Information processing system, information processing method, and program

Abstract

[Problem] To make it possible to issue a conversational content in which a user has an interest. [Solution] An information processing system characterized by comprising: a response data generation unit that generates response data with which a character responds in response to conversation data for the character from a user and that generates the response data such that a line included in a work in which the character appears is cited; and an output unit that outputs the response data to the user.

Description

Information processing system, information processing method and program

[0001] The present invention relates to an information processing system, an information processing method, and a program.

[0002] A dialogue is taking place between a user and a computer (see Patent Document 1).

[0003] Patent No. 6719747

[0004] Scenario-based interactions may not keep users engaged.

[0005] The present invention has been made in view of the above background, and aims to provide a technology that enables a user to utter conversation content that interests the user.

[0006] The main invention of the present invention for solving the above problem is a response data generation unit that generates response data to which a character responds in accordance with conversation data from a user to a character, the generation unit generating the response data so as to quote lines included in a work in which the character appears, and an output unit that outputs the response data to the user.

[0007] Other problems and solutions disclosed in this application will be made clear in the section on preferred embodiments of the invention and the drawings.

[0008] According to the present invention, it is possible to generate conversation content that interests the user.

[0009] It is a diagram showing an example of the overall configuration of an information processing system. It is a diagram showing an example of the hardware configuration of a management server 2. It is a diagram showing an example of the software configuration of a management server 2. It is a diagram explaining the operation of a management server 2.

[0010] <System Overview> An information processing system according to one embodiment of the present invention will be described below. The information processing system of this embodiment is intended to enable conversations between a user and the system, in which characters appearing in a particular work (e.g., manga, anime, film, novel, game, etc.) engage in conversations and quote lines from the work (which may be lines from the character in question or lines from other characters) during the conversations.

[0011] In this embodiment, a text-based conversation is mainly described as an example, but the present invention is not limited to this. For example, the scope of the present invention also includes conversations in various formats, such as the following: (1) Voice-based conversation: A user's voice input is accepted, and a response is output in a character's voice using voice synthesis technology. (2) Conversation including images: Images and emojis sent by the user are analyzed, and a response is generated accordingly. The response can also include images showing the character's facial expressions and posture. (3) Video-based conversation: A more realistic conversation is realized by combining animations of characters or live-action video. (4) Conversation using AR (augmented reality) or VR (virtual reality): A character is projected into real or virtual space, providing a more immersive conversation environment. (5) Multimodal conversation: A combination of multiple formats, such as text, voice, images, and video, enables richer expression.

[0012] These various formats can be used alone or in combination. The following description will primarily focus on text-based conversations, but it goes without saying that the techniques of the present invention can also be applied to the other formats mentioned above.

[0013] 1 is a diagram showing an example of the overall configuration of an information processing system. The information processing system of this embodiment is configured to include a management server 2. The management server 2 is communicably connected to a user terminal 1 via a communication network. The communication network is, for example, the Internet, and is constructed using a public telephone line network, a mobile phone line network, a wireless communication path, Ethernet (registered trademark), or the like.

[0014] The user terminal 1 is a computer operated by a user, and may be, for example, a smartphone, a tablet computer, or a personal computer.

[0015] The management server 2 may be a general-purpose computer such as a workstation or a personal computer, or may be logically realized by cloud computing.

[0016] <Management Server> FIG. 2 is a diagram illustrating an example of the hardware configuration of the management server 2. Note that the illustrated configuration is an example, and other configurations may also be used. The management server 2 includes a CPU 201, memory 202, storage device 203, communication interface 204, input device 205, and output device 206. The storage device 203 stores various data and programs, and is, for example, a hard disk drive, solid state drive, or flash memory. The communication interface 204 is an interface for connecting to a communication network, and is, for example, an adapter for connecting to Ethernet (registered trademark), a modem for connecting to a public telephone network, a wireless communication device for wireless communication, or a USB (Universal Serial Bus) connector or RS232C connector for serial communication. The input device 205 is, for example, a keyboard, mouse, touch panel, button, microphone, or the like for inputting data. The output device 206 is, for example, a display, printer, speaker, or the like for outputting data. Each functional unit of the management server 2 described below is realized by the CPU 201 reading a program stored in the storage device 203 into the memory 202 and executing it, and each storage unit of the management server 2 is realized as part of the storage area provided by the memory 202 and the storage device 203.

