Memory management system, memory management method, and program

Abstract

This system provides a memory management system that can offer memory information tailored to a work session to an AI assistant. [Solution] The system includes a management unit that assigns hierarchical information indicating the level of abstraction of the stored information, importance information indicating the importance of the stored information, and contextual information indicating the work situation in which the stored information is generated or used to each piece of stored information stored in the storage unit, and manages the hierarchical information, importance information, and contextual information as attribute information associated with the stored information; and a provisioning unit that determines the contextual information corresponding to a work session based on information about the work session, evaluates the relationship of each piece of stored information to the work session based on the determined contextual information and the hierarchical information and importance information assigned to each piece of stored information, selects the stored information related to the work session from the stored information stored in the storage unit based on the evaluation result, and provides the selected stored information to the AI ​​assistant.

Description

【Technical Field】 【0001】 The present invention generally relates to a technique for creating digital works through cooperation between humans and AI assistants. 【Background Art】 【0002】 In recent years, with the development of generative AI, efforts have been expanding for humans and AI assistants to collaborate in creating digital works. In such collaborative work, it is required that the AI assistant generates an output in line with the human intention while referring to past work history and knowledge information. However, conventional generative AI mainly focuses on optimizing the output for a given prompt, and it is difficult to continuously select the optimal information by cross - considering the level of abstraction, importance of the work, and differences in context. 【0003】 In such a situation, the importance of a technology for appropriately managing the information provided to the AI assistant and selectively providing necessary information according to the work situation is increasing. In particular, in continuous collaborative work between humans and AI assistants, since the accumulated memory information becomes enormous, a mechanism for efficiently selecting and providing highly relevant information for each work session is required. 【0004】 Recently, an information processing apparatus that designs and executes tasks through a chatbot using generative AI has been disclosed (Patent Document 1). In the technique described in Patent Document 1, a configuration is adopted in which the first generative AI designs a task and the second generative AI executes the designed task, and it is described that a storage for storing knowledge information and design information is provided. 【Prior Art Documents】 【Patent Documents】 【0005】 【Patent Document 1】 Japanese Patent No. 7691086 【Summary of the Invention】 [Problems that the invention aims to solve] 【0006】 While Patent Document 1 discloses a configuration in which the design and execution of tasks are divided between a first generation AI and a second generation AI, it does not adequately consider how to evaluate the information required for each work session and efficiently select and provide it to the AI ​​assistant within the limited range of input information when the amount of information accumulated increases as collaborative work progresses. 【0007】 Based on the above, this invention proposes a memory management system and the like that can provide memory information corresponding to a work session to an AI assistant. [Means for solving the problem] 【0008】 To solve the above problems, the present invention provides a memory management system for creating digital works through collaboration between a human and an AI assistant, comprising: a memory unit for storing memory information; a management unit that assigns to each piece of memory information stored by the memory unit hierarchical information indicating the level of abstraction of the memory information, importance information indicating the importance of the memory information, and contextual information indicating the work situation in which the memory information is generated or used, and manages the hierarchical information, importance information, and contextual information as attribute information associated with the memory information; and a provisioning unit that determines the contextual information corresponding to the work session based on information about the work session, evaluates the relationship of each piece of memory information to the work session based on the determined contextual information and the hierarchical information and importance information assigned to each piece of memory information stored by the memory unit, selects memory information related to the work session from the memory information stored by the memory unit based on the evaluation result, and provides the selected memory information to the AI ​​assistant. [Effects of the Invention] 【0009】 According to the present invention, memory information corresponding to the work session can be provided to the AI ​​assistant. Problems, means, and effects not shown above will be revealed by the following description of embodiments. [Brief explanation of the drawing] 【0010】 [Figure 1] This figure shows an example of a memory management system according to the first embodiment. [Figure 2] This figure shows an example of a memory management infrastructure according to the first embodiment. [Figure 3] This figure shows an example of a conceptual configuration according to the first embodiment. [Figure 4] This figure shows an example of hierarchical information according to the first embodiment. [Figure 5] This figure shows an example of contextual information according to the first embodiment. [Figure 6] This figure shows an example of importance information according to the first embodiment. [Figure 7] This figure shows an example of the registration process according to the first embodiment. [Figure 8] This figure shows an example of the search processing according to the first embodiment. [Figure 9] This figure shows an example of the classification process according to the first embodiment. [Figure 10] This figure shows an example of the determination process according to the first embodiment. [Figure 11] This figure shows an example of the calculation process according to the first embodiment. [Figure 12] This figure shows an example of an information processing device according to the second embodiment. [Modes for carrying out the invention] 【0011】 (I) First Embodiment The configurations, procedures, or other elements disclosed below are provided to illustrate embodiments of the present invention and do not limit it. The descriptions based on the drawings are intended to aid understanding, and the shapes, arrangements, functions, etc., of elements may be omitted or simplified as necessary. The present invention is not limited to one or more of the disclosed embodiments and can also be implemented by functionally or structurally equivalent means, or other means achieving similar technical objectives, to the extent that a person skilled in the art can understand from this specification. Unless otherwise specified, each component in this specification is construed as including "at least one." Furthermore, a singular form of a word includes its plural form, and a plural form of a word includes its singular form. In addition, terms used herein are not limited to a specific meaning, to the extent that they are clear from the context, and should be appropriately interpreted by a person skilled in the art. For example, the expression "including" is construed as meaning not limited to those listed. 【0012】 Next, the characteristics of information retrieval in a work session with respect to the memory management system according to this embodiment will be described. Generally, humans use information with a high level of abstraction and information with a low level of abstraction depending on the purpose of the work session. Humans prioritize referencing information with a high level of importance. In this regard, conventional methods may employ a configuration that extracts relevant information based on the similarity between the search query and the stored information. Conventional methods may not have a configuration that cross-sectionally combines and evaluates hierarchical information indicating the level of abstraction, importance information, and contextual information indicating the work status. Conventional methods may not be able to sufficiently narrow down the information necessary for a work session, and excessive information may be provided to the AI ​​assistant. 【0013】 As the number of candidate pieces of information presented in a work session increases, the burden on humans to compare and select that information also increases. Humans have a limit to the number of information chunks they can grasp in a short amount of time. Miller's Law is a well-known insight regarding human cognitive load. Miller's Law states that the number of information chunks that humans can simultaneously hold and process is approximately 7 ± 2. 【0014】 In the conventional method, ranking biased toward similarity may be performed in ranking candidate information. The conventional method may not control the number of pieces of candidate information based on the cognitive load of a human. It is desirable for the memory management system according to the present embodiment to provide information useful for a work session while suppressing the cognitive load of a human by controlling the number of pieces of memory information provided to an AI assistant. 【0015】 In a conventional chat-type generative AI, as the conversation progresses, past conversation contents (or summaries thereof) are sequentially input to the model side, and the input information may increase and the cognitive area may be compressed. For this reason, necessary information and unnecessary information for a response may be mixed, and optimal information provision for a work session may be hindered. 【0016】 Also, even in a method of summarizing and inputting past conversations, if it does not have a configuration for crosswise selection of information necessary for a work session based on abstraction level, importance level, and work situation (context), effective utilization of the cognitive area may not be sufficient. 【0017】 This memory management system assigns hierarchical information, importance information, and context information as attribute information to memory information. This memory management system determines context information based on information related to a work session. In addition to hierarchical information, importance information, and context information, this memory management system evaluates the relevance of memory information for a work session using the similarity between a search query and memory information and the keyword matching degree. In addition to ranking and filtering based on relevance evaluation, this memory management system performs count control as cognitive load management based on Miller's law. 【0018】 This memory management system manages memories (external knowledge) by coordinate them using three attributes: abstraction level (Layer), importance (Priority), and work context (Context). Based on the input of the work session, the system automatically determines the context and then selects the information to be injected using a composite evaluation with the three attributes, before injecting it into the cognitive domain of the generating AI. With this configuration, compared to, for example, simply selecting based on vector similarity or inputting a summary of conversation history in bulk, it is possible to stably present information that matches the work objective while suppressing cognitive domain congestion, the inclusion of unnecessary information, and ranking bias. 【0019】 Humans may create digital works such as proposals, reports, design documents, specifications, operating manuals, training materials, presentation materials, explanatory materials including videos, and descriptions of processing procedures performed on information processing devices. AI assistants may generate responses by referring to stored information, including work history, work policies, key points of reference materials, terminology definitions, and reasoning behind decisions, when creating digital works. 【0020】 This memory management system can provide useful information to the AI ​​assistant for each work session by selecting and providing memory information according to the work session, even if the type of digital artwork being created by the human differs. 【0021】 The following will be a detailed explanation using the drawings. In this specification, elements that are identical or functionally similar to the components shown in the drawings are denoted by the same reference numerals. The use of reference numerals in this specification is not limited to the specific embodiments shown, but is also applicable to various modifications including at least one component. 【0022】 Figure 1 shows an example of a memory management system 100. 【0023】 The memory management system 100 comprises a user terminal 110, a message sharing device 120, an external LLM device 130, and a memory management base 140. The user terminal 110, the message sharing device 120, and the memory management base 140 are connected to each other via a network 101 so that they can communicate with one another. The memory management base 140 and the external LLM device 130 are also connected to each other via the network 101 so that they can communicate with one another. 【0024】 The user terminal 110 is a terminal for the user to use the processing results generated by the AI ​​(AI assistant). The user terminal 110 accepts user input. The user terminal 110 sends a message based on the user input to the message sharing device 120. The user terminal 110 displays the response message received from the message sharing device 120. User input may include, for example, a written question, a work instruction, a description of the work object, and at least one of the instructions for correcting the work result. 【0025】 The message sharing device 120 is a device that relays messages between the user terminal 110 and the memory management infrastructure 140. The message sharing device 120 transmits messages received from the user terminal 110 to the memory management infrastructure 140. The message sharing device 120 transmits response messages received from the memory management infrastructure 140 to the user terminal 110. The message sharing device 120 can store the message sequence as work session information. The work session information is, for example, a time-series combination of input messages from the user terminal 110 and response messages generated by the external LLM device 130. As a specific example of the message sharing device 120, a bot operating on a chat service (for example, a bot that relays messages between users in a workspace and the memory management infrastructure 140) may be used. 【0026】 The message sharing device 120 stores input messages received from the user terminal 110 and response messages received from the memory management infrastructure 140 in chronological order as work session information. In addition to input messages and response messages, the message sharing device 120 may also store information about attachments referenced in the work session as work session information. The message sharing device 120 may store at least some of the following as information about attachments: the attachment identifier, the attachment type, the attachment name, the attachment body text, a summary of the attachment body text, and reference information indicating where the attachment was obtained. 【0027】 The memory management infrastructure 140 is a platform for generating input information to be transmitted to the external LLM device 130. When generating input information, the memory management infrastructure 140 may use input messages received from the message sharing device 120, the work session history, information about attached files, and stored information as information about the work session. Stored information includes, for example, information that includes at least one of the following: prerequisites used in the production of a digital work, constraints, specialized knowledge, policies, judgment criteria, and self-check criteria. The memory management infrastructure 140 may manage the number of references, the last access time, and the generation time or update time as metadata associated with each piece of stored information. The memory management infrastructure 140 may use a predetermined number of messages from a sequence of messages that chronologically hold user inputs and responses generated by the external LLM device 130 during the work session as the work session history. 【0028】 The external LLM device 130 is a device that generates response messages using a generation AI. The external LLM device 130 generates response messages based on input information received from the memory management base 140. The external LLM device 130 transmits the generated response messages to the memory management base 140. For example, the external LLM device 130 generates response messages related to drafting design documents, summarizing specifications, proposing implementation policies, and estimating the causes of defects. 