includes evaluation information for the response model, the adjustment parameter determination procedure determines adjustment parameters for the response model based on the model evaluation information and parameter setting criterion information, and the model adjustment procedure adjusts the response model based on the adjustment parameters.

[0078] This disclosure can improve the effectiveness of surrogate behavior using surrogate existence models. Therefore, this disclosure is useful in a wide range of industries that utilize surrogate existence models.

[0079] 10, 10A Proxy Act Support Device for Proxy Existence Model 11 Input Information Acquisition Unit 12 Proxy Action Acquisition Unit 13 Response Information Output Unit 14 Evaluation Information Acquisition Unit 15 Adjustment Parameter Determination Unit 16 Model Adjustment Unit 101 Central Processing Unit 102 Memory 103 Bus 104 Storage Device 105 Input Device 106 Output Device 107 Communication Device 20 Proxy Existence Model Construction Device 21 Knowledge Information Acquisition Unit 22 Construction Information Extraction Unit 23 Model Construction Unit 201 Central Processing Unit 202 Memory 203 Bus 204 Storage Device 205 Input Device 206 Output Device 207 Communication Device

Claims

1. A proxy action support device for a proxy model, comprising an input information acquisition unit, a proxy action acquisition unit, and a response output unit, wherein the input information acquisition unit acquires input information of the user of the proxy model that has been input to the proxy existence model, the proxy existence model is a trained model constructed to mimic a proxy target, the proxy action acquisition unit acquires proxy action information output by the proxy existence model based on the input information, the response output unit outputs response information generated by the response model based on the proxy action information, and the response model is a model trained to learn examples of dialogues of a proxy target and examples of responses of the proxy target corresponding to the dialogues, and to generate response information for the proxy action information.

2. The proxy action support device according to claim 1, wherein the response model is a model that has been given the dialogue examples and response examples, as well as instruction information that instructs the model to behave as the ideal dialogue partner of the proxy target.

3. The proxy action support device according to claim 1 or 2, wherein the response output unit modifies the response information to be generated by the response model based on the input information.

4. The proxy action support device according to claim 3, wherein the response output unit estimates the difference between the response information generated based on the proxy action information and the input information corresponding to the proxy action information, and modifies the response information to be generated next based on the difference.

5. The agent support device according to any one of claims 1 to 4, further comprising an evaluation information acquisition unit, an adjustment parameter determination unit, and a model adjustment unit, wherein the evaluation information acquisition unit acquires model evaluation information, the model evaluation information includes evaluation information for a response model, the adjustment parameter determination unit determines adjustment parameters for the response model based on the model evaluation information and parameter setting criterion information, and the model adjustment unit adjusts the response model based on the adjustment parameters.

6. A method for supporting the proxy actions of a proxy model, comprising an input information acquisition step, a proxy action acquisition step, and a response output step, wherein the input information acquisition step acquires input information of the user of the proxy model that has been input into the proxy model, the proxy model is a trained model constructed to mimic a proxy target, the proxy action acquisition step acquires proxy action information output by the proxy model based on the input information, the response output step outputs response information generated by the response model based on the proxy action information, and the response model is a model trained to learn examples of dialogues of a proxy target and examples of responses of the proxy target corresponding to the dialogues, and to generate response information for the proxy action information, each step of which is performed by a computer.

7. The proxy action support method according to claim 6, wherein the response model is a model that has been given the dialogue examples and response examples, as well as instruction information that instructs the large-scale learning model to behave as the ideal dialoguer of the proxy target.

8. The proxy action support method according to claim 6 or 7, wherein the response output step modifies the response information to be generated by the response model based on the input information.

9. The proxy action support method according to claim 8, wherein the response output step estimates the difference between the response information generated based on the proxy action information and the input information corresponding to the proxy action information, and modifies the response information to be generated next based on the difference.

10. The method for supporting an agent action according to any one of claims 6 to 9, further comprising an evaluation information acquisition step, an adjustment parameter determination step, and a model adjustment step, wherein the evaluation information acquisition step acquires model evaluation information, the model evaluation information includes evaluation information for a response model, the adjustment parameter determination step determines adjustment parameters for the response model based on the model evaluation information and parameter setting criterion information, and the model adjustment step adjusts the response model based on the adjustment parameters.

11. A proxy action support program for a proxy model, which causes a computer to execute each of the following steps: input information acquisition step, proxy action acquisition step, and response output step, wherein the input information acquisition step acquires input information of the user of the proxy model that has been input into the proxy model, the proxy model is a trained model constructed to mimic a proxy target, the proxy action acquisition step acquires proxy action information output by the proxy model based on the input information, the response output step outputs response information generated by the response model based on the proxy action information, and the response model is a model trained to learn examples of dialogues of a proxy target and examples of responses from the proxy target corresponding to the dialogues, and to generate response information for the proxy action information.

12. The proxy action support program according to claim 11, wherein the response model is a model that has been given the dialogue examples and response examples, as well as instruction information that instructs the model to behave as the ideal dialoguer of the proxy target.

13. The proxy action support program according to claim 11 or 12, wherein the response output procedure causes the response model to modify the response information to be generated based on the input information.

14. The proxy action support program according to claim 13, wherein the response output procedure estimates the difference between the response information generated based on the proxy action information and the input information corresponding to the proxy action information, and modifies the response information to be generated next based on the difference.

15. The agency support program according to any one of claims 11 to 14, further comprising an evaluation information acquisition procedure, an adjustment parameter determination procedure, and a model adjustment procedure, wherein the evaluation information acquisition procedure acquires model evaluation information, the model evaluation information includes evaluation information for a response model, the adjustment parameter determination procedure determines adjustment parameters for the response model based on the model evaluation information and parameter setting criterion information, and the model adjustment procedure adjusts the response model based on the adjustment parameters.

16. A computer-readable recording medium recording a proxy action support program for a proxy existence model, each procedure of which is executed by a computer. The program includes an input information acquisition procedure, a proxy action acquisition procedure, and a response output procedure, wherein the input information acquisition procedure acquires input information of a user of the proxy existence model that has been input into the proxy existence model, the proxy existence model is a trained model constructed to mimic a proxy target, the proxy action acquisition procedure acquires proxy action information output by the proxy existence model based on the input information, the response output procedure outputs response information generated by the response model based on the proxy action information, and the response model is a model trained to learn examples of dialogues of a proxy target and examples of responses from the proxy target corresponding to the dialogues, and to generate response information for the proxy action information.

17. The recording medium according to claim 16, wherein the response model is a model that has been given the dialogue examples and response examples, as well as instruction information that instructs the large-scale learning model to behave as the ideal dialoguer of the surrogate target.

18. The recording medium according to claim 16 or 17, wherein the response output procedure modifies the response information to be generated by the response model based on the input information.

19. The recording medium according to claim 18, wherein the response output procedure estimates the difference between the response information generated based on the proxy action information and the input information corresponding to the proxy action information, and modifies the response information to be generated next based on the difference.

20. The recording medium according to any one of claims 16 to 19, further comprising a procedure for acquiring evaluation information, a procedure for determining adjustment parameters, and a procedure for adjusting a model, wherein the procedure for acquiring evaluation information acquires model evaluation information, the model evaluation information includes evaluation information for a response model, the procedure for determining adjustment parameters for the response model is based on the model evaluation information and parameter setting criterion information, and the procedure for adjusting the response model is based on the adjustment parameters.

Images