[0017] 3 is a diagram showing an example of the software configuration of the management server 2. The management server 2 includes a work storage unit 231, an acquisition unit 211, a search unit 212, a generation unit 213, and an output unit 214.

[0018] The work storage unit 231 stores lines included in works of various formats. The works referred to here are not limited to manga, but include any creative works in which characters appear and lines and dialogue are included, such as novels, anime, movies, television dramas, stage plays, games, poems, and lyrics. The work storage unit 231 can store scenes of works and the lines included in the scenes. In this embodiment, it is assumed that scenes of works are image data for displaying pages of manga, and lines are text data. Scenes may also be image data for displaying frames of manga.

[0019] The information stored in the work storage unit 231 can be structured as follows according to the form of the work.

[0020] (1) In the case of manga: Dialogue: text data; Scene: image data for displaying the page or frame; Context information: information about the chapter or volume in which the dialogue appears. (2) In the case of novels: Dialogue: text data; Context: narrative text before and after the dialogue; Scene: a block of text containing the dialogue and narrative text; Metadata: chapter, page number, etc. (3) In the case of anime or movies: Dialogue: text data (subtitles or script); Audio data: audio file of the dialogue; Scene: a still image or short video clip of the scene; Time stamp: the time the dialogue appears in the work. (4) In the case of games: Dialogue: text data; Scene: description of the situation or event in the game in which the dialogue appears, captured video; Character status: the state of the character when the line is spoken (e.g., stamina, emotion, etc.). (5) In the case of stage plays or plays: Dialogue: text data; Scene: description of the scene in which the dialogue appears, captured video; Stage directions: acting instructions accompanying the dialogue; Act information: information about the act or scene in which the dialogue appears.

[0021] The work storage unit 231 can store the vector data in which the dialogue has been embedded in association with the dialogue. This embedding process can be applied regardless of the format of the work, and enables efficient dialogue search and similarity calculation.

[0022] The work storage unit 231 can also store additional information related to the work and characters. For example, the work storage unit 231 can store information related to the work's genre, production year, author information, character characteristics, personality, background setting, important events and turning points in the work, information related to the work's worldview and setting, etc. This additional information can be used to select appropriate lines that are more in line with the context and to generate responses that reflect the character's characteristics.

[0023] The acquisition unit 211 acquires conversation data for a character from a user. In this embodiment, the conversation data is text data input by the user to the user terminal 1. The acquisition unit 211 can receive the conversation data from the user terminal 1.

[0024] The search unit 212 searches for lines related to the acquired conversation data. The search unit 212 can search for lines related to the conversation data from the work storage unit 231. As a method for searching for lines, the following multiple methods can be adopted.

[0025] (1) Cosine similarity: The cosine distance between the vector data in which the conversation data has been embedded and the vector data stored in the work storage unit 231 is calculated, and lines with a short distance are determined to be highly related.

[0026] (2) Euclidean distance: Calculate the straight-line distance in vector space and select the lines that are close.

[0027] (3) Manhattan distance: The sum of the absolute values ​​of the differences in each dimension of the vector is calculated, and the lines with the closest distance are selected.

[0028] (4) Jaccard similarity: The conversation data and the lines are treated as a set of words, and the similarity is determined by calculating the percentage of common words.

[0029] (5) Edit distance (Levenshtein distance): The minimum number of string edit operations between the conversation data and the lines is calculated, and the lines with the closest distance are selected.

