User interaction device, user interaction method, and user interaction program

Abstract

A first acquisition unit acquires user information of a target user who is the target of an interaction from a first database in which user information including at least information on a genre of interest and information on knowledge is registered for each user. A second acquisition unit acquires a list of episode data associated with the genre of interest of the target user from a second database in which episode data associated with at least one of a plurality of genres is registered. A generation unit uses the information on the knowledge of the target user acquired from the first database and the list of episode data acquired from the second database to generate a query instructing a language model to select episode data assumed to be of interest to the target user and about which the target user is assumed to have no knowledge.

Description

User interaction device, user interaction method, and user interaction program

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

[0002] With the recent rise of large-scale language models (hereinafter also known as LLMs), technologies have been provided to realize dialogue systems that can engage in natural conversations with users based on past conversations (see Non-Patent Document 1, etc.).

[0003] Wanjun Zhong, Lianghong Guo, Qiqi Gao, He Ye, YanlinWang, "MemoryBank: Enhancing Large Language Models with Long-Term Memory", [online] [Retrieved October 7, 2024], Internet <URL: https: / / ojs.aaai.org / index.php / AAAI / article / download / 29946 / 31654> Qiang Zhang, Jason Naradowsky, Yusuke Miyao, "Mind the Gap Between Conversations for Improved Long-Term Dialogue Generation", [online] [Retrieved October 7, 2024], Internet <URL: https: / / arxiv.org / pdf / 2310.15415.pdf> "Turn your enterprise data into production-ready LLM applications, [online] LlamaIndex [searched October 7, 2024], Internet <URL: https: / / www.llamaindex.ai / > "MemGPT" [online] [searched October 7, 2024], Internet <URL: https: / / memgpt.ai / > "Cotomo" [online] [searched October 7, 2024], Internet <URL: https: / / cotomo.ai / >

[0004] However, there is room for improvement in the prior art. For example, in the prior art, since it interacts with the user passively while referring to past utterances, it asks the user for topics, which places a heavy burden on the user and makes it difficult to maintain the user's willingness to interact with the system. Thus, there is room for improvement in the prior art from the perspective of promoting the use of the user's dialogue system. Therefore, in order to realize a dialogue or the like that can promote the use of the user's dialogue system, it is desirable to enable the selection of information useful to the user.

[0005] The present invention has been made in view of the above, and an object thereof is to enable the selection of information useful to the user.

[0006] In order to solve the above-described problems and achieve the object, a user dialogue apparatus according to the present invention includes: a first acquisition unit that acquires the user information of a target user who is a target of dialogue from a first database in which user information including at least information regarding an interest genre and information regarding knowledge is registered for each user; a second acquisition unit that acquires a list of episode data associated with the interest genre of the target user from a second database in which episode data associated with at least one of a plurality of genres is registered; and a generation unit that generates a query for instructing a language model to select episode data that the target user is assumed to be interested in and have no knowledge of, using the information regarding the knowledge of the target user acquired from the first database and the list of the episode data acquired from the second database.

[0007] According to the present invention, it becomes possible to enable the selection of information useful to the user.

[0008] Figure 1 is a diagram showing an example of the overview of a user interaction system. Figure 2 is a diagram showing an example of the display on a terminal device. Figure 3 is a diagram showing an example of the configuration of a user interaction system according to an embodiment. Figure 4 is a diagram showing an example of the configuration of a user interaction device according to an embodiment. Figure 5 is a diagram showing an example of user information. Figure 6 is a diagram showing an example of episode data. Figure 7 is a flowchart showing an example of a processing procedure executed by the user interaction system. Figure 8 is a diagram showing an example of processing related to episode selection. Figure 9 is a diagram showing an example of processing related to episode selection. Figure 10 is a diagram showing an example of a prompt. Figure 11 is a diagram showing an example of a prompt. Figure 12 is a flowchart showing an example of a processing procedure executed by the user interaction system. Figure 13 is a flowchart showing an example of a processing procedure executed by the user interaction system. Figure 14 is a diagram showing an example of a prompt. Figure 15 is a diagram showing an example of a prompt. Figure 16 is a diagram showing an example of a prompt. Figure 17 is a diagram showing an example of a prompt. Figure 18 is a diagram showing an example of processing related to episode updating. Figure 19 is a diagram showing an example of processing related to episode updating. Figure 20 is a flowchart showing an example of a processing procedure executed by a user interaction device. Figure 21 is a diagram showing an example of a computer that executes an information processing program.

[0009] Hereinafter, one embodiment of the present invention will be described in detail with reference to the drawings. However, the present invention is not limited to this embodiment. Furthermore, in the drawings, the same parts are denoted by the same reference numerals.

[0010] [Embodiment] First, an overview of the information processing performed by the user dialogue system 1 (see Figure 3), which includes the user dialogue device 100 according to the embodiment, will be described with reference to Figure 1. Figure 1 is a diagram showing an example of the overview of the user dialogue system.

[0011] Figure 1 shows an example where the terminal device 10 is a notebook PC (Personal Computer), but the terminal device 10 may be any device that can be used by the user to interact with (also called "conversation") the AI ​​(Artificial Intelligence). The term "interaction" here is not limited to voice communication, but includes various forms of communication between the user (also called the "target user") and the AI. For example, the term "interaction" here includes language-based communication such as text input by the user and text output (display, voice output, etc.) by the device. Furthermore, the term "utterance" below is not limited to voice, but may also include text input, text display, etc., and includes various forms of communication by the user, device (AI), etc.

[0012] The user dialogue system 1 provides a dialogue service in which the user interacts with the AI. The user dialogue system 1 includes a user information acquisition unit 131, a topic acquisition unit 132, a memory setting unit 133, a speech generation unit 134, a dialogue history management unit 135, a user information update unit 136, an information retrieval unit 137, an episode creation unit 138, an episode memory creation unit 139, an information storage unit 140, a conversation end detection unit 141, and a long-term memory DB 200. The long-term memory DB 200 stores user information and episode data, which will be described later. In Figure 1, the long-term memory DB 200 functions as the first database and the second database, which will be described later.

[0013] In addition to user information and episode data, the long-term memory DB 200 stores various other information used to provide dialogue services. For example, the long-term memory DB 200 stores various information such as language models (also simply called "models") used in dialogue services. For example, the long-term memory DB 200 stores various models, such as Model M1, which will be described later, as language models used in dialogue services. Details of each component other than the long-term memory DB 200, such as the user information acquisition unit 131, will be described later.

[0014] Furthermore, the user dialogue system 1 includes a client application 11 that is placed on the terminal device 10 used by the user. For example, the client application 11 is an application installed on the terminal device 10 in order to provide dialogue services by the user dialogue system 1. The terminal device 10 receives input (speech) from the user via the client application 11 and outputs (speech) from the AI ​​to the user by displaying it or outputting it as audio.

[0015] The dashed lines in Figure 1 illustrate an example of the distribution of components (functions) between the terminal device 10 and the user interaction device 100 in the user interaction system 1. In Figure 1, the components to the left of the dashed lines correspond to the components on the terminal device 10 side, and the components to the right of the dashed lines correspond to the components on the user interaction device 100 side. In Figure 1, each component shown to the left of the dashed lines is present in the terminal device 10. Also, in Figure 1, each component shown to the right of the dashed lines is present in the user interaction device 100.

[0016] Note that the boundaries (interfaces) of the device configuration in the user interaction system 1 shown by dashed lines in Figure 1 are not limited to the dashed lines, and the components (functions) assigned to the terminal device 10 and the user interaction device 100 can be any combination as long as the desired processing can be executed. For example, the long-term storage DB 200 may be provided in a device separate from the user interaction device 100. The user interaction system 1 may include a database device having the long-term storage DB 200. In this case, the user interaction device 100 is connected to a database device on the cloud, for example, and communicates with the database device to receive (acquire) information stored in the long-term storage DB 200 from the database device, or to register information in the long-term storage DB 200 by sending information to the database device.

