Information processing method, information processing program, and information processing device

Abstract

To provide an information processing method, information processing program, and information processing device for generating and outputting multiple prompts for causing a language model-based chat system to behave as an agent.SOLUTION: An information processing method disclosed herein comprises: acquiring multiple elements to be bases of prompts including instructions regarding behavior of a language model; generating multiple feature definitions by combining the multiple mutually different elements; generating multiple prompts, based on each of the generated feature definitions, to be fed to the language model; and outputting the generated multiple prompts.SELECTED DRAWING: Figure 10

Description

[Technical Field]

[0001] The present invention relates to an information processing method, an information processing program, and an information processing device that create and output a plurality of prompts to be given to a language model. [Background technology]

[0002] In recent years, the use of digital twins has been increasing. Digital twins are a mechanism and concept that allows for the placement of real-world data in a virtual space like a twin, enabling monitoring and simulation. Until now, digital twins have been used to describe physical objects, environments, or facilities, but there have been proposals to extend this to include the reproduction of people (for example, Non-Patent Document 1).

[0003] Furthermore, Patent Document 1 proposes an expert answer prediction system that includes a learning model storage unit that stores a learning model constructed to be able to predict answers that a personal expert will give to questions by learning from the life log of the personal expert as a digital clone of the personal expert. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2023-82970 [Non-patent literature]

[0005] [Non-Patent Document 1] Mitsunori Tada and two others, "Current Status and Future Prospects of Human Digital Twins," Journal of the Robotics Society of Japan, [online], September 15, 2022, Robotics Society of Japan, Vol. 40, No. 7, pp. 579-584, Internet<URL:https: / / www.jstage.jst.go.jp / article / jrsj / 40 / 7 / 40_40_579 / _pdf / -char / ja> Summary of the Invention [Problem to be solved by the invention]

[0006] However, Non-Patent Document 1 does not disclose how to specifically build a human digital twin that replicates a person. Furthermore, the digital clone disclosed in Patent Document 1 requires the construction of a learning model through supervised learning.

[0007] The present invention has been made in light of the above circumstances, and its purpose is to provide an information processing method, an information processing program, and an information processing device that create and output multiple prompts for making a chat system that uses a language model behave as an agent, without building a learning model through supervised learning. [Means for solving the problem]

[0008] An information processing method according to one aspect of the present application includes obtaining a plurality of elements that form the basis of prompts containing instructions regarding the behavior of a language model, creating a plurality of characteristic definitions that combine the elements that are different from each other, creating a plurality of prompts based on each of the created characteristic definitions to be applied to the language model, and outputting the created plurality of prompts. [Effects of the Invention]

[0009] In one aspect of the present application, it is possible to create and output a plurality of prompts that cause a chat system using a language model to behave as an agent. [Brief explanation of the drawings]

[0010] [Figure 1] FIG. 1 is an explanatory diagram illustrating an example of the configuration of a dialogue system. [Figure 2] FIG. 2 is a block diagram illustrating an example of a hardware configuration of a dialogue server. [Figure 3] FIG. 2 is a block diagram illustrating an example of the hardware configuration of a user terminal. [Figure 4] FIG. 2 is a conceptual diagram showing an example of the contents of a user personality DB. [Figure 5] FIG. 10 is a conceptual diagram showing an example of the contents of a knowledge DB. [Figure 6] FIG. 10 is an explanatory diagram showing an example of an agent personality DB. [Figure 7] FIG. 10 is an explanatory diagram showing an example of an ad-lib DB. [Figure 8] 10 is a flowchart illustrating an example of a procedure for initial setting processing. [Figure 9] 10 is a flowchart illustrating an example of a procedure for interactive processing. [Figure 10] 10 is a flowchart illustrating an example of a procedure for an inference process. [Figure 11] FIG. 10 is an explanatory diagram showing an example of an object and an example of a question. [Figure 12] FIG. 10 is an explanatory diagram showing an example of a first prompt and an answer obtained in response to the first prompt. [Figure 13] FIG. 10 is an explanatory diagram showing an example of a second prompt and an answer obtained in response to the second prompt. [Figure 14] FIG. 10 is an explanatory diagram showing an example of a third prompt and an answer obtained in response to the third prompt. [Figure 15] FIG. 10 is an explanatory diagram showing an example of a fourth prompt and an answer obtained in response to the fourth prompt. [Figure 16] FIG. 10 is an explanatory diagram showing an example of a dialogue screen. [Figure 17] FIG. 10 is an explanatory diagram showing an example of an initial setting screen. DETAILED DESCRIPTION OF THE INVENTION