[0030] (6) Latent Semantic Analysis (LSA): Calculates similarity by taking into account the latent semantic relationship between the conversation data and the lines.

[0031] (7) BM25 algorithm: A ranking algorithm widely used in information retrieval is applied to select highly relevant lines.

[0032] The search unit 212 can use these methods alone or in combination with multiple methods, and can also dynamically select an appropriate method depending on the context of the conversation and the required accuracy.

[0033] The generation unit 213 generates text (hereinafter, response data) that responds to the conversation data. The generation unit 213 can generate the response data so as to quote lines from works in which the characters appear. Note that the generation unit 213 does not use the lines themselves as response data, but generates response data so that the lines are included as quotes along with the main text. The following multiple methods can be used to generate response data.

[0034] (1) Large-scale language models (LLMs): Using LLMs such as GPT-3, GPT-4, PaLM, and LLaMA, responses are generated using conversational data and prompts containing dialogue.

[0035] (2) Rule-based system: Generates responses by combining conversation data and lines based on predefined response patterns and rules.

[0036] (3) Search-based system: Searches for examples of dialogue similar to the conversation data from a large dialogue corpus, and generates new responses based on those responses.

[0037] (4) Template-based generation: A response is generated by embedding conversation data and dialogue information into a pre-prepared template.

[0038] (5) Machine translation approach: Responses are generated by converting conversation data into an intermediate representation and then “translating” it to match the character’s tone and setting.

[0039] (6) Reinforcement learning model: A reward function is defined and responses are generated using a reinforcement learning model trained to maximize character-likeness and naturalness of conversation.

[0040] (7) Neural dialogue model: Responses are generated using a neural network model specialized for dialogue, such as a sequence-to-sequence model or a model with an attention mechanism.

[0041] The generator 213 can use these techniques alone or in combination. For example, a rule-based system can be used to generate a basic response structure, and then LLM can be used to refine the response. It is also possible to dynamically select an appropriate technique depending on the characteristics of the character and the complexity of the conversation.

[0042] If the search unit 212 finds lines related to the conversation data, the generation unit 213 generates response data that quotes the lines, and if the lines are not found, the generation unit 213 can generate response data as a response to the conversation data without quoting the lines. When quoting lines, the generation unit 213 appropriately selects a quoting method and incorporates the lines into the response data in a natural manner. Even when lines are not quoted, an appropriate response is generated that takes into account the characteristics of the character and the context of the conversation.

[0043] In this embodiment, the generation unit 213 can generate response data that does not quote lines by providing a prompt including conversation data and an instruction to create a response to the conversation data to the large-scale language model. Also, the generation unit 213 can generate response data that quotes lines by providing a prompt including conversation data, searched lines, and an instruction to create a response to the conversation data so as to quote the lines to the large-scale language model.

[0044] The generation unit 213 can employ the following methods for quoting lines: (1) Direct quotation: Inserting the retrieved lines as they are into the response data. (2) Partial quotation: Extracting a portion of the retrieved lines and inserting it into the response data. (3) Paraphrase: Incorporating the retrieved lines into the response data using a different expression while retaining their meaning. (4) Free translation: Capturing the essential meaning and emotion of the retrieved lines, changing the expression to match the current conversation context, and incorporating it into the response data. The generation unit 213 can appropriately select these quotation methods based on the flow of the conversation, the length of the retrieved lines, the user's preferences, and the like. In this embodiment, direct quotation is assumed as the method for quoting lines.

[0045] Furthermore, the generation unit 213 can use the following criteria to select lines to quote: (1) Relevance: Lines with a semantic similarity between the conversation data and the lines that exceeds a predetermined threshold are selected. (2) Emotional matching: Lines whose emotional expression matches the emotional state of the user estimated from the conversation data are preferentially selected. (3) Character matching: Lines spoken by characters in conversation are preferentially selected. (4) Importance: Lines that are highly important or memorable within the work are preferentially selected. (5) Diversity: Lines that have not been used in past conversations are preferentially selected to avoid excessive repetition of the same lines. The generation unit 213 can select more appropriate lines by using a combination of these criteria.