[0017] The above-described configuration of User Dialogue System 1 will be used as an example to explain the overview of processing performed by User Dialogue System 1. Below, as an example of processing performed by User Dialogue System 1, the processing performed by User Dialogue System 1 when the target user, Mr. C (also referred to as "User C"), begins a dialogue with the AI ​​will be explained as an example. Note that the processing described below with User Dialogue System 1 as the processing entity may be performed by any device capable of executing that processing, depending on the device configuration included in User Dialogue System 1.

[0018] First, the user dialogue system 1 acquires information to provide dialogue services to user C at the time it begins a dialogue with user C. For example, the user dialogue system 1 acquires user information of user C from the long-term memory DB 200, including information about genres of interest (also called "interest genres") and information about knowledge that user C is assumed to already possess. Also, for example, the user dialogue system 1 acquires a list of episode data associated with user C's interest genres from the long-term memory DB 200.

[0019] The user dialogue system 1 then uses the acquired information to generate utterances to be output to user C. For example, the user dialogue system 1 uses information about user C's knowledge and a list of episode data to generate a prompt, which is an example of a query instructing the language model, model M1, to select episode data that user C is interested in and that user C is not expected to know. For example, episode data containing topics output in past dialogues with user C are excluded from episode data that user C is not expected to know, as they are considered episode data that user C has knowledge of.

[0020] Model M1 is a Large Language Model (LLM) that outputs information corresponding to a prompt in response to the input of that prompt. For example, in response to a prompt instructing user C to select episode data that user C is interested in and that user C is not expected to know, model M1 outputs episode data that user C is interested in and that user C is not expected to know, but more details will be described later. Note that a prompt is merely one example of information (queries) to be input to a model such as model M1, and queries to be input to a model are not limited to prompts; any form of information (queries) can be used.

[0021] In Figure 1, the user dialogue system 1 generates utterances for user C using episode data output by model M1, which is assumed to be of interest to user C and of which user C is not aware of the information. For example, the user dialogue system 1 generates utterances that include information contained in the episode data output by model M1, which is assumed to be of interest to user C and of which user C is not aware of the information, as information that is of interest and new to user C.

[0022] The user dialogue system 1 outputs utterances to user C that are of interest to user C and contain new information (topics). For example, the user dialogue system 1 outputs utterance information INF1 as shown in Figure 2. Figure 2 is a diagram showing an example of the display on the terminal device. The example shown in Figure 2 is an example of user C having a conversation about games in the previous session, such as when using the previous dialogue service.

[0023] The terminal device 10 used by user C outputs speech information INF1 that includes a topic about the DLC (Downloadable Content) of game X, which is of interest to user C and is new information. In Figure 2, the terminal device 10 displays content CT1, which includes the speech information INF1. As a result, the user dialogue system 1 can proactively output speech to user C that includes information (topics) that are of interest to user C and are new information.

[0024] The content CT1 may be generated by the user dialogue device 100 and provided to the terminal device 10, or it may be generated by the terminal device 10 after acquiring the speech information INF1. Also, although Figure 2 shows an example where the speech information INF1 is displayed, the terminal device 10 may also output the speech information INF1 as audio.

[0025] For example, conventional chatbot systems are designed to always prompt the user to initiate a conversation, which places a significant burden on the user to start a conversation and makes it difficult to maintain their motivation to interact with the system.

[0026] On the other hand, the user dialogue system 1, through the processing described above, can provide an autonomous chatbot that can proactively continue to offer topics that are of interest to the user and contain new information, based on past conversations with the user. The user dialogue system 1 refers to information about the user's areas of interest and knowledge, and uses LLM to select episode data that is of interest to the user and that the user is unaware of from a list of episode data related to the areas of interest. In this way, the user dialogue system 1 can proactively continue to offer topics that are of interest to the user and contain information that the user is unaware of.

[0027] Thus, in User Dialogue System 1, the burden on the user at the start of a conversation is reduced because User Dialogue System 1 presents the topic, making it possible to maintain and improve the user's willingness to converse more effectively than in conventional systems that provide conventional chatbots. For example, User Dialogue System 1 can contribute to increasing the user's willingness to converse with the chatbot (AI) again after the conversation has ended.

[0028] For example, the user dialogue system 1 performs the following processes, although details will be described later. For example, the user dialogue system 1 estimates the user's interests and knowledge through conversation with the user, stores and manages this information as user information. For example, the user dialogue system 1 retrieves user information and topics that are of interest to the user and contain new information from the long-term memory DB 200.

[0029] For example, User Dialogue System 1 autonomously works to acquire topics to provide to the user, structures and stores the results, and manages them by genre. The AI's autonomous activities (activity content) include various activities such as information retrieval, conversations with other users, conversations with other chatbots, and the creation of activities that chatbots cannot actually perform.

[0030] For example, User Dialogue System 1 selects and provides topics containing new information of interest to the user based on the information it manages. For example, User Dialogue System 1 manages episodes by genre so that they can be linked to the user's interests. For example, User Dialogue System 1 creates prompts to proactively provide topics based on the information it has acquired. For example, User Dialogue System 1 acquires new information of interest to the user from external sources or creates it itself.

[0031] This allows the user dialogue system 1 to proactively output topics that contain new information of interest to the user. Furthermore, the user dialogue system 1 can dynamically update topics in response to the conversation with the user and as time passes.

[0032] Furthermore, the queries (prompts) generated by the user dialogue system 1 may be used for various purposes other than the processing (internal processing) shown in Figure 1. For example, the queries (prompts) generated by the user dialogue system 1 may be used as metadata.

[0033] The user interaction system 1 may retain the prompts used to select (generate) the episode data so that they can be checked later. In this case, the user interaction system 1 may register the prompts used to select (generate) the episode data in the long-term memory DB 200, associating them with the selected (generated) episode data, etc.

[0034] For example, the user interaction system 1 may output the generated query in any manner, such as by displaying the query, sending it to another server on the network, or outputting it as a file. For example, the terminal device 10 or the user interaction device 100 may display the query used for processing. The user interaction device 100 may also send the query used for processing to another server (such as a database device) on the network. The user interaction device 100 may also output the query used for processing as a file.

[0035] [Configuration of the User Dialogue System] From here, an example of the user dialogue system 1 that performs the information processing described above will be explained using Figure 3. Figure 3 is a diagram showing an example configuration of the user dialogue system according to the embodiment.

[0036] As shown in Figure 3, the user interaction system 1 includes a terminal device 10 and a user interaction device 100. The user interaction device 100 is connected to the terminal device 10 via a predetermined network N, enabling communication by wire or wireless means. Figure 3 is a diagram showing an example configuration of a user interaction system according to an embodiment.

[0037] Furthermore, the user interaction system 1 shown in Figure 3 may include multiple terminal devices 10 and multiple user interaction devices 100. Also, the system configuration shown in Figure 3 is merely an example, and the user interaction system 1 can employ any device configuration as long as it can perform the desired processing. As mentioned above, the user interaction system 1 may include other devices besides the terminal devices 10 and user interaction devices 100, such as a database device having a long-term storage DB 200.

[0038] Terminal device 10 is an information processing device (computer) used by the user. Terminal device 10 is a client (device) that sends and receives information with the user interaction device 100, which acts as a server (device). Terminal device 10 may be a smartphone, tablet, notebook PC, desktop PC, mobile phone, PDA (Personal Digital Assistant), or other device. Furthermore, terminal device 10 can be any device as long as it can perform the desired processing, and may be a wearable device such as a smartwatch.