[0011] An embodiment will be described below with reference to the drawings. Fig. 1 is an explanatory diagram showing an example of the configuration of a dialogue system. The dialogue system 100 includes a dialogue server 1, a user terminal 2, and a dialogue AI 3. The dialogue server 1, the user terminal 2, and the dialogue AI 3 are connected to each other via a network N so that they can communicate with each other.

[0012] An end user of the dialogue system 100 in this specification can virtually experience a text-based conversation with a real person by engaging in text-based chat with an agent built into a chatbot.

[0013] In this embodiment, an agent is a software or the like that imitates an expert with knowledge and expertise in a particular field and outputs the answer that the expert is predicted to give in response to a given question.

[0014] The dialogue server 1 cooperates with the dialogue AI 3 to provide a chatbot that enables chatting with an agent. The dialogue server 1 is composed of a server computer, a workstation, a PC (Personal Computer), etc. The dialogue server 1 may also be composed of a multi-computer consisting of multiple computers, a virtual machine virtually constructed by software, or a quantum computer. Furthermore, the functions of the dialogue server 1 may be realized by a cloud service.

[0015] The user terminal 2 is a terminal used by an end user (user). The user terminal 2 is configured as a smartphone, a tablet computer, a notebook computer, etc. Although only one user terminal 2 is shown in FIG. 1, there may be two or more terminals.

[0016] The conversational AI 3 is a generation AI (Artificial Intelligence) that provides a chatbot. The control unit 31 generates an answer to a question using a language model M. The conversational AI 3 may be incorporated into the dialogue server 1.

[0017] 2 is a block diagram showing an example of the hardware configuration of the dialogue server 1. The dialogue server 1 includes a control unit 11, a main memory unit 12, an auxiliary memory unit 13, a communication unit 14, and a reading unit 15. Each component is connected by a bus B.

[0018] The control unit 11 has one or more arithmetic processing devices such as a CPU (Central Processing Unit), an MPU (Micro-Processing Unit), a GPU (Graphics Processing Unit), etc. The control unit 11 reads and executes a control program 1P (program, program product) stored in the auxiliary storage unit 13, thereby performing various information processing, control processing, etc. related to the dialogue server 1 and realizing various functional units.

[0019] The main memory unit 12 is a static random access memory (SRAM), a dynamic random access memory (DRAM), a flash memory, etc. The main memory unit 12 mainly temporarily stores data required for the control unit 11 to execute arithmetic processing.

[0020] The auxiliary storage unit 13 is a hard disk or an SSD (Solid State Drive) or the like, and stores the control program 1P and various DBs (Databases) required for the control unit 11 to execute processing. The auxiliary storage unit 13 stores a user personality DB 131, a knowledge DB 132, an agent personality DB 133, and an ad-lib DB 134. The auxiliary storage unit 13 may be separate from the dialogue server 1 and may be an external storage device connected externally. The various DBs and the like stored in the auxiliary storage unit 13 may be stored in a database server or cloud storage different from the dialogue server 1.

[0021] The communication unit 14 communicates with the user terminal 2 and the interactive AI 3 via the network N. In addition, the control unit 11 may use the communication unit 14 to download the control program 1P from another computer via the network N, etc., and store it in the auxiliary memory unit 13.

[0022] The reading unit 15 reads the portable storage medium 1a including a CD (Compact Disc)-ROM and a DVD (Digital Versatile Disc)-ROM. The control unit 11 may read the control program 1P from the portable storage medium 1a via the reading unit 15 and store it in the auxiliary storage unit 13. The control unit 11 may also read the control program 1P from the semiconductor memory 1b.