【0029】 The memory management infrastructure 140 is a platform for managing memory information supplied to the external LLM device 130. The memory management infrastructure 140 can be provided as a cloud service. The memory management infrastructure 140 determines the context representing the work status from the work session information. The memory management infrastructure 140 manages each of the multiple memory information stored by the memory management infrastructure 140 by associating it with hierarchical information, importance information, and contextual information. Based on the determined context, hierarchical information, and importance information, the memory management infrastructure 140 selects the memory information to provide to the external LLM device 130. The memory management infrastructure 140 integrates the selected memory information as part of the input information and transmits it to the external LLM device 130. 【0030】 The memory management infrastructure 140 can switch the memory information provided to the external LLM device 130 according to the context. For example, when the context of implementation is determined, the memory management infrastructure 140 can prioritize selecting memory information representing implementation constraints and design policies. For example, when the context of investigation is determined, the memory management infrastructure 140 can prioritize selecting memory information representing terminology definitions and evaluation perspectives. By referring to the input information (including the selected memory information) received from the memory management infrastructure 140, the external LLM device 130 can more easily generate context-appropriate response messages, even if the amount of work session information increases. The memory management infrastructure 140 transmits the response messages received from the external LLM device 130 to the message sharing device 120. 【0031】 In this embodiment, a configuration in which the memory management base 140 generates input information and transmits it to the external LLM device 130 is illustrated, but the embodiment is not limited to this. For example, the message sharing device 120 may transmit the input message received from the user terminal 110 to the external LLM device 130, and the memory management base 140 may transmit the storage information necessary for the external LLM device 130 to generate a response message to the external LLM device 130. The external LLM device 130 may generate a response message based on the input message received from the message sharing device 120 and the storage information received from the memory management base 140, and transmit the generated response message to the message sharing device 120. The memory management base 140 may acquire the response message from the external LLM device 130 or information regarding the generation of the response message via the message sharing device 120 or through a communication path separate from the message sharing device 120, and use this information to update the usage history and importance information. 【0032】 Figure 2 shows an example of the memory management infrastructure 140. 【0033】 The memory management infrastructure 140 comprises a server device 210, a database device 220, and an object storage device 230. The server device 210 is connected to the database device 220 and the object storage device 230 in a communicative manner. The server device 210 can provide a container execution environment as an execution environment for executing the functions of the memory management infrastructure 140. The server device 210 may be a single device or may consist of multiple devices. If it consists of multiple devices, the server device 210 may distribute and execute at least a portion of the functions of the memory management infrastructure 140 across the multiple devices. 【0034】 The server device 210 comprises a storage unit 211, a management unit 212, a provision unit 213, and an update unit 214. The storage unit 211 stores memory information used in the process of creating digital works through collaboration between a human and an AI assistant. 【0035】 The storage unit 211 may be implemented, for example, as a program (storage control unit) that controls writing to and reading from the storage device. The storage information managed by the storage unit 211 may be stored, for example, as structured files (e.g., multiple JSON files) in a local environment, or as object storage and a database in a cloud environment (e.g., a configuration in which storage data is stored in object storage and metadata is stored in a database). Furthermore, when accommodating multiple users, a multi-tenant configuration that separates data on a per-user basis may be adopted. The storage device is a storage layer that stores the storage information, and may include, for example, an object storage device 230 that stores the storage data and a database device 220 that stores metadata associated with the storage data. 【0036】 The management unit 212 assigns hierarchical information, importance information, and contextual information to each piece of stored information stored by the memory unit 211, and manages these as attribute information associated with the stored information. The management unit 212 uses the hierarchical information as a hierarchy consisting of the first, second, third, fourth, and fifth levels. The first level represents basic cognition, the second level represents collaboration patterns, the third level represents specialized knowledge, the fourth level represents strategic thinking, and the fifth level represents metacognition. For example, the management unit 212 may assign the first level to stored information including term definitions and preconditions, the second level to stored information including the division of roles between humans and AI assistants, types of instructions, or types of confirmations, the third level to stored information including knowledge and design policies in a specific domain, the fourth level to stored information including selection criteria and evaluation perspectives for multiple policies, and the fifth level to stored information including inspection perspectives for the decision-making process and self-inspection perspectives to reduce errors. Hierarchical information may be structured so that the level of abstraction increases from concrete to abstract. This makes it easier to select more concrete or more abstract memory information depending on the purpose of the work session. 【0037】 The management unit 212 may adopt one or more definition systems for contextual information representing the work status in a work session, depending on the application. For example, the management unit 212 may use a system of contextual information consisting of implementation and development context, vision and strategy context, planning and design context, problem-solving context, documentation context, creativity and innovation context, and research and analysis context. Furthermore, the management unit 212 may use phase information indicating the stage of work progress as information to supplement the contextual information. Phase information may, for example, indicate one of the following: research and planning phase, design phase, material creation and assembly phase, or integration and publication phase. 【0038】 The management unit 212 may acquire information regarding the work session, including the type of work to be performed, the purpose of the work, the content of the input request, the type of referenced material, and phase information. The management unit 212 may use this information to determine contextual information. 【0039】 The management unit 212 may switch the context to prioritize depending on the acquired phase information. For example, if the investigation and planning phase is indicated, the investigation and analysis context may be prioritized; if the design phase is indicated, the planning and design context may be prioritized. If the material creation and assembly phase is indicated, the documentation context or the creation and innovation context may be prioritized; if the integration and publication phase is indicated, the implementation and development context or the problem-solving context may be prioritized. 【0040】 Even when identical phase information is presented, the management unit 212 may further switch the preferred context depending on the words or expressions contained in the work session information. For example, in the integration and publication phase, if the work session information includes words related to implementation, testing, deployment, or publication, the implementation and development context may be preferred. Conversely, in the integration and publication phase, if the work session information includes words related to failure, cause, avoidance, or correction, the problem-solving context may be preferred. 【0041】 The management unit 212 may define contextual information at a different level of granularity than the system described above. For example, the management unit 212 may adopt a system for defining contextual information consisting of project execution context, problem-solving context, learning context, and retrospective context. In this case, the management unit 212 may associate the project execution context with the implementation and development context, vision and strategy context, planning and design context, documentation context, creativity and innovation context, and research and analysis context. The management unit 212 may associate the problem-solving context with the problem-solving context, and the learning context with the research and analysis context. The management unit 212 may define the retrospective context as the work status related to recording or evaluating outcomes in a work session. 【0042】 The management unit 212 may adopt a system consisting of project work (PROJECT) context, debugging / problem solving (DEBUG) context, learning / investigation (LEARN) context, and review / reflection (REVIEW) context as another example of a definition system for contextual information. The management unit 212 may automatically determine the context according to the adopted definition system based on information about the work session (e.g., search query or input statement). 【0043】 The management unit 212 may assign the above-mentioned context information (which may include phase information) to the stored information generated or used in a work session and stored by the storage unit 211. The provision unit 213 may determine the context information that is suitable for the work session according to the adopted context information definition system and use that context information to select the stored information to be provided to the external LLM device. 【0044】 The provisioning unit 213 determines contextual information corresponding to a work session based on information about the work session. The provisioning unit 213 evaluates the relevance of each piece of stored information to the work session based on the determined contextual information and the hierarchical and importance information assigned to each piece of stored information stored by the storage unit 211. Based on the evaluation results, the provisioning unit 213 selects the stored information related to the work session from the stored information stored by the storage unit 211. The provisioning unit 213 provides the selected stored information to the AI ​​assistant. For example, the provisioning unit 213 can calculate a relevance score using the similarity between the search query indicating what the user wants in the work session and the stored information, the degree of agreement between the keywords extracted from the search query and the keywords extracted from the stored information, and the hierarchical and importance information. Based on the calculated score, the provisioning unit 213 can rank the candidate stored information, filter the candidate stored information based on the importance information, and select a predetermined number of stored information from the filtered candidate stored information to provide to the AI ​​assistant. The provisioning unit 213 can select information related to the work session from among multiple pieces of stored information based on the degree of agreement between the context information of the current work session and the context information attached to each piece of stored information stored by the storage unit 211. 【0045】 The update unit 214 updates importance information according to usage statistics based on the usage history of the stored information stored by the memory unit 211. The update unit 214 calculates a weighted sum by multiplying the usage frequency of each stored information, the newness of each stored information, and the last access time of each stored information by a predetermined coefficient, and can update the importance information of each stored information based on the weighted sum. For example, the update unit 214 can update the importance information of stored information that is repeatedly referenced in a work session to a higher value, and update the importance information of stored information that has decreased in reference frequency to a lower value. 【0046】 The database device 220 functions as a database that stores attribute information assigned to each piece of stored information. Attribute information includes, for example, hierarchy information, importance information, and contextual information. The database device 220 may use, for example, a NoSQL database. The database device 220 may store the attribute information for each piece of stored information in association with separate storage areas for each user identifier. 【0047】 The object storage device 230 stores stored information (for example, content data including preconditions, constraints, policies, etc.) in cloud storage. The object storage device 230 may store each user's stored information in a storage area separated for each user identifier. 【0048】 The server device 210 logically identifies the storage area to be referenced for each user based on the user identifier, and manages the storage information and attribute information belonging to that storage area. The server device 210 refers to the attribute information stored in the database device 220, retrieves the storage information identified by that attribute information from the object storage device 230, and uses it for the provisioning unit 213 to evaluate and select relationships and provide them to the external LLM device 130. 【0049】 The provisioning unit 213 determines the storage area to be referenced based on the user identifier indicated by the information regarding the work session. The provisioning unit 213 then performs a relationship evaluation and selection using attribute information for the storage information belonging to the storage area to be referenced. 【0050】 Figure 3 is a diagram illustrating an example of a conceptual configuration that organizes the process of managing and providing stored information, performed by the device configuration shown in Figures 1 and 2, along with the flow of information in a work session. The layers shown in Figure 3 are logical divisions for organizing the stages in which the multiple devices and functions constituting the stored information management system 100 are involved in receiving input information, referencing stored information, selecting candidates, inputting to the external LLM device 130, receiving response messages, and updating importance information in a work session. The AI ​​assistant is implemented, for example, as the external LLM device 130 that generates response messages using a generation AI. Providing the stored information selected by the providing unit 213 to the external LLM device 130 includes providing the stored information selected by the providing unit 213 to the AI ​​assistant. 【0051】 The user layer 310 represents the stage in which a human user inputs a request using a user terminal 110, receives a response message generated by an external LLM device 130, and proceeds with working on the digital artwork. The human user sends a message from the user terminal 110 to the message sharing device 120. The message sharing device 120 stores the message received from the user terminal 110 as information related to the work session and transfers it to the memory management base 140. 【0052】 The memory management interface layer 320 represents the stage in which the memory management infrastructure 140 receives information about the work session and sequentially executes the processing of the storage unit 211, management unit 212, provision unit 213, and update unit 214. The memory management interface layer 320 can accept requests for registration processing, retrieval processing, search processing, and importance information update processing. In the registration process, the memory management interface layer 320 can accept a user identifier, memory information, hierarchy information, importance information, and context information. In the search process, the memory management interface layer 320 can accept a user identifier, search query, context information, hierarchy information, importance information, and the number of candidates. The memory management interface layer 320 can provide APIs for calling each process. As an example of APIs, the memory management interface layer 320 can accept store_memory_3d() for registration processing, retrieve_memory() for retrieval processing, search_memories() for search processing, and update_priority() for importance information update processing. 