[0039] The terminal device 10 transmits information entered by the user to the user dialogue device 100. For example, the terminal device 10 transmits the utterance entered by the user to the user dialogue device 100. The terminal device 10 has a voice sensor such as a microphone and transmits speech information based on the voice uttered by the user detected by the voice sensor to the user dialogue device 100. For example, the terminal device 10 transmits speech information to the user dialogue device 100, which is obtained by converting the user's voice detected by the voice sensor into text (character information).

[0040] Terminal device 10 receives information transmitted from user dialogue device 100. Terminal device 10 receives speech information from user dialogue device 100 indicating speech output by the AI. Terminal device 10 outputs the information received from user dialogue device 100. Terminal device 10 displays the speech information received from user dialogue device 100.

[0041] The user dialogue device 100 is an information processing device (computer) used to provide dialogue services to the user. The user dialogue device 100 receives information from other information processing devices used to provide dialogue services to the user. The user dialogue device 100 transmits AI (chatbot) speech information to be output to the user to the terminal device 10.

[0042] [Configuration of User Interaction Device] Next, the configuration of a user interaction device 100, which is an example of a user interaction device that performs information processing according to the embodiment, will be described. Figure 4 is a diagram showing an example of the configuration of the user interaction device 100 according to the embodiment.

[0043] As illustrated in FIG. 4, the user interaction device 100 of the present embodiment is implemented by a general-purpose computer such as a personal computer, and includes a communication unit 110, a storage unit 120, and a control unit 130. Note that the user interaction device 100 may include an input unit (for example, a keyboard, a mouse, etc.) that receives various operations from an administrator or the like of the user interaction device 100, a display unit (for example, a liquid crystal display, etc.) for displaying information, a voice output unit (for example, a speaker, etc.) for outputting information as voice, and the like.

[0044] The communication unit 110 is realized by, for example, a NIC (Network Interface Card) or the like. The communication unit 110 is connected to a predetermined network such as the Internet by wire or wirelessly, and transmits and receives information to and from other information processing devices such as the terminal device 10.

[0045] The storage unit 120 is realized by, for example, a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory, or a storage device such as a hard disk or an optical disk. The storage unit 120 is an example of the long-term storage DB 200 in FIG. 1. As shown in FIG. 4, the storage unit 120 according to the embodiment includes a user information storage unit 121 and an episode storage unit 122.

[0046] The user information storage unit 121 according to the embodiment stores various information about the user. For example, the user information storage unit 121 stores various information about the user who interacts with the AI. The user information storage unit 121 functions as a first database in which user information including at least information about the user's interest genre and information about knowledge is registered for each user. The user information storage unit 121 accumulates and manages user information.

[0047] As shown in FIG. 5, the user information storage unit 121 stores user information UD corresponding to each user. FIG. 5 is a diagram showing an example of user information. As shown in FIG. 5, the user information storage unit 121 stores a plurality of user information UD, such as user information UD1 of Mr. A (also referred to as "user A"), user information UD2 of Mr. B (also referred to as "user B"), user information UD3 of Mr. C (user C), user information UD4 of Mr. D (also referred to as "user D"), user information UD5 of Mr. E (also referred to as "user E"), and user information UD6 of Mr. F (also referred to as "user F").

[0048] For example, the user information storage unit 121 stores user information UD in which information related to each user is associated with the user name of that user. Each user information UD includes various information related to that user, such as "gender": gender, "age": age, "occupation": occupation, "location": place of residence, "interest": having an interest (such as a hobby), "knowledge": knowledge estimated from past conversations, "episode": a list of ID of episodes with the user, and "used_episode": a list of ID of used episodes.

[0049] "Interest" is information indicating the genre in which the user has an interest. "Knowledge" indicates the knowledge that is assumed to be possessed by the user, estimated from past conversations that the user has had in the user dialogue system 1. "Episode" is information indicating an ID (identification information) that identifies the episode corresponding to that user. "Used_episode" is information indicating an ID (identification information) that identifies the episode that the user dialogue system 1 has already used in the conversation with that user.

[0050] For example, user information UD is stored in the format key:<username>, value:json (JSON format). Note that user information UD is not limited to JSON format; it may be stored in the user information storage unit 121 in any format. Furthermore, the user information storage unit 121 may store various types of information depending on the purpose, not limited to the above. The user information may also consist of various types of information, such as information entered by the user themselves, or information obtained from past conversations the user had with the user dialogue system 1. Note that user information may be acquired by any method, not limited to the above. For example, user information may be acquired from user behavior information in other services. For example, user information may be acquired from the content of internet content such as web pages viewed by the user. In this case, the genres of interest of the user, the knowledge the user is assumed to possess, etc., may be estimated based on the information contained in the internet content such as web pages viewed by the user.

[0051] The episode storage unit 122 according to this embodiment stores information about episodes. For example, the episode storage unit 122 stores information about episodes of an AI and a user interacting with that AI. The episode storage unit 122 functions as a second database in which episode data associated with at least one of a plurality of genres is registered. The episode storage unit 122 stores each piece of information in the form of structured episodes and manages these episodes by genre.

[0052] As shown in Figure 6, the episode memory unit 122 stores episode information ED, in which episodes are associated with genres. Figure 6 is a diagram showing an example of episode data. The bidirectional arrows in Figure 6 indicate an example of the association between interest genres and episode data. As shown in Figure 6, the episode memory unit 122 stores information about multiple interest genres (also called "interest genre information"), such as interest genre GE1, which is the genre name "games", interest genre GE2, which is the genre name "travel", and interest genre GE3, which is the genre name "food".

[0053] For example, the episode memory unit 122 stores episode information ED, in which the genre name of each interest genre is associated with identification information (e.g., episode ID) of the episode data corresponding to that interest genre. Figure 6 shows that the episode corresponding to the source (tail) of the speech bubble of the episode data EP (the episode closest to the beginning of episode data EL3 in Figure 6) corresponds to interest genre GE3 and is associated with interest genre GE3. Note that the interest genres are not limited to interest genres GE1 to GE3, but may include four or more interest genres such as interest genre GE4.

[0054] As shown in Figure 6, the episode memory unit 122 includes episode data EL1, which is episode data of the AI's own episodes; episode data EL2, which is episode data of episodes between the AI ​​and person A (user A); episode data EL3, which is episode data of episodes between the AI ​​and person B (user B); and so on. Episode data EL1 is episode data of the AI's own episodes in which it interacts with the user. For example, episode data EL1 is episode data of episodes that the AI ​​itself acquired or created, without interacting with the user.

[0055] Episode data EL2 is the episode data for an episode between the AI ​​and user A. For example, episode data EL2 is the episode data for an episode obtained through a dialogue between the AI ​​and user A. Episode data EL3 is the episode data for an episode between the AI ​​and user B. For example, episode data EL3 is the episode data for an episode obtained through a dialogue between the AI ​​and user B. Note that episode data is not limited to episode data EL1 to EL3 shown in Figure 6, but also includes episode data such as episode data EL4, which is the episode data for an episode between the AI ​​and person C (user C).

[0056] As shown in Figure 6, each episode data (EP) contains various pieces of information about the episode, such as "name": the name of the person being spoken to, "time": the date and time, "detail": the details of the memory, "summary": the summary, "keyword": the keyword, and "category": the genre.

[0057] "name" is information indicating the user (conversation partner) with whom the AI ​​was interacting when the episode corresponding to that episode data was acquired. "time" is information indicating the date and time corresponding to the episode corresponding to that episode data (for example, the date and time the conversation took place). "detail" is information indicating the detailed content of the episode corresponding to that episode data. "summary" is information indicating a summary of the episode corresponding to that episode data. "keyword" is information indicating the keywords of the episode corresponding to that episode data. "category" is information indicating the interest genre to which the episode corresponding to that episode data belongs. If the episode corresponding to that episode data belongs to multiple interest genres, information indicating multiple interest genres may be registered in "category".