[0023] 3 is a block diagram showing an example of the hardware configuration of a user terminal 2. The user terminal 2 includes a control unit 21, a main memory unit 22, an auxiliary memory unit 23, a communication unit 24, a display panel 25, and an operation unit 26. Each component is connected by a bus B.

[0024] The control unit 21 has one or more arithmetic processing units such as a CPU, an MPU, a GPU, etc. The control unit 21 provides various functions by reading and executing a control program 2P (program, program product) stored in the auxiliary storage unit 23.

[0025] The main memory unit 22 is an SRAM, a DRAM, a flash memory, etc. The main memory unit 22 mainly temporarily stores data necessary for the control unit 21 to execute arithmetic processing.

[0026] The auxiliary storage unit 23 is a hard disk or SSD, etc., and stores various data necessary for the control unit 21 to execute processing. The auxiliary storage unit 23 may be an external storage device that is separate from the user terminal 2 and externally connected. The various DBs, etc. stored in the auxiliary storage unit 23 may be stored in a database server or cloud storage.

[0027] The communication unit 24 communicates with the dialogue server 1 via the network N. The control unit 21 may use the communication unit 24 to download the control program 2P from another computer via the network N or the like and store it in the auxiliary storage unit 23.

[0028] The display panel 25 can be configured with a liquid crystal panel, an organic EL (Electro Luminescence) display, or the like. The operation unit 26 can be configured with, for example, a touch panel incorporated in the display panel 25, and allows the user to perform predetermined operations on the display panel 25. The operation unit 26 can also perform operations on a software keyboard displayed on the display panel 25. The operation unit 26 may also be a hardware keyboard, a mouse, or the like.

[0029] Next, we will explain the database used in the dialogue system 100. Figure 4 is a conceptual diagram showing an example of the contents of the user personality DB. User personality DB 131 stores episodes that represent the experiences and inner thoughts of end users. User personality DB 131 stores a combination of metadata and documents as one record. The metadatas column stores information that identifies the end user. The documents column stores sentences that represent episodes as embedding vectors. In Figure 4, the sentences enclosed in corner brackets are examples of sentences before they are converted into vectors. The array enclosed in square brackets above the sentences is an example of an embedding vector obtained by vectorizing the sentences. User personality DB 131 does not store sentences, but rather embedding vectors.

[0030] Figure 5 is a conceptual diagram showing an example of the contents of a knowledge DB. Knowledge DB 132 stores knowledge acquired while the agent is being operated. As initial information, knowledge of the end user and information obtained by what the end user has seen and heard may be stored in knowledge DB 132. Similar to user personality DB 131, knowledge DB 132 stores a combination of metadata and documents as one record. The metadatas column stores information that identifies the end user. Documents stores knowledge. The documents column stores knowledge as embedding vectors rather than sentences. In Figure 5, the entries enclosed in corner brackets are examples of sentences before they are converted into vectors. The array enclosed in square brackets above the sentences shows an example of an embedding vector obtained by vectorizing the sentences. Knowledge DB 132 does not store sentences, but rather stores embedding vectors.

[0031] The user personality DB 131 and the knowledge DB 132 may be configured using a relational database, but it is more preferable to configure them using a vector database suitable for storing embedding vectors. Known vector databases include Chroma, LanceDB, and Vespa.

[0032] Figure 6 is an explanatory diagram showing an example of an agent personality DB. Agent personality DB 133 stores information about the characteristics of an agent desired by the end user. Characteristics are descriptions of the personality that the end user desires for the virtual human realized by the agent, such as age, gender, occupation, the agent's purpose, personality, and expertise. Agent personality DB 133 includes a user column and a settings column. The user column stores identification information of the end user. The settings column stores information about the characteristics. In Figure 6, the information about the characteristics is stored as sentences in natural language.

[0033] FIG. 7 is an explanatory diagram showing an example of an ad-lib DB. When an agent creates an answer and it is determined that the information on which the answer is based is insufficient or that new information is needed because the information is not available, the ad-lib DB 134 stores reference information to be referenced or reference destination information indicating the location of the reference information, such as a URL (Uniform Resource Locator). The ad-lib DB 134 includes a user column and a settings column. The user column stores identification information of the end user. The settings column stores reference destination information. In FIG. 7, in addition to the URL of the reference destination information, instructions to the agent are stored in the settings column. In the following explanation, the reference information and reference destination information will be collectively referred to as ad-lib information.