【0053】 The 3D memory system layer 330 represents a framework in which the management unit 212 assigns attribute information to each piece of memory information stored by the memory unit 211, and manages the attribute information in association with the memory information. In the 3D memory system layer 330, hierarchical information represents the level of abstraction of the memory information, importance information represents the level of importance of the memory information, and contextual information represents the work situation in which the memory information is generated or used. In registration or update processing, the management unit 212 associates the hierarchical information, importance information, and contextual information with the memory information. 【0054】 The automatic classification and optimization layer 340 represents a stage that supports the assignment and updating of attribute information performed by the management unit 212 and the update unit 214. In the automatic classification and optimization layer 340, the management unit 212 can estimate hierarchical information based on the contents of stored information. In the automatic classification and optimization layer 340, the management unit 212 can estimate contextual information based on information about work sessions. In the automatic classification and optimization layer 340, the update unit 214 can update importance information based on usage status. 【0055】 The search and ranking layer 350 represents the stage in which the provisioning unit 213 extracts candidate memory information from memory information stored by the storage unit 211 and selects memory information to provide to the external LLM device 130 from among the candidate memory information. A human user inputs a search query indicating the request content during a work session. The provisioning unit 213 determines contextual information based on information about the work session. The provisioning unit 213 extracts memory information with contextual information that matches or approximates the determined contextual information as candidate memory information. The provisioning unit 213 may filter the candidate memory information based on importance information. The provisioning unit 213 calculates the semantic similarity between the search query and the memory information, the degree of match between the search query and the keywords extracted from the memory information, the importance information attached to the memory information, the hierarchical score based on the hierarchical information attached to the memory information, and the contextual match bonus according to the match or approximation of the contextual information, using the candidate memory information after filtering based on importance information. The provisioning unit 213 integrates each value using weight coefficients adopted by the provisioning unit 213 and calculates an overall score indicating the relevance of the candidate memory information. The supply unit 213 ranks the filtered candidate memory information based on the overall score, selects a predetermined number of memory information from the ranking results, and provides the selected memory information to the external LLM device 130. 【0056】 The provisioning unit 213 can define each value used in calculating the overall score as follows: The provisioning unit 213 can define semantic similarity as a value indicating the closeness between a vector representing the search query and a vector representing the stored information. The provisioning unit 213 can define keyword match as a value indicating the degree of match between keywords extracted from the search query and keywords extracted from the stored information. The provisioning unit 213 can define importance information as a value indicating the utility value of the stored information. The provisioning unit 213 can define hierarchical score as a value calculated based on the level of abstraction indicated by the hierarchical information assigned to the stored information. The provisioning unit 213 can define contextual match bonus as a bonus value given according to the match or approximation between the contextual information determined by the provisioning unit 213 and the contextual information assigned to the candidate stored information. The provisioning unit 213 can normalize the semantic similarity, keyword match, importance information, hierarchical score, and contextual match bonus so that they can be compared, for example, by converting each value to a range of 0 to 1. The supply unit 213 may use the following formula as an example of a formula for calculating the overall score. 【0057】 Overall score = 0.5 × Semantic similarity + 0.3 × Importance information + 0.1 × Hierarchy score + 0.1 × Keyword match + Contextual match bonus 【0058】 The provisioning unit 213 can add a contextual matching bonus when the contextual information matches, reduce the bonus when the contextual information is similar, and reduce the bonus when the contextual information does not match or is not similar. The provisioning unit 213 can set the contextual matching bonus to a value in the range of 0 to 1, for example. 【0059】 The provisioning unit 213 can explain specific examples of the calculation and ranking of the overall score as follows: When a human is creating a design document and inputs words indicating functional constraints as a search query, the provisioning unit 213 determines the planning and design context and extracts memory information with the planning and design context as candidate memory information. The provisioning unit 213 may increase the semantic similarity and keyword match of memory information containing constraints among the candidate memory information, add a contextual match bonus, and increase the overall score. When a human is estimating the cause of a defect and inputs words indicating the cause or correction as a search query, the provisioning unit 213 determines the problem-solving context and extracts memory information with the problem-solving context as candidate memory information. The provisioning unit 213 may increase the semantic similarity and keyword match of memory information related to the cause or correction among the candidate memory information, add a contextual match bonus, and increase the overall score. 【0060】 The persistence and storage layer 360 represents the stage of persisting memory information and attribute information. The object storage device 230 can store memory information as, for example, a file. As an example of a file for storing memory information, the object storage device 230 can use MEMORIES_MAIN.JSON, which stores the body of the memory information. The database device 220 can store hierarchical information, importance information, and context information associated with the memory information as attribute information. As an example of a file for storing attribute information, the database device 220 can use CONTEXT_INDEX.JSON, which stores an index of context information; LAYER_INDEX.JSON, which stores an index of hierarchical information; PRIORITY_INDEX.JSON, which stores an index of importance information; and ID_COORDINATE_INDEX.JSON, which stores an index of management coordinates consisting of hierarchical information, importance information, and context information. The server device 210 can refer to the index stored in the database device 220, retrieve memory information from the object storage device 230 based on the referencing result, and use it for candidate extraction and selection by the provisioning unit 213. 【0061】 The multi-tenant extension layer 370 represents the stage of separate management of storage areas for each user identifier. The storage management infrastructure 140 determines the storage area to be referenced based on the user identifier and processes the storage information and attribute information belonging to the determined storage area. The provisioning unit 213 evaluates the storage information belonging to the determined storage area as candidate storage information and provides the selected storage information to the external LLM device 130 based on the evaluation result. The external LLM device 130 generates a response message based on information about the work session and the provided storage information. The storage management infrastructure 140 transmits the response message received from the external LLM device 130 to the message sharing device 120, and the message sharing device 120 transmits the received response message to the user terminal 110. 【0062】 Figure 4 shows an example of hierarchical information. 【0063】 Figure 4 shows the first tier 410 representing basic cognition, the second tier 420 representing collaboration patterns, the third tier 430 representing specialized knowledge, the fourth tier 440 representing strategic thinking, and the fifth tier 450 representing metacognition. The hierarchical information is attribute information used to distinguish the level of abstraction of memory information. 【0064】 The level of abstraction is a measure that represents the degree to which the content of the memory information is specific or general. The subject represents the subject area handled by the memory information belonging to each hierarchical level. The upper limit of the number of memory items represents the upper limit of the number of memory items stored by the memory unit 211 for each hierarchical level. The cognitive load represents the weight assigned to each hierarchical level in the configuration of memory information provided to the external LLM device 130. The weight may be expressed as a percentage. The weight is a value that represents the relative degree assigned to each hierarchical level, and the sum of the weights does not necessarily have to be 100. The provisioning unit 213 can normalize the weights based on the sum of the weights and use the normalized values ​​to distribute the number of memory items selected from each hierarchical level. The provisioning unit 213 can determine the distribution of the number of memory items selected from each hierarchical level based on the weights used by the provisioning unit 213. Keywords represent examples of words that can be extracted from the content of the memory information belonging to each hierarchical level. The provisioning unit 213 can refer to the level of abstraction and weights and adjust the combination and number of hierarchical levels of memory information provided to the external LLM device 130 according to the work session. 【0065】 The first level, 410, represents foundational cognition. It is defined as the level with the most concrete level of abstraction. The objects of the first level, 410, are basic collaborative principles and core cognitive patterns. The upper limit for the number of items that can be memorized in the first level, 410, is, for example, 7. The cognitive load of the first level, 410, is, for example, 40%. Keywords for the first level, 410, include, for example, foundation, core, basic, principle, collaboration, and basis. 【0066】 The second level, 420, is a level that represents collaboration patterns. The second level, 420, is defined as a level with a higher level of abstraction than the first level, 410. The subjects of the second level, 420, are patterns of team collaboration, workflow patterns, and processes. The upper limit of the number of items that can be memorized in the second level, 420, is, for example, 12. The cognitive load of the second level, 420, is, for example, 25%. Keywords for the second level, 420, are, for example, collaboration, team, workflow, coordination, and process. 【0067】 The third level, 430, represents expertise. It is defined as a higher level of abstraction than the second level, 420. The subject matter of the third level, 430, includes technical knowledge, implementation insights, and tool usage. The maximum number of items that can be memorized in the third level, 430, is, for example, 20. The cognitive load of the third level, 430, is, for example, 35%. Keywords for the third level, 430, include, for example, technology, implementation, program (code), tool, and programming language names (programming languages). 【0068】 The fourth level, 440, represents strategic thinking. It is defined as a higher level of abstraction than the third level, 430. The subjects of the fourth level, 440, are project strategy, design policies, and structural policies. The maximum number of items that can be memorized in the fourth level, 440, is, for example, 15. The cognitive load of the fourth level, 440, is, for example, 20%. Keywords for the fourth level, 440, include, for example, strategy, plan, design, structure (architecture), and policy. 【0069】 The fifth level, 450, represents metacognition. It is defined as the level with the highest degree of abstraction. The subjects of the fifth level, 450, are the optimization of cognitive processes, learning patterns, and improvement strategies. The upper limit of memory capacity for the fifth level, 450, is, for example, 10 items. The cognitive load for the fifth level, 450, is, for example, 15%. Keywords for the fifth level, 450, include, for example, creativity, innovation, improvement, meta, learning, and optimization. 【0070】 The provisioning unit 213 determines contextual information based on information about the work session and evaluates the relationships between memory information based on contextual information, hierarchical information, and importance information. For example, if an implementation and development context is determined, memory information of specialized knowledge belonging to the third hierarchical level 430 can be treated as a priority candidate, and memory information of strategic thinking belonging to the fourth hierarchical level 440 can be treated as a secondary candidate. 【0071】 The providing unit 213 may adjust the weighting of the selection of hierarchical information based on the user's role information included in the work session information. The providing unit 213 may relatively increase the number of selected memory information belonging to the first hierarchical level 410 and the second hierarchical level 420 if the user has little development experience. The providing unit 213 may relatively increase the number of selected memory information belonging to the fourth hierarchical level 440 and the fifth hierarchical level 450 if the user is responsible for decision-making. 【0072】 Figure 5 shows an example of contextual information. 【0073】 Figure 5 shows the implementation and development context 510, the vision and strategy context 520, the planning and design context 530, the problem-solving context 540 (problem-solving and analysis context), the documentation context 550, the creativity and innovation context 560, and the research and analysis context 570. Contextual information is attribute information that indicates the working conditions under which memory information is generated or used. 【0074】 The target is the subject area handled by the memory information belonging to each context. Keywords (example keywords) are examples of words that can be extracted from the content of the memory information belonging to each context. Usage scenarios are examples of work situations in which each context is likely to occur during a work session. Priority is an example of a scale that the providing unit 213 refers to when selecting memory information. The providing unit 213 can use priority to adjust the weighting of the selection of candidate memory information for each context. 【0075】 The implementation and development context 510 is a context that describes the work status related to code implementation, task execution, and specific tasks. The subjects of the implementation and development context 510 are, for example, code implementation, task execution, specific tasks, and implementation execution. Keywords for the implementation and development context 510 are, for example, implementation, code, execution, work, task, and program. The usage scenarios for the implementation and development context 510 are, for example, during development work, during implementation task execution, and coding. 【0076】 Vision and Strategy Context 520 is a context that describes the work status related to long-term vision, strategy formulation, goal setting, and direction determination. The subject of Vision and Strategy Context 520 is, for example, long-term vision, strategy formulation, goal setting, and direction determination. Keywords for Vision and Strategy Context 520 include, for example, vision, strategy, goals, future, and direction. Situations in which Vision and Strategy Context 520 is used include, for example, the initiation of new projects, direction considerations, and strategy formulation. 【0077】 The planning and design context 530 is a context that describes the work status related to planning, task design, and scheduling. The subjects of the planning and design context 530 include, for example, specific planning, cycle planning, task design, and scheduling. Keywords for the planning and design context 530 include, for example, planning, design, cycle, task, and schedule. The usage scenarios for the planning and design context 530 include, for example, the planning phase and design work. 