Intelligent agent interaction memory cooperation method and device based on memory grading and medium

Abstract

The invention discloses an agent interaction memory cooperation method and device based on memory grading and a medium, and the method comprises the steps: obtaining the original interaction data of a user, and carrying out the context interaction analysis of the original interaction data, so as to obtain an interaction state variable; based on the interaction state variable, determining user interaction multiplexing state data through cross-session multiplexing of the user unique identifier; obtaining an iteratable shared knowledge base through high-frequency multiplexing knowledge precipitation according to the user interaction multiplexing state data; performing data interaction control on the interaction state variable, the user interaction multiplexing state data and the iteratable shared knowledge base to determine a directional calling state of the memory data; and based on the directional calling state of the memory data, through multi-level cache matching of agent demand analysis, obtaining agent interactive memory collaborative storage data. Through the method, the technical problem that an artificial intelligence interaction memory function cannot meet real-time interaction dynamic and personalized requirements in the prior art is solved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, device and medium for intelligent agent interaction and memory collaboration based on memory hierarchy. Background Technology

[0002] With the widespread application of artificial intelligence technologies such as large language models, the core challenge of intelligent dialogue systems has gradually shifted from generating fluent responses to achieving continuous interaction with memory and cognition. Currently, how to enable intelligent agents to remember the context, user preferences, and task progress in different conversations, just like humans, and thus achieve truly personalized services, has become a key bottleneck restricting the improvement of user experience and intelligence level.

[0003] In existing AI interaction systems, memory functions are typically characterized by singularity and isolation. Short-term conversational memory can only temporarily maintain the context of a single dialogue and is discarded after the conversation ends, causing users to need to repeat information in multiple rounds of interaction, resulting in a fragmented experience. Meanwhile, existing long-term knowledge bases are mostly static and general knowledge collections that cannot be correlated with the real-time dynamics and personalized needs of specific users. Summary of the Invention

[0004] This application provides a method, device, and medium for intelligent agent interactive memory collaboration based on memory hierarchy, which solves the technical problem that the existing artificial intelligence interactive memory function cannot meet the needs of real-time interactive dynamics and personalization.

[0005] In a first aspect, embodiments of this application provide a method for intelligent agent interaction memory collaboration based on memory hierarchy. The method includes: acquiring original user interaction data and performing contextual interaction analysis on the original interaction data to obtain interaction state variables; determining user interaction reuse state data based on the interaction state variables through cross-session reuse of the user's unique identifier; obtaining an iteratively shared knowledge base based on the user interaction reuse state data through high-frequency reuse knowledge accumulation; performing data interaction control on the interaction state variables, user interaction reuse state data, and iteratively shared knowledge base to determine the memory data directional call state; and obtaining intelligent agent interaction memory collaborative storage data through multi-level cache matching based on the memory data directional call state and intelligent agent demand analysis.

[0006] In one implementation of this application, contextual interaction analysis is performed on the original interaction data to obtain interaction state variables. Specifically, this includes: sorting the original interaction data in a FIFO queue to obtain a context data queue; monitoring the context window threshold of the context data queue to obtain semantic summary processing data, wherein the semantic summary processing data includes: refined summary fragments and removed original data; performing local cache dumping on the removed original data to obtain historical data backup; determining task state variables based on the refined summary fragments through key-value pair state updates of multiple rounds of tasks; and obtaining interaction state variables by filtering through session continuation instructions based on historical data backups and task state variables.

[0007] In one implementation of this application, user interaction reuse state data is determined based on interaction state variables and through cross-session reuse of unique user identifiers. Specifically, this includes: determining user profile parameters based on interaction state variables through real-time information parsing; wherein the user profile parameters include: user preferences and user attributes; obtaining historical user profile parameters, and obtaining updated user profile data by fusing the confidence scores of the profile data with the user profile parameters and historical user profile parameters; performing high-frequency knowledge recognition on the updated user profile data to determine user-dimensional knowledge vectors; binding the user-dimensional knowledge vectors to the user ID, and determining user interaction reuse state data through vector similarity retrieval based on stage state persistence when a new session is initiated by the user ID.

[0008] In one implementation of this application, an iteratively shareable knowledge base is obtained based on user interaction reuse status data and through high-frequency reuse knowledge accumulation. Specifically, this includes: screening high-value answers from user interaction reuse status data to obtain candidate high-quality question-answer pairs to be added to the database; analyzing feedback differences from user interaction reuse status data to obtain new knowledge question-answer pairs to be added to the database; determining multi-level indexed knowledge question-answer pairs based on the candidate high-quality question-answer pairs and the new knowledge question-answer pairs to be added to the database through multi-level classification indexing; and obtaining an iteratively shareable knowledge base based on the multi-level indexed knowledge question-answer pairs through performance monitoring.