[0034] The agent personality DB 133 and the ad-lib DB 134 are assumed to be configured using a relational database, but this does not exclude the possibility of configuring them using the vector database described above.

[0035] The episodes stored in the user personality DB 131 and the knowledge stored in the knowledge DB 132 are called memory objects. Information about the agent's characteristics stored in the agent personality DB 133 and the ad-lib information stored in the ad-lib DB 134 are called setting objects. The memory objects and setting objects are collectively called simply objects (elements).

[0036] Next, information processing performed by the dialogue system 100 will be described. FIG. 8 is a flowchart showing an example of the procedure of the initial setting processing. The initial setting processing is processing for setting information to be stored in the user personality DB 131, the knowledge DB 132, the agent personality DB 133, and the ad-lib DB 134. The end user operates the user terminal 2 to start the initial setting processing. The control unit 21 of the user terminal 2 sends a request for a setting screen to the dialogue server 1 (step S1). The control unit 11 of the dialogue server 1 receives the request (step S2). The control unit 11 sends the setting screen to the user terminal 2 (step S3). The control unit 21 of the user terminal 2 receives and displays the setting screen (step S4). The user inputs setting information into the setting screen. It is desirable that the user input all information that can be input into the setting screen, but some information may be input arbitrarily. The control unit 21 accepts the input (step S5). The control unit 21 sends the setting information to the dialogue server 1 (step S6). The control unit 11 of the dialogue server 1 receives the setting information (step S7). Based on the setting information, the control unit 11 generates information to be stored in the above-mentioned user personality DB131, knowledge DB132, agent personality DB133, and ad-lib DB134. The control unit 11 performs vector embedding of the setting information to be stored in the user personality DB131 and knowledge DB132 (step S8). Vector embedding means converting setting information written in natural language into a vector. The control unit 11 stores the vector obtained from the setting information in the user personality DB131 or knowledge DB132. The control unit 11 stores the setting information in the agent personality DB133 or ad-lib DB134 (step S9). The control unit 11 transmits a completion to the user terminal 2 (step S10). The control unit 21 of the user terminal 2 receives the completion (step S11). The control unit 21 displays the completion of the initial setting (step S12) and ends the initial setting process.

[0037] The end user does not need to perform the initial setting process every time he or she uses the dialogue system 100. The end user performs the initial setting process when he or she uses the dialogue system 100 for the first time or when he or she wants to change the behavior of the agent.

[0038] FIG. 9 is a flowchart showing an example of the procedure for dialogue processing. Dialogue processing is processing for controlling a dialogue between an end user and an agent. The end user starts dialogue processing, for example, by selecting "Start dialogue" from a menu displayed on the user terminal 2. The control unit 21 of the user terminal 2 sends a start request to the dialogue server 1 (step S31). The control unit 11 of the dialogue server 1 receives the request (step S32). The control unit 11 sends a dialogue screen to the user terminal 2 (step S33). The control unit 21 of the user terminal 2 receives the dialogue screen (step S34). The end user inputs a question about the matter they want to know into the dialogue screen. Here, the question includes not only a clear question such as "What types of investments do you recommend?" but also an indirect question such as "I'd like to know what types of investments you recommend." If the dialogue ends after repeated dialogues, the end user inputs an end command, such as closing the dialogue screen. The control unit 21 accepts the input (step S35). The control unit 21 determines whether to end the dialogue (step S36). If the control unit 21 determines not to terminate (NO in step S36), it transmits the received question to the dialogue server 1 (step S37). The control unit 11 of the dialogue server 1 receives the question (step S38). The control unit 11 creates a query (step S39). The query is for searching the user personality DB 131, knowledge DB 132, agent personality DB 133, and ad-lib DB 134 for data that matches the end user and the question. The control unit 11 performs the search (step S40). Since the user personality DB 131 and knowledge DB 132 store vector data, the control unit 11 calculates the similarity between the embedding vector obtained by converting the question into a vector and the embedding vector stored in each database, and extracts data with high similarity. For the agent personality DB 133 and ad-lib DB 134, a keyword search using words obtained by morphological analysis of the question is performed to extract matching data. The control unit 11 performs inference processing using information obtained from the four databases (step S41). The control unit 11 transmits a response sentence from the agent obtained by the inference processing to the user terminal 2 (step S42). The control unit 21 of the user terminal 2 receives the response sentence and displays it on the interactive screen (step S43).The control unit 21 returns the process to step S35. If the control unit 21 determines that the process is to be ended (YES in step S36), the process is ended.