【0078】 Problem-solving context 540 is a context that describes the work status related to error resolution, problem analysis, and investigation. The subjects of problem-solving context 540 include, for example, error resolution, debugging, problem analysis, and investigation. Keywords for problem-solving context 540 include, for example, error, problem, analysis, debugging, and investigation. Problem-solving context 540 is used in situations such as troubleshooting, error handling, and problem resolution. 【0079】 Document creation context 550 is a context that describes the work status related to documentation and record-keeping. The subjects of document creation context 550 include, for example, document creation, completion report and guide creation, and documentation. Keywords for document creation context 550 include, for example, document, report, guide, document, and record. Examples of situations in which document creation context 550 is used include during document creation, report creation, and documentation work. 【0080】 The Creative and Innovative Context 560 is a context that describes the working state related to new ideas and creative thinking. The subjects of the Creative and Innovative Context 560 include, for example, new ideas, innovative approaches, and creative thinking. Keywords for the Creative and Innovative Context 560 include, for example, creativity, innovation, ideas, new, and experimentation. Examples of situations in which the Creative and Innovative Context 560 is used include brainstorming, new feature development, and improvement proposals. 【0081】 The Investigation and Analysis Context 570 is a context that describes the work status related to information gathering, examination, and technical evaluation. The subjects of Investigation and Analysis Context 570 are, for example, investigation, analysis, examination, information gathering, and technical evaluation. The keywords of Investigation and Analysis Context 570 are, for example, investigation, analysis, examination, information gathering, and technical evaluation. The situations in which Investigation and Analysis Context 570 is used are, for example, during investigation work, analysis tasks, technical evaluation, and information gathering. 【0082】 The provisioning unit 213 determines contextual information based on information about the work session and evaluates the relationship of memory information based on the determined contextual information, hierarchical information, and importance information. For example, if an implementation and development context 510 is determined, memory information including expertise related to implementation can be treated as a priority candidate. Also, for example, if a vision and strategy context 520 is determined, memory information related to strategic thinking and memory information related to metacognition can be treated as priority candidates. 【0083】 Figure 6 shows an example of importance information. 【0084】 Figure 6 shows importance categories 610, 620, 630, and 640. Importance information is attribute information indicating the importance of the stored information. 【0085】 Importance represents the degree of priority to which the memory information is referenced in a work session. The scope is the range of content handled by the memory information belonging to each importance category. Automatic calculation refers to the process of calculating importance information based on usage statistics based on the memory information's usage history or the detection results of words contained in the memory information's content. Keywords are examples of words that can be extracted from the content of the memory information belonging to each importance category. The update unit 214 can calculate a weighted sum by multiplying the frequency of use, recency, and last access time by predetermined coefficients, and update the importance information based on the calculated weighted sum. 【0086】 The importance level 610 is the highest level of importance. The importance level 610 ranges from 0.8 to 1.0. This level applies to urgent tasks, critical decisions, and memory information requiring immediate attention. Automatic calculation of the importance level 610 can be based on the detection of words indicating urgency, recent access, and high frequency of use. Keywords for the importance level 610 include, for example, CRITICAL, urgency, immediate, and important. 【0087】 Importance level 620 is the next highest level of importance after importance level 610. Importance level 620 corresponds to an importance information value in the range of 0.6 to 0.79. Importance level 620 applies to memory information that is important but not urgent and is referenced regularly. Automatic calculation of importance level 620 can be based on regular access and moderate usage frequency. Keywords for importance level 620 include, for example, important, necessary, and reference. 【0088】 Importance level 630 is a level of moderate importance. Importance level 630 includes items with an importance value between 0.4 and 0.59. Importance level 630 applies to reference information, supplementary materials, and infrequently referenced memory information. Automatic calculation of importance level 630 can be performed based on infrequent access and usage statistics indicating that the information is old. Keywords for importance level 630 include, for example, reference, memo, and supplement. 【0089】 Importance level 640 represents the lowest level of importance. This level includes importance information values ​​between 0.0 and 0.39. Importance level 640 applies to stored information intended for archiving, inactive stored information, and stored information related to historical records. Automatic calculation of importance level 640 can be based on long-term inaccessibility and usage statistics indicating completed tasks. Keywords for importance level 640 include, for example, archive, completed, and preserved. 【0090】 The provisioning unit 213 determines contextual information based on information about the work session and evaluates the relationships between stored information based on contextual information, hierarchical information, and importance information. By filtering based on importance information, the provisioning unit 213 can, for example, treat stored information belonging to importance categories 610 and 620 as priority candidates and stored information belonging to importance categories 630 and 640 as auxiliary candidates. 【0091】 Figure 7 shows an example of the registration process for stored information in the memory management system 100. 【0092】 Through the process shown in Figure 7, the memory management infrastructure 140 generates memory information from information that can be stored, assigns hierarchical information, importance information, and contextual information as attribute information to the memory information, and stores the assigned attribute information in association with the memory information using the memory unit 211. 【0093】 The memory management infrastructure 140 may store the contents of the memory information stored by the memory unit 211 as text data. As an example of text data, the memory management infrastructure 140 may store structured text that follows a markup notation capable of representing heading structures and bulleted list structures. The memory management infrastructure 140 may express the level of abstraction, basis for judgment, and constraints of the memory information using structured text. 【0094】 In step S701, the memory management infrastructure 140 acquires information that can be used as memory candidates. For example, the memory management infrastructure 140 acquires instruction sentences entered by a human during a work session, response sentences generated by the AI ​​assistant, and constraints used during the work session as information that can be used as memory candidates. For example, if preconditions are clarified during a work session, the memory management infrastructure 140 acquires a sentence containing those preconditions as information that can be used as memory candidates. 【0095】 The memory management infrastructure 140 may acquire, as an example of information that can be stored, a document containing configuration procedures related to communication control of a web application. If the configuration procedure is a configuration procedure related to cross-origin resource sharing, the memory management infrastructure 140 may, for example, perform the processing described below to assign hierarchical information of the second layer 420, importance information of importance category 610 or importance category 620, and contextual information of implementation and development context 510 to the stored information representing the configuration procedure. If the configuration procedure is intended for sharing work procedures, the memory management infrastructure 140 may, for example, perform the processing described below to assign hierarchical information of the second layer 420, importance information of importance category 630, and contextual information of document creation context 550 to the stored information representing the configuration procedure. 【0096】 In step S702, the memory management infrastructure 140 analyzes the memory contents. The memory management infrastructure 140 analyzes the subject, terminology, and work situation contained in the information that is a candidate for memory, and extracts the parts to be stored as memory information. For example, if the information that is a candidate for memory contains multiple subjects, the memory management infrastructure 140 can divide the text according to the subject and treat the divided text as memory information. 【0097】 In step S703, the memory management infrastructure 140 assigns hierarchical information to the stored information. The memory management infrastructure 140 uses one of the first to fifth hierarchical levels as the hierarchical information, and selects the hierarchical information according to the level of abstraction of the stored information's content. For example, the memory management infrastructure 140 may assign the first level to stored information containing term definitions or preconditions, the second level to stored information containing the division of roles or instruction types between humans and AI assistants, and the third level to stored information containing specialized knowledge or implementation insights. For example, the memory management infrastructure 140 may assign the fourth level to stored information containing selection criteria or evaluation perspectives for multiple options, and the fifth level to stored information containing inspection perspectives for the decision-making process or self-inspection perspectives to reduce errors. The process of assigning hierarchical information will be described later using Figure 9. 【0098】 In step S704, the memory management infrastructure 140 determines the contextual information corresponding to the work session and uses that contextual information to add contextual information to the memory information. The memory management infrastructure 140 may use one of the following contextual contexts: implementation and development context, vision and strategy context, planning and design context, problem solving context, document creation context, creativity and innovation context, or research and analysis context. For example, the memory management infrastructure 140 may use the implementation and development context when the content of the work session relates to implementation work, and use the problem solving context when the content of the work session relates to investigating the cause of a defect. The process of determining contextual information will be described later with reference to Figure 10. 【0099】 In step S705, the memory management infrastructure 140 calculates importance information. The memory management infrastructure 140 assigns importance information as a value indicating the importance of the memory information. For example, the memory management infrastructure 140 can set the importance information high when the information to be stored contains words indicating urgency, and set the importance information low when it is positioned as reference material. The memory management infrastructure 140 can update the importance information according to usage statistics based on the usage history of the memory information. The calculation process of importance information will be described later using Figure 11. 【0100】 In step S706, the memory management infrastructure 140 determines a three-dimensional coordinate. Based on the assigned hierarchical information, importance information, and context information, the memory management infrastructure 140 determines a three-dimensional coordinate for identifying the memory information. The three-dimensional coordinate is expressed, for example, as "(Layer, Priority, Context)", which is a combination of hierarchical information, importance information, and context information. For example, for memory information that has been assigned hierarchical information corresponding to the third hierarchical level, importance information corresponding to importance classification 620, and context information corresponding to the implementation and development context, the memory management infrastructure 140 can determine a three-dimensional coordinate corresponding to "(Layer 3, Priority 0.6, Context: Implementation)". 【0101】 In step S707, the memory management infrastructure 140 stores memory information and attribute information. The memory management infrastructure 140 can store memory information in association with memory areas separated for each user identifier. For example, the memory management infrastructure 140 stores hierarchical information, importance information, and contextual information in the database device 220. For example, the memory management infrastructure 140 stores memory information in the object storage device 230. 【0102】 Figure 8 shows an example of the process (search process) in which the memory management platform 140 selects and transmits memory information to be provided to the AI ​​assistant. 【0103】 Through the process shown in Figure 8, the memory management infrastructure 140 can determine contextual information based on information about the work session, acquire candidate memory information from the 3D coordinate space, perform filtering and ranking based on importance information and selection based on cognitive load management, configure and transmit input information to be provided to the AI ​​assistant, and update importance information based on usage statistics. 【0104】 In step S801, the memory management infrastructure 140 determines context information based on the work content or user input. The memory management infrastructure 140 obtains a string from the work session information and performs preprocessing on the obtained string. The memory management infrastructure 140 may perform word segmentation by morphological analysis, removal of unnecessary words, and normalization of variations in spelling as preprocessing. The memory management infrastructure 140 extracts keywords from the preprocessed string and calculates the degree of match between the extracted keywords and the keyword groups associated with each context. The memory management infrastructure 140 determines the context with the highest degree of match as the context information corresponding to the work session. 【0105】 The memory management infrastructure 140 determines the implementation and development context 510 if, for example, the extracted keywords include "implementation," "code," or "task," and the degree of matching is above a predetermined threshold. The memory management infrastructure 140 also determines the vision and strategy context 520 if, for example, the extracted keywords include "vision," "strategy," or "goal," and the degree of matching is above a predetermined threshold. 【0106】 The memory management infrastructure 140 may, as a method for determining contextual information, adopt a similarity-based method using document embedding instead of a keyword matching-based method. The memory management infrastructure 140 may, as a method for determining contextual information, adopt an estimation method using a classifier instead of a rule-based method. 【0107】 In step S802, the memory management infrastructure 140 retrieves memory information related to contextual information from the three-dimensional coordinate space. For example, the memory management infrastructure 140 can extract memory information that has contextual information that matches or approximates the determined contextual information as candidate memory information. The three-dimensional coordinate space is represented as a space that identifies memory information by a combination of hierarchical information, importance information, and contextual information, namely "(Layer, Priority, Context)". For example, the memory management infrastructure 140 can retrieve candidate memory information related to the implementation and development context 510 from a region containing three-dimensional coordinates corresponding to "(Layer 3, Priority 0.6, Context: Implementation)". 【0108】 In step S803, the memory management infrastructure 140 filters candidate memory information based on importance information. The memory management infrastructure 140 obtains the importance information value for each candidate memory information, extracts candidate memory information whose importance information value is equal to or greater than a predetermined threshold, and excludes candidate memory information whose importance information value is less than a predetermined threshold. For example, the memory management infrastructure 140 extracts candidate memory information that satisfies the importance information values ​​corresponding to importance categories 610 and 620, and excludes candidate memory information that does not satisfy the importance information values ​​corresponding to importance categories 630 and 640. 