[0058] For example, the interest genre information in episode information ED is stored in the format key:<genre name>, value: list of episode IDs. For example, the episode data EP in episode information ED is stored in the format key:<episode ID> (JSON format). Note that the episode data EP is not limited to JSON format and may be stored in the episode storage unit 122 in any format. Note that the episode storage unit 122 may store various types of information depending on the purpose, not limited to the above.

[0059] Furthermore, the above is merely an example, and the storage unit 120 may store various types of information not limited to those mentioned above. The storage unit 120 stores various types of information used for information processing. For example, the storage unit 120 stores a dialogue history showing past conversations with each user, associated with information that identifies each user (such as a user ID). For example, the storage unit 120 stores information about models such as Model M1, which is an LLM.

[0060] Returning to Figure 4, let's continue the explanation. The control unit 130 is implemented, for example, by a processor such as a CPU (Central Processing Unit) executing a program (for example, an information processing program) stored inside the user interaction device 100 using RAM or the like as a working area. Alternatively, the control unit 130 can be implemented by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

[0061] As shown in Figure 4, the control unit 130 includes a user information acquisition unit 131, a topic acquisition unit 132, a memory setting unit 133, a speech generation unit 134, a dialogue history management unit 135, a user information update unit 136, an information retrieval unit 137, an episode creation unit 138, an episode memory creation unit 139, an information storage unit 140, and a conversation end detection unit 141.

[0062] The user information acquisition unit 131 executes an acquisition process to acquire information. The user information acquisition unit 131 acquires information from other information processing devices such as the management server that manages the project, sensor devices, and terminal devices 10. The user information acquisition unit 131 receives various types of information from other information processing devices and external information processing devices such as terminal devices 10. For example, the user information acquisition unit 131 receives information collected by other information processing devices. For example, the user information acquisition unit 131 acquires information by receiving information from other information processing devices via the communication unit 110.

[0063] For example, the user information acquisition unit 131 acquires information from the terminal device 10, etc. For example, the user information acquisition unit 131 acquires information by receiving information from other information processing devices such as the terminal device 10 via the communication unit 110. For example, the user information acquisition unit 131 receives information from the terminal device 10 regarding users who use the terminal device 10. The user information acquisition unit 131 receives information entered by users who use the terminal device 10 from the terminal device 10.

[0064] The user information acquisition unit 131 acquires various types of information from the storage unit 120. For example, the user information acquisition unit 131 acquires information from the user information storage unit 121, etc. The user information acquisition unit 131 functions as a first acquisition unit that acquires user information of the target user who is the subject of the interaction from the user information storage unit 121.

[0065] The user information acquisition unit 131 acquires user information from the user information storage unit 121 based on the user ID transmitted from the terminal device 10 (client application 11 in Figure 1, etc.) and passes the user information to the topic acquisition unit 132. The user information acquisition unit 131 acquires the past conversation history with the user and passes it to the topic acquisition unit 132.

[0066] The topic acquisition unit 132 performs an acquisition process to acquire information. The topic acquisition unit 132 acquires various types of information from the memory unit 120. For example, the topic acquisition unit 132 acquires information from the episode memory unit 122, etc. The topic acquisition unit 132 functions as a second acquisition unit that acquires a list of episode data associated with the target user's interest genre from the episode memory unit 122.

[0067] The topic acquisition unit 132 functions as a generation unit that generates queries instructing the language model to select episode data that the target user is expected to be interested in and to have no prior knowledge of, using information about the target user's knowledge acquired from the user information storage unit 121 and a list of episode data acquired from the episode storage unit 122.

[0068] The topic acquisition unit 132 generates a query using the information updated by the user information update unit 136. The topic acquisition unit 132 retrieves a list of episode data associated with the target user's interest genre from the episode storage unit 122, which contains episode data including new episode data generated based on external information retrieved by the information retrieval unit 137.

[0069] The topic acquisition unit 132 acquires a list of episode data associated with the target user's interest genre from the episode memory unit 122, which has registered episode data including new episode data corresponding to the new episode (also called "new episode") generated by the episode creation unit 138. The topic acquisition unit 132 acquires a list of episode data associated with the target user's interest genre from the episode memory unit 122, which has registered episode data including new episode data generated by the episode memory creation unit 139.

[0070] The topic acquisition unit 132 acquires information from the episode memory unit 122, etc., that is likely to be of interest to the user but is also likely to be unfamiliar with the user, based on the user information received from the user information acquisition unit 131, generates a prompt for providing a topic, and passes it to the speech generation unit 134.

[0071] The topic acquisition unit 132 selects the interest genre indicated by the "interest" in the target user's user information as the interest genre that the target user is interested in (also called the "target genre"), and acquires episode data of episodes corresponding to the selected target genre from the episode storage unit 122.

[0072] The topic acquisition unit 132 excludes episodes that have already been used in a conversation with the target user from among the acquired episodes. The topic acquisition unit 132 excludes episodes that have already been used from among the acquired episodes, specifically those indicated by "used_episode" in the target user's user information.

[0073] The topic acquisition unit 132 lists summaries of the remaining episodes (also called "remaining episodes") after excluding used episodes, and selects an episode from the remaining episodes that is likely to contain new information for the target user. For example, the topic acquisition unit 132 inputs a prompt to model M1 to select an episode, and outputs information indicating the selected episode to model M1, thereby selecting an episode that is likely to contain new information for the target user. The topic acquisition unit 132 passes the summary and details of the selected episode to the speech generation unit 134.

[0074] The memory setting unit 133 sets the dialogue history in a format that the speech generation unit 134 can refer to and passes it to the speech generation unit 134. The memory setting unit 133 aggregates the target user's dialogue history into a single text and passes the indexed information to the speech generation unit 134. For example, the memory setting unit 133 aggregates all dialogue history into a single text, generates an index using a technology such as Llama-index disclosed in Non-Patent Document 3, and passes the generated index to the speech generation unit 134.

[0075] The speech generation unit 134 performs speech generation processing to generate information about utterances to be used in dialogue with the user. The speech generation unit 134 uses the episode data selected by the language model to which the query has been input to generate speech information that includes content that the target user is expected to not have knowledge of, in a genre that the target user is interested in.

[0076] For example, the speech generation unit 134 uses a speech generation model, which is an LLM (Limited Language Model) that generates utterances for the user, to generate speech information that includes content that the target user is not expected to know in a genre that the target user is interested in. The speech generation model may be model M1, or it may be a different model from model M1. For example, the speech generation unit 134 inputs a prompt to the speech generation model to generate utterances for the user, and causes the speech generation model to output speech information, thereby generating speech information that includes content that the target user is not expected to know in a genre that the target user is interested in.

[0077] The dialogue history management unit 135 manages the dialogue history. The dialogue history management unit 135 registers the utterances made during the dialogue between the AI ​​and the target user as the target user's dialogue history in the storage unit 120. For example, until it receives information from the conversation end detection unit 141 indicating that the end of the dialogue has been detected (also called "detection information"), the dialogue history management unit 135 registers information indicating each utterance made during the dialogue between the AI ​​and the target user as the target user's dialogue history. When the dialogue history management unit 135 receives detection information, it passes the dialogue history up to that point to the user information update unit 136, the episode creation unit 138, the episode memory creation unit 139, etc.

[0078] The user information update unit 136 executes an update process to update user information. The user information update unit 136 estimates at least one of the target user's areas of interest and knowledge from the content of the conversation with the target user, and updates the target user's user information using the estimated information.

[0079] The user information update unit 136 estimates the target user's interests (hobbies), knowledge, etc., from the content of the conversation with the target user, and updates the user information according to the estimation result. For example, the user information update unit 136 registers the updated target user information in the user information storage unit 121. For example, the user information update unit 136 passes the updated target user information to the information retrieval unit 137.