[0039] FIG. 10 is a flowchart showing an example of the procedure for the inference process. The inference process corresponds to step S41 in FIG. 9. The control unit 11 of the dialogue server 1 creates a first prompt (step S51). The first prompt includes a question entered by the end user, an episode acquired from the user personality DB 131, knowledge acquired from the knowledge DB 132, information about the agent's characteristics acquired from the agent personality DB 133, and ad-lib information acquired from the ad-lib DB 134. The control unit 11 transmits the first prompt to the dialogue AI 3 (step S52). The control unit 11 receives a response from the dialogue AI 3 (step S53).

[0040] The control unit 11 creates a second prompt (step S54). The second prompt includes the question entered by the end user, the answer obtained from the first prompt, instructions to correct the answer, anecdotes obtained from the user personality DB 131, knowledge obtained from the knowledge DB 132, and information about the agent's characteristics obtained from the agent personality DB 133. The second prompt does not include ad-lib information obtained from the ad-lib DB 134. The control unit 11 sends the second prompt to the interactive AI 3 (step S55). The control unit 11 receives the answer (corrected sentence) from the interactive AI 3 (step S56).

[0041] The control unit 11 creates a third prompt (step S57). The third prompt includes the question entered by the end user, the answer obtained from the second prompt, instructions to correct the answer, an episode obtained from the user personality DB 131, and knowledge obtained from the knowledge DB 132. The third prompt does not include information about the agent's characteristics obtained from the agent personality DB 133, or ad-lib information obtained from the ad-lib DB 134. The control unit 11 sends the third prompt to the interactive AI 3 (step S58). The control unit 11 receives the answer (corrected sentence) from the interactive AI 3 (step S59).

[0042] The control unit 11 creates a fourth prompt (step S60). The fourth prompt includes the question entered by the end user, the answer obtained from the third prompt, instructions to correct the answer, and an episode obtained from the user personality DB 131. The fourth prompt does not include knowledge obtained from the knowledge DB 132, information about the agent's characteristics obtained from the agent personality DB 133, or ad-lib information obtained from the ad-lib DB 134. The control unit 11 sends the fourth prompt to the interactive AI 3 (step S61). The control unit 11 receives the answer (corrected sentence) from the interactive AI 3 (step S62).

[0043] The control unit 11 formats the received response into a format suitable for display on the user terminal 2 (step S63). Note that this formatting may be omitted if it is not necessary. The control unit 11 returns the formatted response or the received response as a return value to the caller of the process.

[0044] Next, the process of creating an answer to an end user's question will be explained using a specific example. Figure 11 is an explanatory diagram showing example objects and example questions. In Figure 11, an example of an end user's episode included in a storage object is set as "When I was three years old, my father went bankrupt due to real estate investment." Also, an example of knowledge is set as "I will recommend something no matter what past experience I have."

[0045] An example of information about the agent's characteristics included in the setting object (hereinafter referred to as agent's characteristic information) is "You are an agent who recommends investments." Also, an example of ad-lib information is "Now is the time to definitely invest in real estate." Furthermore, suppose that an end user inputs a question such as "What type of investment do you recommend?"

[0046] FIG. 12 is an explanatory diagram showing an example of the first prompt and an answer obtained from the first prompt. The first prompt includes a question from the end user, an anecdote, knowledge, characteristic information of the agent, and ad-lib information. What is distinctive about the example answer shown in FIG. 12 is that, based on the ad-lib information, the answer begins with a statement such as "My recommendation is real estate investment." Considering that the end user's anecdote is set to "his father went bankrupt due to real estate investment when he was three years old," this is an undesirable answer.