【0109】 The memory management infrastructure 140 calculates an overall score indicating relevance for each of the filtered candidate memory information. The memory management infrastructure 140 calculates the similarity between the search query and the candidate memory information, calculates the degree of match between the keywords extracted from the search query and the keywords extracted from the candidate memory information, and obtains the values ​​of hierarchical information and importance information. The memory management infrastructure 140 may calculate bonus points based on the match or approximation between the contextual information determined based on information about the work session and the contextual information attached to the candidate memory information, and add these to the overall score. The memory management infrastructure 140 takes the similarity, match rate, hierarchical information, importance information, and bonus points as input and calculates an overall score based on predetermined weighting. 【0110】 The memory management infrastructure 140 can weight and integrate similarity, agreement, hierarchical information, and importance information when calculating the overall score. The memory management infrastructure 140 may calculate the overall score as a weighted sum of a weight coefficient of 0.5 for similarity, a weight coefficient of 0.3 for importance information, a weight coefficient of 0.1 for hierarchical information, and a weight coefficient of 0.1 for agreement. The memory management infrastructure 140 may add bonus points to the weighted sum. The memory management infrastructure 140 may use fixed values ​​for the weight coefficients, or it may be possible to change them according to the operating conditions. 【0111】 The memory management infrastructure 140 may define evaluation values ​​for hierarchical information as hierarchical importance. As an example of hierarchical importance, the memory management infrastructure 140 may define evaluation values ​​of 1.0 for the 5th hierarchical level, 0.8 for the 4th hierarchical level, 0.6 for the 3rd hierarchical level, 0.4 for the 2nd hierarchical level, and 0.2 for the 1st hierarchical level. Based on the definition of hierarchical importance, the memory management infrastructure 140 may calculate the contribution of hierarchical information to the overall score. 【0112】 The memory management infrastructure 140 may employ a method for calculating the overall score based on a weighted sum of similarity, agreement, hierarchical information, and importance information. The memory management infrastructure 140 may also employ a method for estimating ranks using a trained ranking model instead of a rule-based sorting method for calculating the overall score. The memory management infrastructure 140 may use the ranks estimated by the ranking model to sort the candidate memory information. 【0113】 In step S804, the memory management infrastructure 140 ranks the filtered candidate memory information by sorting it in descending order of overall score. The memory management infrastructure 140 determines the number of memory information items to provide to the external LLM device 130 according to the information about the work session. The memory management infrastructure 140 may change the selection criteria based on hierarchical information according to the information about the work session. For example, if the implementation and development context 510 is determined, the memory management infrastructure 140 may relatively increase the contribution of hierarchical importance so that candidate memory information belonging to the third hierarchical level 430 is more likely to be ranked higher. For example, if the implementation and development context 510 is determined, the memory management infrastructure 140 may relatively decrease the contribution of hierarchical importance so that candidate memory information belonging to the fifth hierarchical level 450 is less likely to be ranked higher. 【0114】 The memory management infrastructure 140 controls the number of memory information items to be provided to the external LLM device 130 as a measure of item count control. As an example of item count control, the memory management infrastructure 140 may control the number of memory information items to be provided to a range of 7 ± 2. The memory management infrastructure 140 may determine the number of memory information items to be provided to be the top 7 items. Depending on the information regarding the work session, the memory management infrastructure 140 may determine the number of memory information items to be provided to be the top 5 items or the top 9 items. 【0115】 The memory management infrastructure 140 selects top candidate memory information corresponding to the determined number based on the ranking results. The memory management infrastructure 140 provides the selected candidate memory information to the external LLM device 130. In the above selection, the memory management infrastructure 140 may refer to the upper limit of the number of items associated with the hierarchical information. In the above selection, the memory management infrastructure 140 may, in addition to the upper limit of the number of items associated with the hierarchical information, use the number of items distribution associated with the hierarchical information, set selection frames for each hierarchical level, and adopt a method of selecting top candidate memory information within the range of the selection frame for each hierarchical level. 【0116】 In step S805, the memory management infrastructure 140 generates input information to be sent to the external LLM device 130 (generation AI) and sends the generated input information to the external LLM device 130. The memory management infrastructure 140 acquires information about the work session and extracts the request content from the acquired work session information. The memory management infrastructure 140 acquires the memory information selected in step S804 and includes the acquired memory information in the input information in association with the request content. The memory management infrastructure 140 generates a text column containing the request content and memory information and formats the generated text column as input information. 【0117】 The memory management infrastructure 140 sets the number of memory information items to be included in the input information to be the same as the number determined in step S804. The memory management infrastructure 140 includes the above number of memory information items selected in step S804 in the input information. 【0118】 The memory management infrastructure 140 may place the request content first in the input information, the above number of stored items after the request content, and the output format instructions after the stored items. The memory management infrastructure 140 may place information regarding the work session history and attached files between the request content and the stored items, or between the stored items and the output format instructions. The memory management infrastructure 140 may arrange the stored items in the input information in order of importance, in order of hierarchical information, or in order of the degree of relevance of contextual information. 【0119】 The memory management infrastructure 140 transmits the generated input information to the external LLM device 130. The external LLM device 130 generates a response message based on the received input information. The memory management infrastructure 140 assists the external LLM device 130 in generating the response message by including memory information related to the request in the input information. 【0120】 The memory management infrastructure 140 may, as a method for structuring input information, adopt a method of generating request content and stored information as structured text separated into items, instead of simply concatenating the request content and stored information. The memory management infrastructure 140 may, as a method for structuring input information, adopt a method of including information regarding work session history and attached files as items, in addition to the request content and stored information. 【0121】 In step S806, the memory management infrastructure 140 updates the importance information assigned to the memory information. The memory management infrastructure 140 identifies the memory information provided to the external LLM device 130 in step S805 and records the usage history for each identified memory information. The memory management infrastructure 140 updates the usage statistics based on the usage history. The usage statistics include the frequency of use, which indicates the number of times the memory information has been provided; the recency, which indicates the time elapsed since the memory information was created or updated; and the last access time, which indicates the time when the memory information was last provided. 【0122】 The memory management infrastructure 140 multiplies the frequency of use, recency, and last access time by predetermined coefficients and calculates a weighted sum of the multiplication results. The memory management infrastructure 140 converts the weighted sum into importance information values ​​and stores the converted values ​​as importance information associated with the memory information. For example, the memory management infrastructure 140 sets the coefficients so that the weighted sum increases for memory information with a high frequency of use, thereby increasing the importance information of memory information that has been repeatedly referenced in a work session. For example, the memory management infrastructure 140 sets the coefficients so that the weighted sum decreases for memory information with an old last access time, thereby decreasing the importance information of memory information that has not been referenced for a long period of time. 【0123】 The memory management infrastructure 140 may, instead of using a weighted sum-based method for updating importance information, employ a method that decays importance information at regular intervals. The memory management infrastructure 140 may, instead of using a usage statistics-based decision method for updating importance information, employ a rule-based, stepwise classification change method. 【0124】 Figure 9 shows an example of the hierarchical information classification process in step S703. 【0125】 Based on the processing shown in Figure 9, the memory management infrastructure 140 selects hierarchical information corresponding to one of the first hierarchical levels 410 to the fifth hierarchical level 450, and outputs the selected hierarchical information. 【0126】 In step S901, the memory management infrastructure 140 generates text to be used for hierarchical information determination and inputs the generated text as content. The memory management infrastructure 140 accepts the candidate information for memory obtained in step S701 shown in Figure 7 as input data and extracts candidate sentences from the input data. The memory management infrastructure 140 performs a cutting process on the candidate sentences based on length, delimiter, or line break position, and confirms the cut-out sentences as the text to be determined. The memory management infrastructure 140 passes the confirmed text to be determined as content to the subsequent processing. For example, the memory management infrastructure 140 confirms one of the following as the text to be determined: a sentence representing the preconditions determined in the work session, a sentence representing the constraints determined in the work session, or a sentence representing the work policy in the work session, and inputs it as content. 【0127】 The memory management infrastructure 140 may, as a method for generating text to be judged, adopt a method that generates text to be judged by summarizing information that can be used as memory candidates, instead of a text extraction process. The memory management infrastructure 140 may, as a method for generating text to be judged, adopt a method that generates text to be judged based on sentence boundary estimation, instead of a rule-based extraction process. 【0128】 In step S902, the memory management infrastructure 140 preprocesses the text input as content and generates a keyword set based on the preprocessing results. As preprocessing, the memory management infrastructure 140 sequentially performs character type unification, normalization of variations in notation, removal of unnecessary words, and word segmentation. For example, the memory management infrastructure 140 converts synonymous expressions into identical expressions using predetermined regular expressions or dictionaries, removes unnecessary words such as particles, and generates word sequences based on word boundaries. From the word sequences, the memory management infrastructure 140 extracts nouns and verbs based on part-of-speech information and registers the extracted words as keywords in the keyword set. For example, the memory management infrastructure 140 extracts technical terms or design terms contained in the input text as keywords and registers them in the keyword set. 【0129】 The memory management infrastructure 140 may, as a method for generating keywords, adopt a method of selecting top words based on frequency of occurrence instead of an extraction method based on part-of-speech information. The memory management infrastructure 140 may, as a method for generating keywords, adopt a method of extracting important words based on partial matching of strings instead of a method based on word segmentation. 【0130】 In step S903, the memory management infrastructure 140 calculates a matching score with the hierarchical definition keywords. The memory management infrastructure 140 refers to the keyword groups associated with each hierarchical level from the first level 410 to the fifth level 450 and calculates the degree of agreement between the extracted keywords and the keyword groups. Based on the degree of agreement, the memory management infrastructure 140 calculates a matching score for each hierarchical level. For example, if the extracted keywords include "collaboration" and "cooperation," the memory management infrastructure 140 increases the matching score corresponding to the second level 420. For example, if the extracted keywords include "strategy" and "design," the memory management infrastructure 140 increases the matching score corresponding to the fourth level 440. 【0131】 In step S904, the memory management infrastructure 140 selects the hierarchy with the highest score. The memory management infrastructure 140 compares the matching scores of each hierarchy and selects the hierarchy showing the maximum value as the hierarchy information. If multiple hierarchies show the same maximum value, the memory management infrastructure 140 may determine the hierarchy using priority based on importance information or contextual information. 【0132】 In step S905, the memory management infrastructure 140 outputs a hierarchy number corresponding to the selected hierarchy. For example, if the third hierarchy 430 is selected, the memory management infrastructure 140 outputs "3" as the hierarchy number. The memory management infrastructure 140 uses the outputted hierarchy number as hierarchy information to be assigned to the memory information. 【0133】 The memory management infrastructure 140 may, as a method for determining hierarchical information, adopt a method that estimates the hierarchy using a classifier based on text features, instead of a method based on keyword matching. The memory management infrastructure 140 may, as a method for determining hierarchical information, adopt a method that estimates the hierarchy based on text similarity, instead of a method based on keyword matching. 【0134】 Figure 10 shows an example of the detailed processing of the context information determination process in steps S704 and S801. 【0135】 As shown in Figure 10, the memory management infrastructure 140 inputs content (text) and work session history, calculates a confidence score for contextual information based on the input information, selects the contextual information with the highest confidence score, and outputs the type of the selected contextual information. 【0136】 In step S1001, the memory management infrastructure 140 inputs content (text) and work session history. For example, the memory management infrastructure 140 generates a text to be judged from the candidate information for memory acquired in step S701 shown in Figure 7, and accepts the generated text to be judged as content. The memory management infrastructure 140 accepts information as work session history, which holds user input and AI assistant responses in chronological order during a work session. The memory management infrastructure 140 inputs a predetermined number of most recent message sequences as work session history. 