[0046] The generation unit 213 can perform the following processes to naturally incorporate the selected lines into the response data: (1) Preface generation: Generate an appropriate preface before quoting the lines (e.g., "Come to think of it, there was a line like this"). (2) Post-explanation: Generate a sentence after quoting the lines to explain the intention and relevance of the quote. (3) Context adjustment: Generate sentences that match the flow of the conversation before and after the selected lines, thereby incorporating the lines naturally. (4) Maintaining character: Generate sentences that reflect the character's tone and personality, even in response data other than the quoted parts. Through these processes, the generation unit 213 can generate more natural and seamless response data.

[0047] The large-scale language model (LLM) used by the generation unit 213 may be, for example, GPT-3, GPT-4, PaLM, LLaMA, or a model with equivalent performance. The following procedure may be adopted as a method for using the LLM.

[0048] (1) Initializing the LLM: Load the LLM to be used and fine-tune it as necessary. During fine-tuning, the model is trained using a related dataset to reflect the character's characteristics and the worldview of the work. In this embodiment, it is assumed that the trained LLM is used as is.

[0049] (2) Context setting: A system prompt containing background information for the conversation and character settings is created and given to the LLM. (3) Conversation history management: A conversation history with the user is maintained and input to the LLM for each turn. (4) Prompt generation: A prompt containing conversation data, searched lines, and instructions for response generation is dynamically generated. (5) Input to LLM: The generated prompt is input to the LLM and a response is obtained. The LLM may be provided on the management server 2, or the response may be obtained by calling the API of an external server that performs generation processing using the LLM. (6) Post-processing: The output of the LLM is adjusted as necessary to produce the final response data.

[0050] Examples of prompts include the following:

[0051] Example 1: When quoting a line System: Act as [character name]. [Brief character description]. Quote the given line naturally in the conversation. User: [user's conversation data] Related line: "[Line from the work]" Instructions: Respond to the user's conversation data in a way that sounds like [character name], quoting the above line naturally.

[0052] Example 2: If no lines were found: System: Please act as [character name]. [Brief character description]. User: [User's conversation data] Instructions: Please respond to the user's conversation data in a way that is appropriate for [character name]. Please do not quote lines from the work, but generate your own responses that reflect the character's personality and tone.

[0053] The generator 213 uses these prompts as templates and can dynamically change the content depending on the actual conversation situation. For example, the generator 213 replaces the character name, character description, user conversation data, related lines, etc. with appropriate values ​​each time and inputs them into the LLM. It is also possible to flexibly change the structure and content of the prompts depending on the flow of the conversation and the characteristics of the retrieved lines.

[0054] The generation unit 213 may include some of the lines in the prompt and perform learning by short-shot learning.

[0055] The output unit 214 outputs the generated response data to the user. The output unit 214 transmits the response data to the user terminal 1, which can display the response data. The output unit 214 can output a scene corresponding to the searched line, which is stored in the work storage unit 231, to the user. After outputting the response data, the output unit 214 can output a scene in response to a request from the user.

[0056] <Operation> FIG. 4 is a diagram illustrating the operation of the management server 2.

[0057] The management server 2 acquires conversation data (S301), searches for lines related to the conversation data (S302), and if the lines are found (S303: YES), generates response data quoting the lines (S304), and if the lines are not found (S303: NO), generates response data without quoting the lines (S305), and transmits the response data to the user terminal 1 (S306). In response to a request from the user terminal 1, the management server 2 can read out a scene corresponding to the lines from the work storage unit 231 and transmit the scene to the user terminal 1 (S307).

[0058] As described above, according to the information processing system of this embodiment, in a conversation between a user and the system, it is possible to respond by quoting lines from a work related to conversation data from the user.