[0080] For example, the user information update unit 136 uses a user information estimation model, which is an LLM (Limited Literacy Model) that generates estimation results of user information such as the user's interests (genres) and the user's knowledge, to generate user information estimation results that show the estimated user's interests, genres of interest, and knowledge. The user information estimation model may be model M1 or a different model from model M1. For example, the user information update unit 136 inputs prompts including the user's dialogue history and processing conditions into the user information estimation model, and causes the user information estimation model to output user information estimation results, thereby generating user information estimation results that include the user's interests and knowledge in genres of interest to the user.

[0081] The information retrieval unit 137 performs a search process to retrieve information. Based on the information updated by the user information update unit 136, the information retrieval unit 137 searches for external information to be used in generating new episode data. For example, the information retrieval unit 137 creates search keywords based on the user information of the target user received from the user information update unit 136 and the target user's most recent dialogue history, and then searches for external information. The information retrieval unit 137 passes the results of the external information search to the episode memory creation unit 139.

[0082] For example, the information retrieval unit 137 generates keywords for searching for information that the user might be interested in, using a keyword generation model, which is an LLM (Limited Licensing Model) that generates search keywords. The keyword generation model may be model M1, or it may be a different model from model M1. For example, the information retrieval unit 137 inputs prompts containing user information and instructions such as processing conditions into the keyword generation model, and causes the keyword generation model to output keywords, thereby generating keywords for searching for information that the user might be interested in.

[0083] The episode creation unit 138 performs a creation process to generate an episode. The episode creation unit 138 generates a new episode through the creation process. The episode creation unit 138 generates a new episode that is not included in the episode data stored in the episode storage unit 122.

[0084] The episode creation unit 138 performs a creative process to create the AI ​​(chatbot)'s own experience based on the target user's most recent conversation history. The episode creation unit 138 then passes the result of the creative process to the episode memory creation unit 139.

[0085] For example, the episode creation unit 138 uses an episode creation model, which is an LLM (Limited Language Model) that creates new episodes, to create an episode as the AI's own experience. The episode creation model may be model M1, or it may be a different model from model M1. For example, the episode creation unit 138 inputs a prompt to the episode creation model to create a new episode based on the target user's most recent dialogue history, and causes the episode creation model to output information indicating a new episode, thereby creating an episode as the AI's own experience.

[0086] For example, the episode creation unit 138 generates an episode for the AI ​​(chatbot) itself using an episode creation model, which is an LLM (Limited Language Model) that creates episodes for the AI. The episode creation model may be model M1 or a different model from model M1. For example, the episode creation unit 138 inputs prompts including user dialogue history and processing conditions into the episode creation model, and causes the episode creation model to output information indicating the detailed content of the episode, thereby generating an episode as the AI's own experience in a genre of interest to the user.

[0087] The episode memory creation unit 139 executes an episode memory creation process to generate episode data to be registered in the episode memory unit 122. The episode memory creation unit 139 generates episode data for a new episode using information based on the interaction history with the target user, and passes the new episode data to the information storage unit 140.

[0088] The episode memory creation unit 139 generates episode data for a new episode using information based on the dialogue history with the target user, estimates the genre of the generated new episode, and passes the new episode data to the information storage unit 140, associating it with the estimated genre. For example, the episode memory creation unit 139 estimates the genre of the dialogue history received from the dialogue history management unit 135, the search results received from the information retrieval unit 137, and the episode summary received from the episode creation unit 138, structures each of these as individual episodes, and passes them to the information storage unit 140.

[0089] For example, the episode memory creation unit 139, based on the dialogue history received from the dialogue history management unit 135, structures the episodes acquired in the dialogue between the AI ​​and the target user, and passes them to the information storage unit 140 as episode data. For example, if the target user is user A, the episode memory creation unit 139 generates episode data of the episodes acquired in the dialogue between the AI ​​and user A based on the dialogue history with user A, associates that episode data with the corresponding genre as episode data EL2, and passes it to the information storage unit 140.

[0090] For example, the episode memory creation unit 139, based on the search results received from the information retrieval unit 137, structures the episodes acquired through the dialogue between the AI ​​and the target user, and passes them to the information storage unit 140 as episode data. For example, if the target user is user B, the episode memory creation unit 139 generates episode data of the episodes acquired through the dialogue between the AI ​​and user B based on the search results received from the information retrieval unit 137, associates that episode data as episode data EL3 with the corresponding genre, and passes it to the information storage unit 140.

[0091] For example, the episode memory creation unit 139, based on the episodes received from the episode creation unit 138, structures the AI's own episodes and passes them to the information storage unit 140 as episode data. For example, the episode memory creation unit 139 generates episode data of the AI's own experiences based on the information indicating the episodes received from the episode creation unit 138, associates that episode data as episode data EL1 with the corresponding genre, and passes it to the information storage unit 140.

[0092] For example, the episode memory creation unit 139 generates episode data using an episode data generation model, which is an LLM (Low-Level Memory) that generates episode data. The episode data generation model may be model M1, or it may be a different model from model M1. For example, the episode memory creation unit 139 generates episode data by inputting prompts containing instructions such as episode details and processing conditions to the episode data generation model, causing the episode data generation model to output episode data.

[0093] The information storage unit 140 performs information storage processing to store information. For example, the information storage unit 140 functions as a long-term memory DB storage unit that performs information storage processing to store information in the long-term memory DB 200. The information storage unit 140 performs information storage processing to store information in the storage unit 120, which is an example of the long-term memory DB 200.

[0094] The information storage unit 140 stores new episodes (new episodes) received from the episode memory creation unit 139 in the long-term memory DB 200. For example, the information storage unit 140 registers the episode data received from the episode memory creation unit 139 in the episode memory unit 122.

[0095] For example, the information storage unit 140 checks the genre described in the new episode, and if that genre does not exist in the genre information managed in the long-term memory DB 200, it adds that genre as a new genre. For example, if the genre of the new episode is a new genre not registered in the episode memory unit 122, the information storage unit 140 adds the new genre and registers the new episode in the episode memory unit 122, associating it with that new genre.

[0096] Furthermore, if the genre described in the new episode is a genre managed in the long-term memory DB 200 (also called an "existing genre"), the information storage unit 140 associates the new episode with that existing genre. For example, if the genre described in the new episode is an existing genre registered in the episode memory unit 122, the information storage unit 140 associates the new episode with that existing genre and registers it in the episode memory unit 122.

[0097] The conversation end detection unit 141 executes a detection process to detect the end of a conversation with the target user. The conversation end detection unit 141 detects the end of the conversation between the target user and the AI ​​and notifies the conversation history management unit 135 of the detection information indicating that the end has been detected. For example, if the conversation end detection unit 141 obtains information indicating that the user has pressed the conversation end button provided in the client application 11, it determines that the conversation between the target user and the AI ​​has ended. For example, the conversation end detection unit 141 determines that the conversation has ended if a certain amount of time has elapsed since the last utterance from the AI ​​(chatbot). For example, if the conversation end detection unit 141 determines that the conversation has ended, it passes the detection information to the conversation history management unit 135.

[0098] [Processing Example] From here, assuming the configuration of User Dialogue System 1 described above, we will explain an example of processing by User Dialogue System 1 in interaction with a user. Note that explanations of points that are the same as those described above will be omitted as appropriate.

[0099] [Initiating Dialogue with the User] First, an example of the processing procedure of the user dialogue system 1 when initiating dialogue with a user, such as when a user logs in, will be explained using Figure 7. Figure 7 is a flowchart of an example of the processing procedure executed by the user dialogue system. Below, the processing flow will be explained as an example when the target user is User C and the AI ​​side of the user dialogue system 1 speaks to the user.

[0100] The user dialogue system 1 obtains a user ID (step S101). For example, the user dialogue device 100 obtains the user ID of user C, who is the target user with whom it will engage in dialogue, by receiving the user ID of user C entered by user C when logging into the client application 11 from the terminal device 10.