【0137】 In step S1002, the memory management infrastructure 140 analyzes keywords within the content. The memory management infrastructure 140 performs character type unification, normalization of spelling variations, removal of unnecessary words, and word segmentation on the content to generate a preprocessed word sequence. The memory management infrastructure 140 extracts nouns and verbs from the preprocessed word sequence and registers the extracted words as keywords in the keyword set. For example, if the content contains "implementation," "code," or "task," the memory management infrastructure 140 increases the candidate score corresponding to the implementation and development context 510. For example, if the content contains "vision," "strategy," or "goal," the memory management infrastructure 140 increases the candidate score corresponding to the vision and strategy context 520. 【0138】 In step S1003, the memory management infrastructure 140 analyzes the work session history. The memory management infrastructure 140 extracts the most recent topic words, request expressions, and response expressions from the work session history. The memory management infrastructure 140 calculates the frequency of occurrence and the most recent occurrence position of the words included in the work session history and generates the calculation results as history features. For example, if "error," "analysis," or "investigation" repeatedly appears in the work session history, the memory management infrastructure 140 increases the history score corresponding to the problem-solving context 540. For example, if "document," "report," or "record" repeatedly appears in the work session history, the memory management infrastructure 140 increases the history score corresponding to the document creation context 550. 【0139】 In step S1004, the memory management infrastructure 140 calculates a confidence score for contextual information. The memory management infrastructure 140 integrates the candidate scores obtained in step S1002 and the history scores obtained in step S1003 for each context type and calculates the integrated result as a confidence score. For example, the memory management infrastructure 140 calculates a weighted sum of a score based on the degree of matching of keyword sets and a score based on history features, and registers the calculated value as the confidence score. For example, the memory management infrastructure 140 increases the weight of the history score when the content is short, and increases the weight of the candidate score when the work session history is short. 【0140】 In step S1005, the memory management infrastructure 140 selects the context information with the highest confidence level. The memory management infrastructure 140 compares the confidence scores for each context from the implementation and development context 510 to the investigation and analysis context 570 and selects the context showing the maximum value. If multiple contexts show the same maximum value, the memory management infrastructure 140 may determine the priority based on the phase information of the work session or the recency of user input and confirm the selected context. 【0141】 The memory management infrastructure 140 may determine the selected context using a default context if multiple contexts show the same maximum value and the selected context cannot be determined even by prioritizing based on the up-to-dateness of phase information and user input. The memory management infrastructure 140 may use the implementation and development contexts as examples of default contexts. When the memory management infrastructure 140 determines the selected context as the default context, it may record the reason for the determination in association with a confidence score. 【0142】 In step S1006, the memory management infrastructure 140 outputs the type of contextual information selected. The memory management infrastructure 140 uses the outputted type of contextual information as the result of the contextual information assignment process in step S704 shown in Figure 7 and the contextual information determination result in step S801 shown in Figure 8. 【0143】 The memory management infrastructure 140 may use a similarity method based on text embeddings instead of a weighted sum of keyword match and historical features as a method for calculating the confidence score. The memory management infrastructure 140 may use a method that estimates the context type using a classifier instead of a rule-based method as a method for calculating the confidence score. 【0144】 Figure 11 shows an example of the process for calculating importance information in step S705 and step S806 shown in Figure 8. 【0145】 Figure 11 shows an example of the procedure for calculating importance information to be assigned to stored information. In Figure 11, the storage management infrastructure 140 calculates a standard importance for each piece of stored information, adjusts it by time decay according to the elapsed time since the last reference time, corrects it by frequency of use according to the number of references, calculates importance information that reflects the adjustments, corrections, and correction terms, and normalizes the calculated importance information. 【0146】 In step S1101, the memory management infrastructure 140 inputs metadata to be used to calculate importance information. The memory management infrastructure 140 accepts keywords, reference counts, last access time, and creation time or update time as metadata. The memory management infrastructure 140 accepts words extracted based on words contained in the memory information as keywords. The memory management infrastructure 140 may extract keywords from the memory information in step S1101. The memory management infrastructure 140 may extract keywords from the memory information when the importance information calculation process is called. The memory management infrastructure 140 may accept statistical values ​​of usage history in the work session as metadata. 【0147】 In step S1102, the memory management infrastructure 140 sets a standard importance level based on keywords. The standard importance level is a base value calculated from the content of the stored information and the features of the usage history. The memory management infrastructure 140 analyzes the keywords included in the metadata, calculates the degree of agreement with the keyword groups associated with importance levels 610 to 640, and sets the standard importance level based on the degree of agreement. For example, the memory management infrastructure 140 sets a high standard importance level if the keyword contains "urgent," "immediate," or "important," and a low standard importance level if the keyword contains "reference" or "supplementary." 【0148】 As shown in the example calculation method in Figure 11, the memory management infrastructure 140 may calculate the baseline importance by weighting multiple indicators. For example, the memory management infrastructure 140 calculates the baseline importance as: Baseline Priority = wA × Frequency of Use Index + wF × Newness Index + wR × Recentness Index. The frequency of use index is an indicator that represents the number or frequency of times the memory information has been referenced or provided. The newness index is an indicator that represents the time elapsed since the creation time or update time of the memory information. The recentness index is an indicator that represents the time elapsed since the last access time and is an indicator that represents the contribution of short-term freshness in the calculation of baseline importance. On the other hand, the time decay coefficient, which will be described later, is applied after the calculation of baseline importance and is a coefficient that represents forgetting (obsolescence) due to long-term non-reference. The memory management infrastructure 140 may use wA=0.4, wF=0.3, and wR=0.3 as examples of coefficients. 【0149】 In step S1103, the memory management infrastructure 140 makes adjustments based on the last access time. The memory management infrastructure 140 calculates the elapsed time [h] from the difference between the current time and the last access time, and calculates a time decay coefficient based on the elapsed time [h]. The time decay coefficient is a coefficient that decays importance information based on the last access time. While the recency index included in the standard importance mainly represents short-term freshness, it is used to reflect the decrease in importance due to long-term non-referral. For example, the memory management infrastructure 140 calculates the time decay coefficient as time decay coefficient = max{exp(-λ × elapsed time [h]), lower limit}. The memory management infrastructure 140 may also use λ = 0.01 and lower limit = 0.5 as examples of coefficients. 【0150】 In step S1104, the memory management infrastructure 140 makes adjustments based on the frequency of use. The memory management infrastructure 140 calculates a frequency of use correction value based on the number of references. The frequency of use correction value is a correction term that increases the importance information based on the number of references. For example, the memory management infrastructure 140 calculates the frequency of use correction value as frequency of use correction value = min(number of references × 0.01, upper limit). The memory management infrastructure 140 may use upper limit = 0.15 as an example of the upper limit. 【0151】 In step S1105, the memory management infrastructure 140 calculates the importance information value and normalizes it to a range of 0.0 to 1.0. The memory management infrastructure 140 applies a time decay coefficient to the reference importance and calculates the importance information value by reflecting the frequency of use correction value and the urgency correction value. Here, the residency index included in the reference importance represents short-term freshness, and the time decay coefficient represents decay due to long-term non-reference, so both contribute to importance on different time scales. The urgency correction value is a correction term that increases the importance information based on the detection result of words indicating urgency. For example, the memory management infrastructure 140 calculates the importance information value as Priority value = Reference Priority × Time decay coefficient + Frequency of use correction value + Urgency correction value. The memory management infrastructure 140 normalizes the calculated Priority value by clipping or linear transformation so that it falls within the range of 0.0 to 1.0. 【0152】 In step S1106, the memory management infrastructure 140 outputs the normalized importance information value. The memory management infrastructure 140 uses the outputted importance information value to assign importance information in step S705 shown in Figure 7, and to update importance information in step S806 shown in Figure 8. 【0153】 The formulas, correction terms, and calculation order shown in Figure 11 are examples of calculation methods. The memory management infrastructure 140 may change the coefficients, correction terms, and calculation order. The memory management infrastructure 140 may use a method based on the features of the work session history instead of a method based on keyword detection as a method for calculating the urgency correction value. The memory management infrastructure 140 may use a transformation based on the sigmoid function instead of clipping as a method for normalization. 【0154】 According to this embodiment, even if the amount of stored information increases, it is possible to stably select and present stored information that is suitable for the work session. 【0155】 (II) Second Embodiment In the first embodiment, the memory management infrastructure 140 is implemented by a server device 210, a database device 220, and an object storage device 230. In contrast, the second embodiment implements the functions of the memory management infrastructure 140 using a single information processing device 1200. The second embodiment differs from the first embodiment in that the storage of memory information and attribute information, the assignment of hierarchical information, contextual information, and importance information, the selection and provision of memory information, and the updating of importance information are all performed within the information processing device 1200. 【0156】 Figure 12 shows an example of the information processing device 1200 according to this embodiment. 【0157】 The information processing device 1200 comprises a processor 1210, a storage device 1220, and an interface device 1230. The processor 1210 is a device that performs data arithmetic and control processing. The storage device 1220 is a device for storing programs, data, etc. Programs are read and executed by the processor. The interface device 1230 is a device that sends and receives information with a network or other devices, or a device that performs information input and output between a user and a device, etc. The processor 1210 executes programs stored in the storage device 1220 and realizes the processing of the storage management base 140. The functions of the information processing device 1200 (storage unit 1221, management unit 1222, provision unit 1223, update unit 1224, etc.) are realized by software, hardware, or a combination thereof. Some or all of the functions of the information processing device 1200 may be shared by one or more devices. 【0158】 The storage device 1220 includes metadata 1225 and main data 1226. The memory unit 1221 stores the storage information. The management unit 1222 assigns hierarchical information, importance information, and contextual information to each piece of storage information stored in the storage device 1220 by the memory unit 1221, and manages the hierarchical information, importance information, and contextual information as attribute information in association with the storage information. The provision unit 1223 determines the contextual information based on information about the work session, evaluates the relationship of the storage information based on the determined contextual information and the hierarchical information and importance information assigned to the storage information, selects the storage information, and provides the selected storage information to the AI ​​assistant. The update unit 1224 updates the importance information according to usage statistics based on the usage history of the storage information. 【0159】 Metadata 1225 is information that includes hierarchical information, importance information, and contextual information associated with the stored information. Main data 1226 is data that represents the content of the stored information. The information processing device 1200 stores the content of the stored information as main data 1226 and stores the hierarchical information, importance information, and contextual information attached to the stored information as metadata 1225. For example, the information processing device 1200 includes an identifier for the stored information in the metadata 1225 and associates the main data 1226 and the metadata 1225 using the identifier. 【0160】 The main data 1226 may include text data as data representing the content of the stored information. The information processing device 1200 may store structured text as the main data 1226, as an example of text data, which follows a markup notation capable of representing heading structures and bulleted list structures. The information processing device 1200 may use the structured text to express the level of abstraction, basis for judgment, and constraints of the stored information. 【0161】 The interface device 1230 communicates with, for example, the user terminal 110, the message sharing device 120, or the external LLM device 130 to receive information about the work session and transmit selected stored information. Based on the information about the work session received via the interface device 1230, the information processing device 1200 determines contextual information using the provisioning unit 1223, selects stored information, and transmits the selected stored information to the external LLM device 130 via the interface device 1230. Based on the response received from the external LLM device 130, the information processing device 1200 updates the usage history and updates the importance information using the update unit 1224. 【0162】 The message sharing device 120 may have the function of sending and receiving messages between the user and other people or between the user and the information processing device 1200. The message sharing device 120 may be a communication device equipped with a chat-type user interface. The interface device 1230 may receive information about the work session via the message sharing device 120. The interface device 1230 may transmit selected stored information via the message sharing device 120. 【0163】 The interface device 1230 may include an application programming interface (API) for registering, searching, retrieving, and updating the importance level of stored information. For example, it may include an API for registering stored information associated with three attributes (hierarchical information, importance information, and contextual information), an API for searching stored information based on input related to a work session, an API for retrieving selected stored information, and an API for updating importance information based on usage history. 