[0047] Figure 13 is an explanatory diagram showing an example of the second prompt and the answer obtained in the second prompt. The second prompt includes a question from the end user, the answer obtained in the first prompt, anecdotes, knowledge, and characteristic information about the agent. What is notable about the example answer shown in Figure 13 is that the answer shown in Figure 12 has been corrected, and the sentence at the beginning, "My recommendation is real estate investment," has been deleted. This is thought to be the effect of not including the ad-lib information "Now is the time to definitely proceed with real estate investment" in the second prompt.

[0048] Figure 14 is an explanatory diagram showing an example of the third prompt and the answer obtained in response to the third prompt. The third prompt includes a question from the end user, the answer obtained in the second prompt, the episode, and characteristic information of the agent. What is distinctive about the example answer shown in Figure 14 is that the content of the episode has been taken into consideration, and the sentence "Considering my father's experience, I think it would be best to proceed very carefully" has been added. As the objects included in the second prompt now consist of only two, the episode and the characteristic information of the agent, it is thought that the content has been corrected to focus on the content of the episode.

[0049] Figure 15 is an explanatory diagram showing an example of the fourth prompt and the answer obtained in response to the fourth prompt. The fourth prompt includes a question from the end user, the answer obtained in response to the third prompt, and an anecdote. What is distinctive about the example answer shown in Figure 15 is that the content of the anecdote has been further considered, and the phrase "My recommendation," which was present up to Figure 14, has been removed and replaced with the phrase "One way you can challenge yourself...." This is thought to be because the fourth prompt does not include the agent's characteristic information, "You are an agent who recommends investments." In addition, the sentence "It is also important to learn from your father's experience" appears. This suggests that the end user's anecdote has been given greater consideration.

[0050] Fig. 16 is an explanatory diagram showing an example of a dialogue screen. As shown in Fig. 16, the answer sent by the dialogue server 1 to the question posed to the end user is the answer obtained at the fourth prompt. The progress of the process is not shown to the end user.

[0051] As described above, by having the conversational AI 3 sequentially correct the answer obtained from the first prompt using the second to fourth prompts, it is possible to obtain the answer that is most suitable for the end user. This method is used due to the following characteristics of conversational AI that uses a language model (hereinafter referred to as "LLM," an abbreviation for Large Language Model).

[0052] First, LLMs can forget (or stop considering) the settings they specify. To prevent this, it is necessary to set up a mechanism to check whether the settings have been forgotten. Furthermore, LLMs have internal biases, which can lead to behavior that is not intended by the user. Examples of biases include position bias, self-emphasis bias, and redundancy bias. Position bias is a bias in which priority changes depending on the input position. Self-emphasis bias is a bias in which the LLM's own output is prioritized. Redundancy bias is a bias in which a longer output is selected even if the content is the same. To prevent LLMs from behaving unintendedly due to biases, it is necessary to repeatedly give priority elements to the LLM and emphasize them.

[0053] For the above reasons, in this embodiment, the first prompt for creating an answer to the end user's question includes four objects. Then, when the answer obtained from the interactive AI 3 is corrected by the interactive AI 3 through the second to fourth prompts, the objects included in the prompts are reduced by one. The priority of the objects is, in descending order, episode, agent characteristic information, knowledge, and ad-lib information. Therefore, by reducing the objects from the lowest priority and leaving only the episode in the final fourth prompt, the final answer can be expressed in a way that is expected to resonate most with the end user. The set of objects included in each prompt is an example of a characteristic definition.

[0054] In this embodiment, four types of objects are used, so four prompts are created to create an answer. The number of prompts created may be changed depending on the type of object.

[0055] The first prompt described above is an example of a first prompt. The second prompt is an example of a second prompt. In the relationship between the second prompt and the third prompt, the second prompt is an example of a first prompt, and the third prompt is an example of a second prompt. In the relationship between the third prompt and the fourth prompt, the third prompt is an example of a first prompt, and the fourth prompt is an example of a second prompt.