[0101] The user interaction system 1 acquires user information (step S102). For example, the user interaction device 100 acquires user information of user C, who is the target user indicated by the user ID, from the user information storage unit 121.

[0102] The user dialogue system 1 performs topic acquisition (step S103). For example, the user dialogue device 100 selects a genre that user C is likely to be interested in based on the "interest" information of user C, and acquires all episodes in that genre. For example, as shown in Figure 8, the user dialogue device 100 acquires episodes associated with user C's interest genres, such as games and music, from the episode information ED stored in the episode storage unit 122.

[0103] Figure 8 shows an example of the process related to episode selection. In Figure 8, the user interaction device 100 selects the hatched episode data EP as episode data (also called "candidate episode") associated with the user C's interest genre. The user interaction device 100 then retrieves the selected candidate episode from the episode storage unit 122.

[0104] Furthermore, as shown in Figure 9, the user interaction device 100 excludes used episodes from the acquired episodes. Figure 9 is a diagram illustrating an example of the processing related to episode selection. In Figure 9, the user interaction device 100 excludes episodes that have been used in the interaction with user C from among the episode data EP11, EP12, and EP13 selected as candidate episodes. The user interaction device 100 excludes the episode indicated by "used_episode" in user information of user C as a used episode from among the candidate episodes. In Figure 9, the user interaction device 100 excludes episode data EP13 of the candidate episodes, which is an episode that has been used in the interaction with user C.

[0105] The user dialogue device 100 then generates a summary list of summaries of the remaining episodes after excluding the episode data EP13 of the used episodes, and selects from the remaining episodes that seem likely to contain new information for user C. For example, the user dialogue device 100 generates a summary list of the remaining episodes using the "summary" information of each of the remaining episodes. The user dialogue device 100 then performs speech generation processing using the summary and details of the selected episodes using model M1.

[0106] For example, the user interaction device 100 inputs a prompt PT11 as shown in Figure 10 to model M1 and outputs information indicating the selected episode to model M1, thereby selecting an episode that is likely to contain new information for user C. Figure 10 is a diagram showing an example of a prompt.

[0107] In Figure 10, the string {user_knowledge} will be replaced with the “knowledge” information from the user's information. For example, if the target user is User C, the string {user_knowledge} in Figure 10 will contain the “knowledge” information from User C's user information. Also, the string {episode_list} in Figure 10 will be replaced with the summary list. This string will be replaced with this string. For example, if the target user is User C, the string {episode_list} in Figure 10 will contain the summary list of the remaining episodes selected for User C.

[0108] The user dialogue system 1 performs memory settings (step S104). For example, the user dialogue device 100 aggregates the dialogue history of user C into a single text and generates indexed information. For example, the user dialogue device 100 aggregates all dialogue history with user C into a single text and generates an index using a technology such as Llama-index disclosed in Non-Patent Document 3. The user dialogue device 100 performs speech generation processing using the generated index.

[0109] Then, the user dialogue system 1 generates speech using the information acquired in steps S103 and S104 (step S105). The user dialogue device 100 inputs a prompt PT11 as shown in Figure 11 to the speech generation model and causes the speech generation model to output speech information, thereby generating speech information that includes content that user C is assumed to have no knowledge of in a genre that user C is interested in. Figure 11 is a diagram showing an example of a prompt.

[0110] In Figure 11, the string {referred_info} is replaced with the summary list of the remaining episodes generated in step S104. For example, if the target user is user C, the string {referred_info} in Figure 11 will contain the summary list of the remaining episodes for user C.

[0111] Then, the user dialogue system 1 makes an initial utterance to introduce a topic (step S106). For example, terminal device 10 receives utterance information generated by user dialogue device 100 from user dialogue device 100 and outputs based on the received utterance information. For example, terminal device 10 used by user C receives utterance information generated by user dialogue device 100 from user dialogue device 100 and makes an utterance based on the received utterance information that includes content that user C is assumed to have no knowledge of in a genre that user C is interested in.

[0112] For example, the terminal device 10 used by user C displays content CT1 as shown in Figure 2. In this case, the terminal device 10 displays speech information INF1 that includes a topic about DLC for game X, which is of interest to user C and is new information.

[0113] [During user interaction] Next, an example of the processing procedure of the user interaction system 1 during user interaction will be explained using Figure 12. Figure 12 is a flowchart showing an example of the processing procedure executed by the user interaction system.

[0114] The user dialogue system 1 receives user utterances (step S201). For example, the terminal device 10 receives input (utterances) from user C and transmits information indicating the received utterances from user C to the user dialogue device 100.

[0115] Then, the user dialogue system 1 performs speech generation (step S202). The user dialogue device 100 generates speech information that includes a response to the content of the utterance of user C received in step S201. The user dialogue device 100 generates speech information that includes a response to the content of user C's utterance by inputting a prompt to generate a response to user C's utterance into the speech generation model and causing the speech generation model to output speech information. For example, the user dialogue device 100 may generate speech information that includes a response to the content of user C's utterance using dialogue system technology without using an LLM such as a speech generation model. Alternatively, the user dialogue device 100 may set a profile for the AI ​​(chatbot) itself using a system prompt and have the AI ​​role-play as a person with that profile.

[0116] The user dialogue system 1 manages the dialogue history (step S203). The user dialogue device 100 registers the information indicating the utterances in steps S201 and S202 in the storage unit 120 as the dialogue history with user C.

[0117] Furthermore, the user dialogue system 1 generates utterances (step S204). For example, terminal device 10 receives utterance information generated by user dialogue device 100 from user dialogue device 100 and outputs based on the received utterance information. For example, terminal device 10 used by user C receives utterance information generated by user dialogue device 100 from user dialogue device 100 and makes an utterance that includes a response to the content of user C's utterance based on the received utterance information.

[0118] [After interaction with the user] Next, an example of the processing procedure of the user interaction system 1 after interaction with the user will be explained using Figure 12. Figure 13 is a flowchart showing an example of the processing procedure executed by the user interaction system.

[0119] The user dialogue system 1 performs a conversation end detection (step S301). If the user dialogue device 100 detects the end of the conversation, it executes the process in step S302. The user dialogue system 1 may either provide a conversation end button in the client application 11 and detect the end of the conversation when that button is pressed, or it may determine that the conversation has ended after a certain period of time has elapsed since the AI ​​(chatbot)'s last utterance.

[0120] The user dialogue system 1 manages the dialogue history (step S302). The user dialogue device 100 updates the dialogue history based on the utterances of user C and the AI ​​received from the terminal device 10 (client application 11) until it detects the end of the conversation. For example, the user dialogue device 100 generates episode data based on the updated dialogue history of user C. The user dialogue device 100 then executes the episode memory creation process in step S306 using the updated dialogue history of user C.

[0121] The user dialogue system 1 updates user information (step S303). The user dialogue device 100 estimates user C's interests (hobbies) and knowledge from the content of the dialogue with user C, and updates the user information as necessary. The user dialogue device 100 stores the updated user information in the long-term memory DB 200 (user information storage unit 121, etc.).

[0122] The user interaction device 100 inputs a prompt PT21, as shown in Figure 14, into a user information estimation model, and causes the user information estimation model to output user information estimation results, thereby generating user information estimation results that include the target user's areas of interest, the knowledge the target user possesses, etc. Figure 14 is a diagram showing an example of a prompt.

[0123] In Figure 14, the string {dialogue_history} is replaced and replaced with the dialogue history updated in step S302. For example, if the target user is user C, the string {referred_info} in Figure 10 will contain the updated dialogue history of user C.