【0164】 In this embodiment, the functions of the memory management infrastructure 140 are realized by a single information processing device 1200. According to this embodiment, the storage of memory information and attribute information, the assignment of hierarchical information, contextual information, and importance information, the selection and provision of memory information, and the updating of importance information are all performed within the information processing device 1200, so that the processing of the memory management infrastructure 140 can be completed and executed within a single information processing device 1200. 【0165】 (III) Addendum The identifiers of components described herein (e.g., prefixes and symbols such as "First" and "Second") are for convenience only and do not limit the number, order, function, arrangement, etc., of the components. The same identifier may refer to different components in different embodiments, and one component may also perform the function of another component. Therefore, the identifiers of components described herein are not intended to limit the technical scope, functional scope, or scope of rights of the components, and each component should be interpreted flexibly according to the context of its embodiment. 【0166】 In this specification, "interface device" means a component that may include one or more interface devices. Such interface devices may include, but are not limited to, I / O (Input / Output) interface devices, communication interface devices, or combinations thereof. For example, an I / O interface device may be configured to function as a user interface and may include at least one input device (e.g., a keyboard, a pointing device) and / or an output device (e.g., a display). These I / O interface devices may be configured to have communication functions that allow connection to remote computing devices, in which case the I / O interface device can also operate as a communication interface device. Furthermore, the communication interface device may include identical communication means (e.g., multiple NICs (Network Interface Cards)) or a combination of different types of communication means (e.g., a NIC and an HBA (Host Bus Adapter)). This enables a flexible configuration that ensures connectivity with heterogeneous systems. Interface devices configured in this way are not limited to a specific hardware configuration and can accommodate future technological advancements and diversification of embodiments. 【0167】 In this specification, “storage device” means a component that may include at least one storage device. Depending on the intended use and system configuration, such storage devices may be classified, for example, into “memory” which temporarily holds data during operation and “persistent storage device” which retains data even after power is lost. Memory can function as a temporary storage medium accessible by the processor and may include volatile, non-volatile, or a combination thereof memory devices. Specifically, examples include, but are not limited to, volatile memory such as DRAM (Dynamic Random Access Memory) and SRAM (Static RAM), and non-volatile memory such as MRAM (Magnetoresistive RAM) and ReRAM (Resistive RAM). Persistent storage devices are components intended for long-term data storage and include devices using non-volatile storage media. Specifically, these may include HDD (Hard Disk Drive), SSD (Solid State Drive), NVMe (Non-Volatile Memory Express) drives, etc. Next-generation storage technologies such as phase-change memory may also be included as examples of storage device configurations. A storage device configured in this way is not limited by the type or architecture of the storage medium, and can accommodate future technological advancements and diversification of implementations. 【0168】 In this specification, "processor" means a component that may include an arithmetic unit or circuit capable of performing at least one processing function. Depending on the application, such a processor may include, for example, a microprocessor device such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit), or it may include implementations using dedicated circuits such as an FPGA (Field-Programmable Gate Array), a CPLD (Complex Programmable Logic Device), or an ASIC (Application Specific Integrated Circuit). A processor may consist of a single core, a multi-core, or a single processor core. A processor may be configured to implement processing functions using a computer program, or it may be directly implemented using hardware circuits, or it may be a hybrid configuration combining these. When processing is performed by a program, the processor may perform processing in cooperation with other components such as memory devices or interface devices. In this specification, a particular function may be described as a "part," but such a function can be realized by a program executed by the processor, an implementation using circuits, or a combination thereof. Therefore, such a function can be considered to be at least a part of the processor. The program may be supplied from an external program source. Examples of program sources include, but are not limited to, network-connected program distribution servers and computer-readable non-temporary storage media. Processors configured in this way are not limited to specific hardware configurations or implementation forms, and can adapt to future technological advancements and diversification of implementations. 【0169】 In this specification, "system" means a set of components that may include at least one computing resource. Such a system may consist of one or more physical computers (dedicated hardware, on-premises servers, etc.) or virtualized computing resources (cloud infrastructure, virtual machines, containers, etc.). A system may include, but is not limited to, cloud computing systems, cluster configurations, serverless environments, etc. A system may be configured within a single device or in a configuration in which multiple computing resources cooperate via a network. Each component may be logically integrated or physically separated. Such a system may also include components such as processors, storage devices, and interface devices, and these components may be implemented as physical devices or realized as virtual configurations. A system configured in this way is not limited to a specific hardware configuration, implementation form, deployment form, etc., and can accommodate future technological advancements and diversification of embodiments. 【0170】 The above-described embodiment has, for example, the following features. 【0171】 (1) A memory management system (e.g., memory management system 100, information processing device 1200) for creating digital works through collaboration between a human and an AI assistant (e.g., generative AI, external LLM device 130, etc.), comprising: a memory unit (e.g., memory unit 211, memory unit 1221) that stores memory information (e.g., information including preconditions, constraints, specialized knowledge, policies, judgment perspectives, or self-assessment perspectives used in the production of digital works); for each piece of memory information stored by the memory unit, hierarchical information indicating the level of abstraction of each piece of memory information (e.g., a hierarchy consisting of the first, second, third, fourth, and fifth levels); importance information indicating the importance of each piece of memory information (e.g., a value indicating the importance of the memory information); and contextual information indicating the work situation in which each piece of memory information is generated or used (e.g., implementation and development context, vision and strategy context, planning and design context, problem-solving context, document creation context, etc.). The system includes a management unit (e.g., management unit 212, management unit 1222) that assigns contextual information (such as a setting context, a creative and innovative context, or a research and analysis context) to the above-mentioned hierarchical information, importance information, and contextual information, and manages them in association with the above-mentioned stored information as attribute information (e.g., attribute information consisting of hierarchical information, importance information, and contextual information); and a provision unit (e.g., provision unit 213, provision unit 1223) that determines contextual information corresponding to the above-mentioned work session based on information about the work session, evaluates the relevance of each piece of stored information to the above-mentioned work session based on the determined contextual information and the hierarchical information and importance information assigned to each piece of stored information stored by the storage unit, selects the stored information related to the above-mentioned work session from the stored information stored by the storage unit based on the evaluation result, and provides the selected stored information to the AI ​​assistant. 【0172】 In the above configuration, the management unit organizes the memory information by assigning hierarchical information, importance information, and contextual information to each piece of memory information, making it easier for the provisioning unit to evaluate the relationships based on the contextual information, hierarchical information, and importance information corresponding to the work session. According to the above configuration, for example, the memory management system can reliably select and present the necessary memory information even if the amount of memory in the memory unit increases, because the provisioning unit selects and provides the memory information that is suitable for the work session based on the evaluation results. 【0173】 (2) The above-described memory management system includes an update unit (e.g., update unit 214, update unit 1224) that updates importance information according to usage statistics (e.g., usage statistics including frequency of use, newness, and last access time) based on the usage history of the memory information stored by the above-described memory unit (e.g., usage history for each piece of memory information). 【0174】 The update unit updates importance information according to usage statistics based on the usage history of the stored information stored by the storage unit. Therefore, the importance information managed by the management unit is updated according to usage statistics based on the usage history of the stored information. With the above configuration, for example, in an operation where the provision unit evaluates the relevance of stored information based on importance information, the storage management system can perform relevance evaluation using importance information updated according to usage statistics. Thus, it can perform relevance evaluation that reflects changes in usage statistics based on usage history in the importance information. 【0175】 (3) The update unit calculates a weighted sum for each piece of stored information stored by the storage unit by multiplying the frequency of use of each piece of stored information, the newness of each piece of stored information, and the last access time of each piece of stored information by a predetermined coefficient, and updates the importance information of each piece of stored information based on the calculated weighted sum (see, for example, Figure 11). 【0176】 In the above configuration, the update unit calculates a weighted sum by multiplying the frequency of use, freshness, and last access time of each piece of stored information stored by the storage unit by a predetermined coefficient, and updates the importance information based on the weighted sum. This makes it easier for the update unit to calculate importance information based on an integrated index that adjusts multiple usage characteristics using coefficients. With the above configuration, for example, in the operation where the provision unit evaluates the relationships between stored information using importance information, the storage management system can perform relationship evaluation using importance information that reflects the frequency of use, freshness, and last access time, so that stored information can be selected with differences in usage reflected in the importance information. 【0177】 (4) The hierarchical information managed by the above-mentioned management department is defined as a hierarchy consisting of the following: the first tier representing basic cognition (e.g., the first tier 410), the second tier representing collaboration patterns (e.g., the second tier 420), the third tier representing specialized knowledge (e.g., the third tier 430), the fourth tier representing strategic thinking (e.g., the fourth tier 440), and the fifth tier representing metacognition (e.g., the fifth tier 450) (see, for example, Figure 4). 【0178】 In the above configuration, the hierarchical information managed by the management department is defined as a hierarchy consisting of a first tier representing basic cognition, a second tier representing collaboration patterns, a third tier representing specialized knowledge, a fourth tier representing strategic thinking, and a fifth tier representing metacognition. Therefore, the management department can easily manage the level of abstraction of memory information as a hierarchy with meaning from the first to the fifth tier. According to the above configuration, for example, in the operation where the provisioning department uses hierarchical information to evaluate the relationships between memory information, the memory management system can distinguish memory information with different levels of abstraction using hierarchical information and evaluate the relationships, so it can select memory information with an appropriate level of abstraction for reference depending on the work session. 【0179】 (5) The contextual information managed by the above-mentioned management department is defined as a context consisting of the following: implementation and development context (e.g., implementation and development context 510), vision and strategy context (e.g., vision and strategy context 520), planning and design context (e.g., planning and design context 530), problem-solving context (e.g., problem-solving context 540), documentation context (e.g., documentation context 550), creativity and innovation context (e.g., creativity and innovation context 560), and research and analysis context (e.g., research and analysis context 570) (see, for example, Figure 5). 【0180】 In the above configuration, the contextual information managed by the management department is defined as a context consisting of implementation and development context, vision and strategy context, planning and design context, problem-solving context, document creation context, creativity and innovation context, and research and analysis context. This makes it easier for the management department to uniformly assign and manage the work status in which stored information is generated or used as one of several types of context. According to the above configuration, for example, in an operation where the provisioning department determines the contextual information corresponding to a work session and selects stored information based on the determined contextual information, the storage management system can distinguish the work status of a work session as contextual information and select stored information accordingly. This allows for the selection of stored information suitable for reference according to differences in work status. 【0181】 (6) The contextual information managed by the above-mentioned management unit includes phase information indicating the progress of work in a work session, and the phase information indicates one of four types of work phases: the investigation and planning phase (e.g., investigation and planning phase), the design phase (e.g., design phase), the material creation and assembly phase (e.g., material creation and assembly phase), and the integration and publication phase (e.g., integration and publication phase). For each work session, the management unit assigns contextual information, including the phase information of each work session, to the stored information generated or used in each work session and stored in the above-mentioned storage unit (see, for example, step S704). The providing unit selects the stored information relevant to the work session from among the multiple stored information stored in the above-mentioned storage unit based on the degree of agreement between the contextual information of the current work session and the contextual information assigned to each stored information stored in the above-mentioned storage unit (see, for example, Figure 8). 【0182】 According to the above configuration, for example, the memory management system can select and present memory information related to the work session using the degree of agreement between the contextual information of the current work session and the contextual information attached to the memory information, as the providing unit selects memory information based on the degree of agreement of the contextual information, including the stage of work progress. 