[0056] FIG. 17 is an explanatory diagram showing an example of an initial setting screen. The initial setting screen d02 includes an episode area d021, an agent information area d022, a reference information area d023, an external information area d024, a send button d025, and a cancel button d026. The episode area d021 is an area where the end user inputs a past episode. The input content is stored in the user personality DB 131. The agent information area d022 is an area where the end user inputs a profile of the agent they desire. The input content is stored in the agent personality DB 133. The reference information area d023 is an area where the end user inputs information they know. The input content is stored in the knowledge DB 132. The external information area d024 is an area where the end user inputs external information sources that they consider useful. The input content is stored in the ad-lib DB 134. When the send button d025 is selected, the input content is sent to the dialogue server 1. If the cancel button d026 is selected, the input information is not sent and the initial setting screen d02 closes.

[0057] As described above, in this embodiment, it is possible to create prompts that control the behavior of the conversational AI 3 so that it responds appropriately to questions from the end user. By using prompts to repeatedly correct the initial answer obtained from the conversational AI 3, it is possible to create an answer that expresses something that the end user can empathize with.

[0058] The technical features (constituent elements) described in each embodiment can be combined with each other, and by combining them, new technical features can be formed. The embodiments disclosed herein are to be considered as illustrative in all respects and not restrictive. The scope of the present invention is defined by the claims, not by the above meaning, and is intended to include all modifications within the meaning and scope of the claims. In addition, the claims are written in a format in which a claim cites two or more other claims (multiple claim format), but this is not limited to this. A multiple claim format in which at least one of multiple claims is cited (multi-multi claim format) may also be used. [Explanation of symbols]

[0059] 100: Dialogue Systems 1: Interactive server 11: Control section 12: Main memory 13: Auxiliary storage section 131: User personality DB 132: Knowledge DB 133: Agent Personality DB 134: Adlib DB 14: Communications Department 15: Reading unit 1P: Control program 1a: Portable storage medium 1b: Semiconductor memory 2: User terminal 21: Control unit 22: Main memory 23: Auxiliary storage section 24: Communications Department 25: Display panel 26:Operation section 2P: Control program 3: Conversational AI 31: Control unit M: Language model B: Bus N: Network

Claims

1. obtaining a plurality of prompt base elements that include instructions for the behavior of the language model; Creating a plurality of characteristic definitions that combine a plurality of different elements; generating a plurality of prompts based on each of the generated feature definitions for application to the language model; Output the multiple prompts you created An information processing method that performs processing.

2. providing a first prompt to the language model to obtain a sentence, and then providing a second prompt to the language model to correct the sentence; Output the corrected sentence output by the language model The information processing method according to claim 1 .

3. Generate a second prompt by deleting the predetermined element from the first prompt. The information processing method according to claim 2 .

4. setting the second prompt to a new first prompt; generating a new second prompt based on the new first prompt; providing the corrected sentence and the new second prompt to the language model to obtain a new corrected sentence; repeatedly until the number of elements in the new second prompt is one. The information processing method according to claim 3 .

5. Accepting questions, creating the first prompt associated with the received question sentence; inputting the question sentence and the first prompt into the language model; A sentence corresponding to the answer sentence is obtained from the language model; outputting the corrected sentence obtained by providing the second prompt to the language model; The information processing method according to any one of claims 2 to 4.

6. The elements are classified into memory objects containing information or knowledge about the user's personality, and setting objects containing information or ad-lib about the personality of the agent simulated by the language model. The information processing method according to any one of claims 1 to 4.

7. obtaining a plurality of prompt base elements that include instructions for the behavior of the language model; Creating a plurality of characteristic definitions that combine a plurality of different elements; generating a plurality of prompts based on each of the generated feature definitions for application to the language model; Output the multiple prompts you created An information processing program that causes a computer to execute a process.

8. An information processing device including a control unit, The control unit obtaining a plurality of prompt base elements that include instructions for the behavior of the language model; Creating a plurality of characteristic definitions that combine a plurality of different elements; generating a plurality of prompts based on each of the generated feature definitions for application to the language model; Output the multiple prompts you created An information processing device that executes processing.

Images