[0124] The user dialogue device 100 adds the estimated "interests" and "knowledge" to the "interest" and "knowledge" fields of the user information, respectively, and stores them in the long-term memory DB 200 (user information storage unit 121, etc.). For example, the user dialogue device 100 inputs the prompt PT 21 to the user information estimation model and causes the user information estimation model to output the user information estimation result, thereby generating a user information estimation result that includes user C's interest genres, the knowledge possessed by user C, etc. In this case, the user dialogue device 100 adds the "interests" and "knowledge" of user C indicated by the user information estimation result to the "interest" and "knowledge" fields of user C's user information, respectively, and stores them in the long-term memory DB 200 (user information storage unit 121, etc.).

[0125] The user dialogue system 1 performs an information retrieval (step S304). The user dialogue device 100 creates search keywords based on the updated user information of user C and user C's most recent dialogue history, and performs an external information retrieval. The user dialogue device 100 uses the results of the external information retrieval to execute the episode memory creation process in step S306.

[0126] The user interaction device 100 inputs a prompt PT22, as shown in Figure 15, into a keyword generation model and causes the keyword generation model to output keywords, thereby generating keywords for searching for information that the target user may be interested in. Figure 15 is a diagram showing an example of a prompt.

[0127] In Figure 15, the string {user_info} will be replaced with the user information of the target user. For example, if the target user is user C, the string {user_info} in Figure 15 will contain the user information of user C.

[0128] The user interaction device 100 uses the generated keywords as search keywords to search for external information. For example, the user interaction device 100 may search for external information by using the generated keywords and any API (Application Programming Interface), such as the Google Custom Search API, to search for articles on web pages.

[0129] The user dialogue system 1 creates an episode (step S305). The user dialogue device 100 performs a creation process to create the AI ​​(chatbot)'s own experience based on the user C's previous dialogue history. The user dialogue device 100 uses the results of the creation process to perform the episode memory creation process in step S306.

[0130] The user dialogue device 100 inputs a prompt PT23, as shown in Figure 16, to the episode creation model, causing the episode creation model to output information indicating the detailed content of the episode, thereby generating an episode as the AI's own experience. Figure 16 is a diagram showing an example of a prompt.

[0131] In Figure 16, the string {dialogue_history} is replaced with the dialogue history updated in step S302. For example, if the target user is user C, the string {referred_info} in Figure 10 will contain the updated dialogue history of user C. Also, the string {ai_name} in Figure 16 is replaced with the name set for the AI, and the string {ai_hobby} in Figure 16 is replaced with the hobby (interest genre, etc.) set for the AI. In this way, the user dialogue system 1 generates an episode for the AI ​​itself. The user dialogue system 1 may also generate an episode with the AI ​​itself having a profile.

[0132] The user dialogue system 1 performs episodic memory creation (step S306). The user dialogue device 100 executes an episodic memory creation process to generate an episodic memory based on the information acquired in steps S302 to S305. The user dialogue system 1 performs dialogue history, search results from information retrieval processing, episode summaries from episode creation processing, and genre estimation, structures each as individual episodes, and executes the information storage process in step S307.

[0133] The user interaction device 100 generates episode data by inputting a prompt PT24, as shown in Figure 17, into the episode data generation model and causing the episode data generation model to output episode data. Figure 17 is a diagram showing an example of a prompt.

[0134] In Figure 17, the string {episode_detail} is replaced with information indicating the detailed content of the episode, including the information obtained in steps S302 to S305. In Figure 17, the user interaction device 100 generates the episode data EP in Figure 17 by inputting prompt PT24 to the episode data generation model and causing the episode data generation model to output episode data. For example, the user interaction device 100 executes the episode data generation model based on prompt PT24 and generates JSON information based on the output result of the episode data generation model.

[0135] The user interaction system 1 performs information storage (step S307). The user interaction device 100 performs information storage processing to store the episode data generated in step S306 in the episode storage unit 122.

[0136] For example, if the genre of the episode data generated in step S306 is an existing genre, the process shown in Figure 18 is executed. Figure 18 is a diagram showing an example of the process related to episode updating. In Figure 18, the user interaction device 100 registers the new episode NE1 in the episode storage unit 122, associating it with the existing genres of interest genre GE2 and interest genre GE3, because the interest genre of the new episode NE1 is the existing genre of travel and dining.

[0137] Furthermore, if the genre of the episode data generated in step S306 is a genre that is not an existing genre (a new genre), the process shown in Figure 19 is executed. Figure 19 is a diagram showing an example of the process related to episode updating. In Figure 19, the user interaction device 100 adds the interest genre GE11, whose genre name is "model", as a new genre because the interest genre of the new episode NE2 is a new genre called "model", and registers the new episode NE2 in the episode storage unit 122, associating it with the interest genre GE11.

[0138] [Flowchart] Here, an example of a processing procedure performed by the user interaction device will be explained using Figure 20. Figure 20 is a flowchart of an example of a processing procedure performed by the user interaction device. For example, Figure 20 is a flowchart of an example of a query generation process performed by the user interaction device. The user interaction device 100 will perform the processing shown in Figure 20 at any time, such as periodically or at the instruction of an administrator. For example, the user interaction device 100 will perform the processing shown in Figure 20 when it starts interacting with the user.

[0139] The user interaction device 100 obtains user information from a first database in which user information is registered (step S11). For example, the user interaction device 100 obtains user information of the target user who is the subject of the interaction from the user information storage unit 121, which is a first database in which user information including at least information about areas of interest and information about knowledge for each user is registered.

[0140] The user interaction device 100 obtains a list of episode data associated with the user's interest genres from a second database in which episode data is registered (step S12). For example, the user interaction device 100 obtains a list of episode data associated with the target user's interest genres from the episode storage unit 122, which is a second database in which episode data associated with at least one of several genres is registered.

[0141] The user interaction device 100 generates a query instructing the language model to select episode data that the user is expected to be interested in and to have no prior knowledge of, using information about the user's knowledge and a list of episode data (step S13). For example, the user interaction device 100 generates a prompt, which is a query instructing the LLM to select episode data that the target user is expected to be interested in and to have no prior knowledge of, using information about the target user's knowledge obtained from the user information storage unit 121 and a list of episode data obtained from the episode storage unit 122.

[0142] [Modifications] The process described above is just one example, and the user dialogue system 1 may perform reliability evaluations using various information. Several examples of modifications in this regard will be explained below. Details that are the same as those in the embodiments described above will be omitted as appropriate.

[0143] For example, the modified user dialogue system 1 may perform processing that takes into account the semantic similarity between interest genres. For example, the user dialogue system 1 may perform processing using the semantic similarity between genre names (words) of interest genres. An example of this type of processing is described below.

[0144] For example, the user dialogue system 1 may process data using a score (also called "similarity") that indicates the semantic closeness between the genre names (words) of each interest genre. The score (similarity) indicating the semantic closeness between the genre names (words) of the interest genres may be calculated using any method such as Word2vec.

[0145] For example, automobiles and car models are related interest genres concerning cars, and their similarity is above a predetermined threshold (for example, any value such as 0.5). In this way, if there are interest genres (also called "related genres") whose similarity is above a predetermined threshold for a given interest genre, the user dialogue system 1 may use the information of those related genres to perform processing.

[0146] For example, if the target user's area of ​​interest is "automobiles," and there is no new information (episodes) available to the target user regarding the area of ​​interest "automobiles," the user dialogue system 1 may use information about "car models," which is a related area to "automobiles." In this case, if the user dialogue system 1 has no new information (episodes) available to the target user regarding the area of ​​interest "automobiles," and there are unused episodes in the conversation with the target user regarding "car models," which is a related area to "automobiles," the system may speak to the target user about those episodes.

[0147] The above is merely an example, and the user dialogue system 1 may perform various processes using the similarity between interest genres. For example, when adding an episode, the user dialogue system 1 may merge interest genres based on their similarity. For instance, when adding a new episode, even if the interest genre of the new episode is not included in an existing genre, if there is an existing genre whose similarity to the interest genre of the new episode is above a predetermined threshold, the interest genre of the new episode may be merged into that genre.