[0059] Although the present embodiment has been described above, the above embodiment is intended to facilitate understanding of the present invention and is not intended to limit the present invention. The present invention may be modified or improved without departing from the spirit thereof, and equivalents thereof are also included in the present invention.

[0060] For example, the processing by each of the functional units of the management server 2 described above may be performed by any of the functional units. Also, a different functional unit that performs part of the processing by each of the functional units described above may be added. Also, the functional units of the management server 2 may be distributed across multiple computers.

[0061] Furthermore, the information stored in each storage unit of the management server may be stored in any of the storage units. That is, the information stored in the above-mentioned multiple storage units may be stored in one storage unit, or part of the information stored in one storage unit may be stored in another storage unit.

[0062] <Modification 1> Modification 1 has a function for handling multimodal input in addition to the configuration of the above embodiment. The following will mainly describe the differences from the above embodiment.

[0063] The management server 2 of the first modification may include a voice recognition unit and an image recognition unit in addition to the configuration of the above embodiment.

[0064] The speech recognition unit analyzes speech data input by the user and converts it into text data. The speech recognition unit uses a speech recognition model based on deep learning, for example, to perform highly accurate speech-to-text conversion. It also extracts prosodic information such as intonation and tone of the speech, which can be used to estimate the user's emotional state.

[0065] The image recognition unit analyzes image data input by the user and recognizes objects, facial expressions, situations, etc. in the image. The image recognition unit uses deep learning models such as convolutional neural networks (CNN) to extract image features. The recognition results are output as text data.

[0066] The acquisition unit 211 can accept not only text data but also voice data and image data as input. When voice data is input, the acquisition unit 211 converts it into text using a voice recognition unit. When image data is input, the acquisition unit 211 converts the image content into text using an image recognition unit.

[0067] The search unit 212 searches for related lines based on the content of the audio data or image data converted into text. For example, if a user sends an image of a sad expression, the search unit 212 can search preferentially for sad scenes or comforting lines within the work.

[0068] The generation unit 213 generates response data taking into consideration the characteristics of multimodal input. For example, it can estimate the emotion of the user from the voice input and quote lines in a tone that matches the emotion, or it can preferentially use lines related to the content of the image sent by the user.

[0069] The output unit 214 not only outputs the generated response data as text, but can also use voice synthesis technology to read the response in a character's voice, or simultaneously display images or videos related to the response.

[0070] <Modification 2> In addition to the configuration of the above embodiment, Modification 2 has a function of simultaneously conversing with multiple characters. The following will mainly explain the differences from the above embodiment.

[0071] The management server 2 of the second modification may include a character management unit and a dialogue control unit in addition to the configuration of the above embodiment.

[0072] The character management unit manages information about multiple characters participating in a conversation. It stores information about each character's settings, personality, and relationships with other characters, and has the function of selecting an appropriate character depending on the context of the conversation.

[0073] The dialogue control unit controls the flow of dialogue between multiple characters. It decides which character will respond to what the user says and what kind of dialogue will take place between the characters. It also manages the dialogue history between characters to maintain naturalness and consistency in the conversation.

[0074] In addition to the configuration of the above embodiment, the work storage unit 231 also stores scenes of dialogue between characters and scenes of group conversations, making it possible to quote appropriate lines and dialogue patterns even in situations involving multiple characters.

[0075] When searching for lines related to the user's conversation data, the search unit 212 searches not only for lines of a single character but also for dialogue scenes involving multiple characters.

[0076] The generation unit 213 has the function of simultaneously generating responses for multiple characters. Response data is generated to reproduce natural dialogue, taking into account the characteristics and relationships of each character. When generating dialogue between characters, the following factors are taken into consideration: (1) Relationships between characters (friendly, antagonistic, hierarchical, etc.) (2) Personality and speaking characteristics of each character (3) Context and situation of the conversation (4) Content and intention of user's remarks (5) Past dialogue history

[0077] The output unit 214 presents the generated responses of the multiple characters to the user in a format that allows each character to be distinguished. For example, in the case of text chat, the comments of each character can be displayed in a different color or speech bubble, or can be accompanied by a character icon.