[0009] In one implementation of this application, data interaction control is performed on the interaction state variables, user interaction reuse state data, and iterable shared knowledge base to determine the memory data targeted call state. Specifically, this includes: performing data linkage control on the interaction state variables, user interaction reuse state data, and iterable shared knowledge base to determine the data linkage state; obtaining the user data scheduling state through data flow scheduling based on the data linkage state; and determining the memory data targeted call state based on the user data scheduling state through rule priority determination.

[0010] In one implementation of this application, based on the memory data directed call state, multi-level cache matching through agent demand analysis is used to obtain agent interactive memory collaborative storage data. Specifically, this includes: providing user-driven feedback on the memory data directed call state to obtain knowledge accumulation data; accumulating the knowledge accumulation data into an iteratively shared knowledge base and storing it through a distributed database to determine the agent interactive memory collaborative storage data.

[0011] In one implementation of this application, distributed database storage is used to determine the collaborative storage data of intelligent agent interactive memory. Specifically, this includes: storing the temporary caching needs of short-term memory through an in-memory database; storing the user-bound persistence needs of medium-term memory through a combination of relational database and vector database; and storing the knowledge accumulation needs of long-term memory through a combination of distributed database and knowledge base system.

[0012] In one implementation of this application, after obtaining the agent interaction memory collaborative storage data by multi-level cache matching through agent demand analysis based on the state of targeted invocation of memory data, the method further includes: preloading the agent interaction memory collaborative storage data to obtain pre-push knowledge data.

[0013] Secondly, embodiments of this application also provide a memory-hierarchical intelligent agent interactive memory collaboration device, characterized in that the device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: acquire user's original interaction data and perform contextual interaction analysis on the original interaction data to obtain interaction state variables; based on the interaction state variables, determine user interaction reuse state data through cross-session reuse of user unique identifiers; based on the user interaction reuse state data, obtain an iteratively shared knowledge base through high-frequency reuse knowledge accumulation; perform data interaction control on the interaction state variables, user interaction reuse state data, and iteratively shared knowledge base to determine the memory data directed call state; based on the memory data directed call state, obtain intelligent agent interactive memory collaborative storage data through multi-level cache matching of intelligent agent demand analysis.

[0014] Thirdly, embodiments of this application also provide a non-volatile computer storage medium for intelligent agent interactive memory collaboration based on memory hierarchy, storing computer-executable instructions. The computer-executable instructions are characterized by: acquiring original user interaction data and performing contextual interaction analysis on the original interaction data to obtain interaction state variables; determining user interaction reuse state data based on the interaction state variables through cross-session reuse of the user's unique identifier; obtaining an iteratively shared knowledge base based on the user interaction reuse state data through high-frequency reuse knowledge accumulation; performing data interaction control on the interaction state variables, user interaction reuse state data, and iteratively shared knowledge base to determine the memory data directional call state; and obtaining intelligent agent interactive memory collaborative storage data through multi-level cache matching based on intelligent agent demand analysis, based on the memory data directional call state.

[0015] This application provides a method, device, and medium for intelligent agent interaction memory collaboration based on memory hierarchy. Through an integrated architecture combining a three-level memory module, a collaborative scheduling layer, and a storage layer, it solves the technical problem that the existing artificial intelligence interactive memory function cannot meet the needs of real-time dynamic and personalized interaction. Through hierarchical memory storage and intelligent collaborative scheduling, it realizes the full lifecycle persistent management of interactive states and the efficient reuse of memory resources. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart of an intelligent agent interaction memory collaboration method based on memory hierarchy provided in this application embodiment; Figure 2 This is a schematic diagram of the internal structure of an intelligent agent interactive memory collaboration device based on memory hierarchy, provided as an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] This application provides a method, device, and medium for intelligent agent interaction memory collaboration based on memory hierarchy. Through an integrated architecture combining a three-level memory module, a collaborative scheduling layer, and a storage layer, it solves the technical problem that the existing artificial intelligence interactive memory function cannot meet the needs of real-time dynamic and personalized interaction. Through hierarchical memory storage and intelligent collaborative scheduling, it realizes the full lifecycle persistent management of interactive states and the efficient reuse of memory resources.

[0019] The technical solutions proposed in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0020] Figure 1 A flowchart illustrating a memory-hierarchical intelligent agent interaction and memory collaboration method provided in this application embodiment. Figure 1 As shown in the figure, the intelligent agent interaction memory collaboration method based on memory hierarchy provided in this application embodiment specifically includes the following steps: Step 101: Obtain the user's original interaction data and perform contextual interaction analysis on the original interaction data to obtain the interaction state variables.

[0021] For example, the three-level memory module is the core data carrier of the system. It is divided into short-term memory module, medium-term memory module and long-term memory module according to the timeliness of memory, the scope of reuse and the storage period, forming a complete memory link from temporary interaction to persistent accumulation. Among them, for the short-term memory module, this application realizes the linkage between data processing logic and large model context window by performing context interaction analysis on the original interaction data, ensuring the real-time interaction response.