[0148] Furthermore, for example, when adding a new episode, if the interest genre of that new episode is not included in an existing genre, and there is no existing genre whose similarity to the interest genre of that new episode exceeds a predetermined threshold, the new episode may be added to that genre as a new interest genre.

[0149] [Effects] As described above, in the user dialogue device 100 of this embodiment, a user information acquisition unit 131, which is an example of a first acquisition unit, acquires user information of the target user who is the subject of the dialogue from a first database in which user information including at least information on interest genres and information on knowledge for each user is registered. A topic acquisition unit 132, which is an example of a second acquisition unit, acquires a list of episode data associated with the target user's interest genre from a second database in which episode data associated with at least one of a plurality of genres is registered. A topic acquisition unit 132, which is an example of a generation unit, uses the information on the target user's knowledge obtained from the first database and the list of episode data obtained from the second database to generate a query that instructs the language model to select episode data that the target user is interested in and is assumed to have no knowledge of.

[0150] In this way, the user dialogue device 100 can instruct the language model to select episode data that is expected to be of interest to the target user but of which the user has no prior knowledge, by generating queries that instruct the language model to select episode data that is expected to be of interest to the target user but of which the user has no prior knowledge. Therefore, the user dialogue device 100 can enable the user to select information that is useful to the user.

[0151] [Program] A program can also be created that describes the processing performed by the user interaction device 100 according to the above embodiment in a language that can be executed by a computer. In one embodiment, the user interaction device 100 can be implemented by installing an information processing program that performs the above information processing as packaged software or online software on a desired computer. For example, by having an information processing device execute the above information processing program, the information processing device can be made to function as the user interaction device 100. In addition, the information processing device can include mobile communication terminals such as smartphones, mobile phones and PHS (Personal Handyphone System), and slate terminals such as PDA (Personal Digital Assistant). Furthermore, the functions of the user interaction device 100 may be implemented on a cloud server.

[0152] Figure 21 shows an example of a computer that executes an information processing program. Computer 1000 includes, for example, memory 1010, CPU 1020, hard disk drive interface 1030, disk drive interface 1040, serial port interface 1050, video adapter 1060, and network interface 1070. These components are connected by a bus 1080.

[0153] Memory 1010 includes ROM (Read Only Memory) 1011 and RAM 1012. ROM 1011 stores, for example, a boot program such as BIOS (Basic Input Output System). The hard disk drive interface 1030 is connected to the hard disk drive 1031. The disk drive interface 1040 is connected to the disk drive 1041. A removable storage medium, such as a magnetic disk or optical disk, is inserted into the disk drive 1041. A serial port interface 1050 is connected to, for example, a mouse 1051 and a keyboard 1052. A video adapter 1060 is connected to, for example, a display 1061.

[0154] Here, the hard disk drive 1031 stores, for example, the OS 1091, the application program 1092, the program module 1093, and the program data 1094. The information described in the above embodiment is stored, for example, in the hard disk drive 1031 or the memory 1010.

[0155] Furthermore, the information processing program is stored in the hard disk drive 1031 as a program module 1093 containing instructions to be executed by the computer 1000, for example. Specifically, the program module 1093 containing instructions for each process executed by the user interaction device 100 described in the above embodiment is stored in the hard disk drive 1031.

[0156] Furthermore, the data used for information processing by the information processing program is stored as program data 1094, for example, in the hard disk drive 1031. The CPU 1020 then reads the program module 1093 and program data 1094 stored in the hard disk drive 1031 into the RAM 1012 as needed and executes the procedures described above.

[0157] Furthermore, the program module 1093 and program data 1094 related to the information processing program are not limited to being stored in the hard disk drive 1031; for example, they may be stored in a removable storage medium and read by the CPU 1020 via a disk drive 1041 or the like. Alternatively, the program module 1093 and program data 1094 related to the information processing program may be stored in another computer connected via a network such as a LAN (Local Area Network) or WAN (Wide Area Network) and read by the CPU 1020 via a network interface 1070.

[0158] Although embodiments applying the invention made by the present inventors have been described above, the present invention is not limited by the descriptions and drawings that constitute part of the disclosure of the present invention in this embodiment. That is, all other embodiments, examples, and operational techniques made by those skilled in the art based on this embodiment are included in the scope of the present invention.

[0159] 1 User Dialogue System 10 Terminal Device 100 User Dialogue Device 110 Communication Unit 120 Storage Unit 121 User Information Storage Unit (First Database) 122 Episode Storage Unit (Second Database) 130 Control Unit 131 User Information Acquisition Unit (First Acquisition Unit) 132 Topic Acquisition Unit (Second Acquisition Unit, Generation Unit) 133 Storage Setting Unit 134 Speech Generation Unit 135 Dialogue History Management Unit 136 User Information Update Unit 137 Information Retrieval Unit 138 Episode Creation Unit 139 Episode Memory Creation Unit 140 Information Storage Unit 141 Conversation End Detection Unit

Claims

1. A user dialogue device comprising: a first acquisition unit that acquires user information of a target user to be the subject of dialogue from a first database in which user information including at least information on areas of interest and information on knowledge for each user is registered; a second acquisition unit that acquires a list of episode data associated with the area of ​​interest of the target user from a second database in which episode data associated with at least one of a plurality of areas is registered; and a generation unit that generates a query instructing a language model to select episode data that the target user is expected to be interested in and not have knowledge of, using the information on the target user's knowledge acquired from the first database and the list of episode data acquired from the second database.

2. The user dialogue device according to claim 1, comprising: a user information update unit that estimates at least one of the target user's interests and knowledge from the content of the dialogue with the target user, and updates the target user's user information using the estimated information, wherein the generation unit generates the query using the information updated by the user information update unit.

3. The user interaction device according to claim 2, further comprising: an information retrieval unit that searches for external information used to generate new episode data based on the information updated by the user information update unit, wherein the second acquisition unit acquires a list of episode data associated with the target user's interest genre from the second database in which the episode data, including the new episode data generated based on the external information retrieved by the information retrieval unit, is registered.

4. The user interaction device according to claim 2, comprising: an episode creation unit for generating new episodes, wherein the second acquisition unit acquires a list of episode data associated with the target user's interest genre from the second database in which the episode data, including new episode data corresponding to the new episode generated by the episode creation unit, is registered.

5. The user dialogue device according to claim 1, comprising: an episode memory creation unit that generates new episode data using information based on the dialogue history with the target user and registers the new episode data in the second database, wherein the second acquisition unit acquires a list of episode data associated with the target user's interest genre from the second database in which the episode data, including the new episode data generated by the episode memory creation unit, is registered.

6. The user dialogue device according to claim 1, further comprising: a speech generation unit that generates speech information including content that the target user is assumed to have no knowledge of in a genre of interest to the target user, using the episode data selected by the language model into which the query has been input.

7. A user interaction method comprising: a first acquisition step of acquiring user information of a target user to be the subject of an interaction from a first database in which user information including at least information on areas of interest and information on knowledge for each user is registered; a second acquisition step of acquiring a list of episode data associated with the area of ​​interest of the target user from a second database in which episode data associated with at least one of a plurality of areas is registered; and a generation step of generating a query that instructs a language model to select episode data that the target user is expected to be interested in and not have knowledge of, using the information on the target user's knowledge acquired from the first database and the list of episode data acquired from the second database.

8. A user dialogue program characterized by causing a computer to execute: a first acquisition procedure for acquiring user information of a target user who is the subject of an interaction from a first database in which user information including at least information about interest genres and information about knowledge for each user is registered; a second acquisition procedure for acquiring a list of episode data associated with the target user's interest genre from a second database in which episode data associated with at least one of a plurality of genres is registered; and a generation procedure for generating a query that instructs a language model to select episode data that the target user is expected to be interested in and not have knowledge of, using the information about the target user's knowledge acquired from the first database and the list of episode data acquired from the second database.

Images