[0078] Although the second modification has been described as a format in which the user directly converses with multiple characters, a role-playing style dialogue is also possible in which the user controls a specific character and converses with other characters as that character. In this case, the generation unit 213 may have a function to suggest appropriate response candidates in consideration of the characteristics of the character controlled by the user.

[0079] <Modification 3> In addition to the configuration of the above embodiment, Modification 3 has a function of generating more appropriate quotes and responses by taking into account past conversation history. The following will mainly describe the differences from the above embodiment.

[0080] The management server 2 of the third modification includes a conversation history storage unit and a context analysis unit in addition to the configuration of the above embodiment.

[0081] The conversation history memory stores the history of past conversations between the user and the system. It saves the user's comments, the system's responses, quoted lines, and conversation time information for each conversation session. It also records important information mentioned during the conversation (e.g., the user's preferences and experiences).

[0082] The context analysis unit has the function of analyzing the current conversation data and past conversation history to understand the context of the conversation. Specifically, it performs the following processes: (1) Tracking the progression of the topic (2) Estimating the user's emotional state (3) Understanding the long-term purpose and direction of the conversation (4) Analyzing the relationship between the user and the character (5) Recording previously quoted lines and avoiding duplication

[0083] The search unit 212 uses the information obtained from the context analysis unit to search for lines that are more appropriate for the context. For example, the following search criteria can be added: (1) Prioritizing lines related to the current topic, (2) Prioritizing lines that have not been quoted in the past, (3) Selecting lines that fit the user's current emotional state, and (4) Selecting lines that are in line with the long-term purpose of the conversation.

[0084] The generation unit 213 generates more appropriate responses by taking into account the information obtained from the context analysis unit. Specifically, it realizes the following functions: (1) Appropriate reference to information mentioned in past conversations (2) Maintaining consistency in the conversation (avoiding contradictory statements) (3) Developing topics based on the user's interests and reactions (4) Structuring long-term conversations (introduction, development, conclusion, etc.) (5) Adjusting the response style according to the development of the relationship with the user

[0085] As described above, according to Modification 3, it is possible to generate more appropriate quotes and responses by taking into account past conversation history, thereby realizing long-term consistency, more natural dialogue, and dialogue that is in line with the context.

[0086] In addition, in Variation 3, it is important to consider the user's privacy by appropriately setting the retention period and scope of use of the conversation history. For example, a function can be implemented that saves the conversation history with the user's consent and automatically deletes it after a certain period of time. It is also possible to clearly explain the purpose of using the conversation history and provide a mechanism that allows the user to request deletion of the history or suspension of use.

[0087] The functions of this embodiment can also be implemented in combination with Modification 1 and Modification 2. For example, a more sophisticated dialogue experience can be provided by taking into account the history of multimodal inputs and tracking the development of relationships with each character in a conversation with multiple characters.

[0088] <Modification 4> In addition to the configuration of the above embodiment, Modification 4 has the function of learning the user's preferences and interests and selecting lines and generating responses according to them. The following will mainly explain the differences from the above embodiment.

[0089] The management server 2 of the fourth modification example includes a user profile storage unit, a preference learning unit, and a preference consideration unit in addition to the configuration of the above-described embodiment.