【0183】 (7) The above-mentioned provisioning unit determines contextual information corresponding to the work session based on information about the work session (for example, step S801 in Figure 8, and steps S1001 to S1006 in Figure 10), extracts memory information that matches or approximates the contextual information from the memory information stored by the above-mentioned storage unit as candidate memory information based on the determined contextual information (for example, step S802 in Figure 8), calculates a score indicating the relevance of each memory information to the work session using the hierarchical information and importance information attached to each memory information, the similarity between the search query indicating the content requested by the user in the work session and each memory information, and the degree of agreement between the keywords extracted from the search query and the keywords extracted from each memory information (for example, step S803 in Figure 8), ranks the candidate memory information based on the calculated score, filters the candidate memory information based on the importance information of each memory information, selects a predetermined number of memory information from the filtered candidate memory information (for example, step S804 in Figure 8) and provides it to the AI ​​assistant. 【0184】 According to the above configuration, for example, the memory management system ranks candidate memory information based on scores, filters the candidate memory information based on importance information, and selects a predetermined number of memory information from the filtered candidate memory information to provide to the AI ​​assistant. This allows for prioritizing and selecting memory information suitable for the work session and presenting it in a controlled, predetermined number. 【0185】 (8) The above-mentioned storage unit stores each user's memory information in cloud storage (e.g., object storage device 230) in a storage area separated for each user identifier in the cloud environment, and the above-mentioned management unit stores hierarchical information, importance information, and contextual information attached to each user's memory information in a database (e.g., database device 220) in a storage area separated for each user identifier in the cloud environment. 【0186】 With the above configuration, for example, the memory management system can select memory information for each user identifier and provide it to the AI ​​assistant. Furthermore, with the above configuration, for example, the memory management system can manage the memory information stored in cloud storage separately from the hierarchical information, importance information, and contextual information stored in the database, making it easier to manage the content and attribute information of the memory information independently. 【0187】 Memory information is a unit of information used in the process of creating digital works through collaboration between humans and artificial intelligence assistants. Memory information may be a data structure that includes text, tables, images, audio, video, design data, programs, settings, history, evaluation results, reference information, and identification information. Memory information may also include information that includes prerequisites, constraints, specialized knowledge, policies, judgment criteria, self-assessment criteria, goals, evaluation indicators, instruction types, division of roles, and production process procedures. Memory information may also be composed of composite data that links multiple elemental data. 【0188】 A work session is a concept that describes a set of tasks performed collaboratively by a human and an artificial intelligence assistant. A work session may be set up to cover part or all of the production process of a digital work. A work session may be defined as a period that includes the subject of production, purpose of production, deadline, person in charge, materials used, means of use, and work history. A work session may be defined as a segment from the start to the end of a dialogue, or as a segment of a dialogue. 【0189】 Information regarding a work session identifies the work session, estimates the work status of the work session, and represents the content of the requests within the work session. This information may include user input, dialogue history, the name of the work object, the status of the production process, the date and time of the work, the processes performed, identification information of referenced stored information, and environmental information. This information may be obtained as structured data, semi-structured data, or unstructured data. 【0190】 Contextual information is classification information that indicates the working conditions under which memory information is generated or used. Contextual information may be defined as a classification system that includes implementation and development contexts, vision and strategy contexts, planning and design contexts, problem-solving contexts, documentation contexts, creativity and innovation contexts, and research and analysis contexts. Contextual information is not limited to the above classification systems and may also include classification systems that represent the type of production object, production process, knowledge domain used, form of request, and form of deliverable. Contextual information may be represented by a single contextual value or by a set of multiple contextual values. 【0191】 Phase information is information that indicates the stage of work progress in a work session. Phase information may be defined as information indicating one of the following phases: research and planning phase, design phase, material creation and assembly phase, or integration and publication phase. Phase information is not limited to the above stages and may be defined as stages including prototyping, verification, revision, and final confirmation. 【0192】 Hierarchical information is classification information that indicates the level of abstraction of memory information. Hierarchical information may be defined as a hierarchical system that includes the first level representing basic cognition, the second level representing collaboration patterns, the third level representing specialized knowledge, the fourth level representing strategic thinking, and the fifth level representing metacognition. Hierarchical information is not limited to the above hierarchical system, and may also be defined as a multi-level abstraction classification based on measures of generality and specificity. 【0193】 Importance information indicates the utility or priority of stored information. Importance information may be expressed as a continuous value, discrete value, segmental value, rank, or weight. Importance information may be updated according to usage statistics, which may include evaluation results indicating frequency of use, recency, last access time, number of references, access date and time, and contribution to the work session. 【0194】 Attribute information is a collection of supplementary information attached to stored information. Attribute information may be defined as information that includes hierarchical information, importance information, and contextual information. Attribute information may further include identification information, creation date and time, update date and time, creator, number of references, reference history, tags, association information, and storage location information. 【0195】 A search query is inquiry information that represents what a user is looking for in a work session. A search query may include natural language sentences, keyword columns, question sentences, directives, conditional expressions, and structured conditions. A search query may also be composed of a compound query that combines multiple conditions. 【0196】 Similarity is a measure of how close a search query is to stored information. Similarity may be calculated based on the distance or similarity between vector representations, string similarity, semantic similarity, or edit distance. It may also be calculated using the dot product of embedding representations from a machine learning model, cosine similarity, or a distance function. 【0197】 The degree of match is a measure that indicates the degree of agreement between keywords extracted from a search query and keywords extracted from stored information. The degree of match may be calculated as the number of matches, weighted number of matches, match rate, partial match rate, synonym match rate, or normalized match rate. Keyword extraction may be performed by morphological analysis, normalization, irrelevant word removal, named entity recognition, and technical term extraction. 【0198】 The relevance score is an evaluation value representing the usefulness of the memory information for a work session. The relevance score may be calculated using a combined function that utilizes hierarchical information, importance information, similarity, and agreement. The combined function may include estimation based on addition, weighted sum, normalization, nonlinear functions, thresholding, and ranking learning. 【0199】 Ranking is the process of ordering candidate memory information based on a score indicating relevance. Filtering is the process of excluding candidate memory information based on importance information, hierarchical information, contextual information, count conditions, and threshold conditions. Selection is the process of selecting a predetermined number of memory information from the results of ranking and filtering. Provision is the process of sending the selected memory information to the artificial intelligence assistant and outputting it in a format that the artificial intelligence assistant can refer to. 【0200】 A cloud environment is a computing environment that allows access to computing and storage resources via a network. Cloud storage is a storage service or storage device that stores stored information. A database is a data management means that stores and makes searchable attribute information. Databases may include relational databases, key-value databases, document databases, and graph databases. 【0201】 A user identifier is identification information that distinguishes multiple users. A separated storage area is a storage area that is logically or physically separated for each user identifier. Separated storage areas may be implemented as partitioning on cloud storage, namespace partitioning on a database, separation based on access control, or separation based on cryptographic keys. A storage management system may determine the storage area to be referenced based on the user identifier and perform a relevance evaluation and selection of the storage information belonging to the referenced storage area. [Explanation of symbols] 【0202】 100...Memory management system, 110...User terminal, 120...Message sharing device, 130...External LLM device, 140...Memory management infrastructure.

Claims

[Claim 1] A memory management system for creating digital works through collaboration between humans and AI assistants, A memory unit that stores memory information, A management unit assigns to each piece of stored information stored by the storage unit hierarchical information indicating the level of abstraction of the stored information, importance information indicating the importance of the stored information, and contextual information indicating the work situation in which the stored information is generated or used, and manages the hierarchical information, importance information, and contextual information as attribute information associated with the stored information. A providing unit that determines contextual information corresponding to a work session based on information about the work session, evaluates the relationship of each piece of stored information to the work session based on the determined contextual information and the hierarchical information and importance information assigned to each piece of stored information stored in the storage unit, selects the stored information related to the work session from the stored information stored in the storage unit based on the evaluation result, and provides the selected stored information to the AI ​​assistant. A memory management system equipped with the following features. [Claim 2] A memory management system according to claim 1, A memory management system comprising an update unit that updates importance information according to usage statistics based on the usage history of the memory information stored in the memory unit. [Claim 3] A memory management system according to claim 2, The update unit calculates a weighted sum for each piece of stored information stored by the storage unit by multiplying the frequency of use of the stored information, the newness of the stored information, and the last access time of the stored information by a predetermined coefficient, and updates the importance information of the stored information based on the calculated weighted sum. [Claim 4] A memory management system according to claim 1, The hierarchical information managed by the aforementioned management department is defined as a memory management system consisting of a first hierarchical level representing basic cognition, a second hierarchical level representing collaborative patterns, a third hierarchical level representing specialized knowledge, a fourth hierarchical level representing strategic thinking, and a fifth hierarchical level representing metacognition. [Claim 5] A memory management system according to claim 1, A memory management system in which contextual information managed by the aforementioned management department is defined as contexts consisting of implementation and development context, vision and strategy context, planning and design context, problem-solving context, documentation context, creativity and innovation context, and research and analysis context. [Claim 6] A memory management system according to claim 5, The contextual information managed by the aforementioned management unit includes phase information indicating the progress of work in a work session. The aforementioned phase information indicates one of four types of work phases: the research and planning phase, the design phase, the material creation and assembly phase, and the integration and publication phase. The management unit, for each work session, adds contextual information, including phase information of the work session, to the stored information generated or used in the work session and stored by the storage unit. The aforementioned provisioning unit is a memory management system that selects memory information related to the work session from among a plurality of memory information stored by the storage unit, based on the degree of agreement between the context information of the current work session and the context information attached to each memory information stored by the storage unit. [Claim 7] A memory management system according to claim 1, The aforementioned supply unit is, Based on information about the work session, determine the contextual information corresponding to the work session. Based on the determined contextual information, the memory unit extracts memory information that matches or approximates the contextual information from among the memory information stored in the memory unit as candidate memory information. For each of the extracted candidate memory pieces, a score indicating the relevance of the memory piece to the work session is calculated using the hierarchical information and importance information assigned to the memory piece, the similarity between the search query indicating the content the user is seeking in the work session and the memory piece, and the degree of agreement between the keywords extracted from the search query and the keywords extracted from the memory piece. A memory management system that ranks the candidate memory information based on the calculated score, filters the candidate memory information based on the importance information of each memory information, and selects a predetermined number of memory information from the filtered candidate memory information to provide to the AI ​​assistant. [Claim 8] A memory management system according to claim 1, The aforementioned storage unit stores each user's memory information in cloud storage in a separate storage area for each user identifier in the cloud environment. The aforementioned management unit is a memory management system that stores hierarchical information, importance information, and contextual information attached to each user's memory information in a database, in memory areas separated for each user identifier in a cloud environment. [Claim 9] A memory management method for creating digital works through collaboration between humans and AI assistants, The memory unit stores memory information, The management unit assigns to each piece of stored information stored by the storage unit hierarchical information indicating the level of abstraction of the stored information, importance information indicating the importance of the stored information, and contextual information indicating the work situation in which the stored information is generated or used, and manages the hierarchical information, importance information, and contextual information as attribute information associated with the stored information. The provisioning unit determines contextual information corresponding to the work session based on information related to the work session, evaluates the relationship of each piece of stored information to the work session based on the determined contextual information and the hierarchical information and importance information assigned to each piece of stored information stored by the storage unit, selects the stored information related to the work session from the stored information stored by the storage unit based on the evaluation result, and provides the selected stored information to the AI ​​assistant. A memory management method that includes this. [Claim 10] A program for causing a computer to execute the memory management method described in claim 9.

Images