[0090] The user profile storage unit stores each user's preference information. Specifically, the following information can be saved: (1) Favorite characters (2) Interesting genres and topics (3) Frequently quoted lines and their characteristics (4) Conversation patterns that users respond well to (5) Frequency and time of use of the user (6) Basic user attribute information (age group, gender, etc., collected with the user's consent)

[0091] The preference learning unit has the function of learning the user's preferences from conversation data with the user and the user's behavior. Specifically, it can perform the following processes: (1) Analyzing the content of user comments (2) Tracking user reactions (e.g., use of the "Like" button, duration of conversations) (3) Extracting topics frequently mentioned by users (4) Analyzing the user's preferred language and expression style (5) Measuring the user's level of interest in specific characters or works

[0092] The preference learning unit periodically updates the information in the user profile storage unit based on the results of these analyses.

[0093] The preference consideration unit has a function of referring to the preference information stored in the user profile storage unit and reflecting it in the selection of lines and the generation of responses.

[0094] The search unit 212 uses information obtained from the preference consideration unit to search preferentially for lines that match the user's preferences. For example, the following search criteria can be added: (1) Prioritizing lines of characters that the user likes (2) Selection of lines related to genres or topics that the user is interested in (3) Selection of lines that are similar to the characteristics of lines to which the user has responded favorably in the past

[0095] The generation unit 213 generates a response that matches the user's preferences, taking into account the information obtained from the preference consideration unit. Specifically, it realizes the following functions: (1) Adopting the user's preferred language and expression style (2) Leading the conversation to topics that interest the user (3) Selecting a character that matches the user's preferences (in the case of multiple characters) (4) Reproducing conversation patterns that the user responded well to

[0096] As described above, according to the fourth modification, it is possible to learn the user's preferences and interests and select lines and generate responses that match those preferences and interests, thereby providing the user with a more engaging and personalized conversation experience.

[0097] In addition, in the fourth modification, sufficient care must be taken in handling user preference information. For example, it is desirable to take the following measures: (1) Collect and use preference information only after obtaining explicit consent from the user. (2) Clearly explain the scope of information to be collected and the purpose of use. (3) Provide a function that allows users to check, modify, and delete their own preference information. (4) Implement appropriate security measures, such as encrypting preference information and controlling access.

[0098] The functions of this embodiment can also be implemented in combination with the other embodiments described above. For example, by combining it with Modification 3, a more advanced dialogue system can be realized that takes into account both past conversation history and user preferences.

[0099] <Modification 5> In addition to the configuration of the above embodiment, Modification 5 has a function of incorporating new works and added lines in real time and reflecting the latest information. The following will mainly explain the differences from the above embodiment.

[0100] The management server 2 of the fifth modification example includes an update monitoring unit, a data acquisition unit, and an integration processing unit in addition to the configuration of the above-described embodiment.

[0101] The update monitoring unit has the function of periodically checking for new work information and additional dialogue. Specifically, it performs the following processes: (1) Periodically accessing external data sources (publisher APIs, official websites, etc.) (2) Receiving update notifications using RSS feeds, Webhooks, etc. (3) Monitoring the update date and time of the work database.

[0102] The data acquisition unit has the function of acquiring new information detected by the update monitoring unit. Specifically, it performs the following processes: (1) Acquisition of metadata for new works (title, author, release date, etc.) (2) Download of new dialogue data (3) Acquisition of updated work information

[0103] The integration processing unit has the function of integrating new information acquired by the data acquisition unit with existing data. Specifically, it performs the following processes: (1) Adding new work data to the work storage unit 231 (2) Adding or updating lines to existing works (3) Converting the format of new data (e.g., converting text data to embedded vectors) (4) Checking data consistency and handling errors

[0104] The work storage unit 231 is always kept up to date by the integration processing unit, which allows the search unit 212 and the generation unit 213 to always use the latest work information and lines.

[0105] As described above, according to the fifth modification, new works and added lines can be incorporated in real time, and dialogue can be always carried out that reflects the latest information.

[0106] In addition, in Modification 5, it is necessary to appropriately set the data update frequency and acquisition timing. In order to avoid excessive system load due to the update process, it is possible to devise measures such as adjusting the update frequency or executing the update process during a time period when the system load is low.

[0107] Quality control of newly added data is also important. By implementing a function in the integrated processing unit to check the validity of new data and filter inappropriate content, the reliability and safety of the system can be ensured.

[0108] The function of Modification 5 can also be implemented in combination with the other modifications described above. For example, by combining it with Modification 4, a more engaging dialogue experience can be realized, such as by preferentially providing the latest lines and work information that match the user's preferences.

[0109] <Disclosures> The present disclosure also includes the following configurations. [Item 1] An information processing system comprising: a response data generation unit that generates response data with which a character responds in accordance with conversation data from a user to a character, the generation unit generating the response data so as to quote a line included in a work in which the character appears; and an output unit that outputs the response data to the user. [Item 2] The information processing system according to Item 1, comprising: a work storage unit that stores the line included in the work; and a search unit that searches the work storage unit for the line related to the conversation data, wherein the generation unit provides a prompt to a large-scale language model, the prompt including the conversation data, the searched line, and an instruction to create a response to the conversation data so as to quote the line, thereby generating the response data. [Item 3] The information processing system according to Item 1, comprising: a work storage unit that stores scenes of the work and lines included in the scenes; and a search unit that searches the work storage unit for the lines related to the conversation data, wherein the output unit acquires the scenes corresponding to the lines from the work storage unit and outputs the acquired scenes to the user. [Item 4] The information processing system according to Item 1, comprising: a search unit that searches for the lines related to the conversation data, wherein, if the lines are found, the generation unit generates the response data so as to quote the lines, and if the lines are not found, generates the response data as a response to the conversation data without quoting the lines. [Item 5] The information processing system according to Item 1, wherein the conversation data generation unit generates the second conversation data so as to quote the lines of a second character different from the first character.[Item 6] An information processing method executed by a computer comprising the steps of: generating response data in response to conversation data from a user to a character, with which the character responds; and outputting the response data to the user, wherein in the generating step, the computer generates the response data so as to quote a line included in a work in which the character appears. [Item 7] A program for causing a computer to execute the steps of: generating response data in response to conversation data from a user to a character, with which the character responds; and outputting the response data to the user, wherein in the generating step, the computer generates the response data so as to quote a line included in a work in which the character appears.

[0110] 1 User terminal 2 Management server

Claims

1. An information processing system comprising: a response data generation unit that generates response data with which a character responds in response to conversation data from a user to a character, the generation unit generating the response data so as to quote lines included in a work in which the character appears; and an output unit that outputs the response data to the user.

2. An information processing system as described in claim 1, comprising: a work storage unit that stores the lines included in the work; and a search unit that searches the work storage unit for the lines related to the conversation data, wherein the generation unit generates the response data by providing a prompt to a large-scale language model that includes the conversation data, the searched lines, and an instruction to create a response to the conversation data that quotes the lines.

3. An information processing system as described in claim 1, comprising: a work storage unit that stores scenes of the work and lines contained in the scenes; and a search unit that searches the work storage unit for the lines related to the conversation data, wherein the output unit obtains the scenes corresponding to the lines from the work storage unit and outputs the obtained scenes to the user.

4. An information processing system as described in claim 1, comprising a search unit that searches for the lines related to the conversation data, and wherein the generation unit generates the response data so as to quote the lines if the lines are found, and generates the response data as a response to the conversation data without quoting the lines if the lines are not found.

5. An information processing system according to claim 1, wherein the conversation data generation unit generates the second conversation data so as to quote the lines spoken by a second character different from the first character.

6. An information processing method executed by a computer, which comprises the steps of: generating response data in response to conversation data from a user to a character; and outputting the response data to the user; wherein in the generating step, the computer generates the response data so as to quote lines from a work in which the character appears.

7. A program for causing a computer to execute the steps of: generating response data in response to conversation data from a user to a character; and outputting the response data to the user, wherein in the generating step, the computer is caused to generate the response data so as to quote lines from a work in which the character appears.

Images