[0022] Specifically, contextual interaction analysis is performed on the original interaction data to obtain interaction state variables, including: sorting the original interaction data into a FIFO queue to obtain a context data queue; monitoring the context window threshold of the context data queue to obtain semantic summary processing data, wherein the semantic summary processing data includes: refined summary fragments and removed original data; local caching and dumping of the removed original data to obtain historical data backup; determining task state variables based on the refined summary fragments through key-value pair state updates of multiple rounds of tasks; and obtaining interaction state variables through session continuation instructions based on historical data backups and task state variables.

[0023] In one embodiment, the short-term memory module focuses on supporting real-time interaction within a single session. Its core objectives are low latency, high responsiveness, and temporary storage to adapt to the context processing needs of large models. Its storage objects cover the entire interaction data of a single session, including the user's continuous question text, the agent's real-time response content, the input parameters and return results of tool calls, the current progress status of multi-round tasks, and temporary variables associated with the session context.

[0024] The storage medium employs a high-performance in-memory database, while a local cache is configured as a backup. This ensures millisecond-level access speeds for data within a session and prevents temporary data loss due to memory overflow. In terms of processing logic, an independent cache space is automatically created at session startup, uniquely identified by the session ID. All interactive data is stored in a FIFO queue structure in timestamp order to ensure the temporal continuity of the context. The system monitors the matching between the total amount of session data and the threshold of the large model's context window in real time. When the data volume approaches 80% of the threshold, the large model calls a semantic summarization algorithm to extract the core information of earlier content, removes redundant expressions, retains concise fragments, and temporarily stores the original complete data in the local cache for later review.

[0025] In multi-round task scenarios, state variables are stored in key-value pairs. The state value is automatically updated after each process node is completed. By default, cached data is retained for 1 hour after the session ends. Temporary data is automatically cleaned up after the timeout. If the user triggers a session continuation command, the core state variables are synchronized to the medium-term memory module.

[0026] Step 102: Based on the interaction state variables, determine the user interaction reuse state data through cross-session reuse of the user's unique identifier.

[0027] For example, the medium-term memory module is positioned as a user or task-level persistent memory unit, bound to a unique user identifier, and used to store user-related information and task status data that can be reused across sessions. Based on the interaction state variable, this application achieves multiple reuses of the interaction for a single user interaction by reusing the unique user identifier across sessions, thereby improving the reuse rate of intelligent agent interaction data.

[0028] Specifically, based on interaction state variables, user interaction reuse state data is determined through cross-session reuse of unique user identifiers. This includes: determining user profile parameters based on interaction state variables through real-time information parsing; where user profile parameters include: user preferences and user attributes; obtaining user historical profile parameters, and obtaining updated user profile data by fusing the confidence scores of profile data based on the user profile parameters and the user historical profile parameters; performing high-frequency knowledge recognition on the updated user profile data to determine user-dimensional knowledge vectors; binding user-dimensional knowledge vectors and user IDs, and determining user interaction reuse state data through vector similarity retrieval based on stage state persistence when a new session is initiated by the user ID.

[0029] In one embodiment, the medium-term memory module focuses on cross-session reuse requirements, with the core objective of enabling cross-session continuation of user information and task status, and all data being strongly bound to the user's unique identifier.

[0030] The stored objects include core user profile information, complete status records of cross-session tasks, and knowledge content that users have recently mentioned frequently. The user profile information includes preference settings, identity attributes, and historical demand keywords. The task status records include the progress of various cross-session tasks, and the high-frequency knowledge content involves commonly used user terms, fixed demand templates, etc.

[0031] The storage medium employs a hybrid approach combining relational and vector databases. The relational database stores structured user profiles and task status data, ensuring data consistency and transactionality. The vector database stores high-frequency knowledge vector data, supporting rapid semantic retrieval. In terms of processing logic, user profile data is maintained through real-time extraction and incremental updates. During the session, the large model parses key information from user input in real time, compares it with existing profile data in the database, and directly supplements it with any new information.

[0032] Conflict information is handled through confidence level determination. Content that the user explicitly modifies is directly overwritten, while ambiguous statements are marked as pending confirmation. High-frequency knowledge needs to be vectorized. The large model identifies knowledge content that appears three or more times in the user session, calls the embedding model to convert it into a fixed-dimensional vector, associates it with the user ID and stores it in the vector database. In subsequent sessions, when the user input triggers semantic matching and the similarity reaches 0.85 or higher, relevant knowledge is retrieved from the vector database first to avoid duplicate input.

[0033] The task status data adopts a phased storage and cross-session recovery mechanism. After each task phase is completed, it is automatically synchronized to the relational database. When the user initiates a related task again, the system retrieves the historical status through the user ID and automatically restores to the last interrupted node, achieving seamless task connection.

[0034] Step 103: Based on the user interaction reuse status data, and through the accumulation of frequently reused knowledge, obtain an iteratively shareable knowledge base.

[0035] For example, the long-term memory module is positioned as a knowledge-level accumulation and sharing unit, targeting intelligent agents across the entire platform and all users. It is used to accumulate general knowledge, industry rules, and frequently reused information. Based on user interaction reuse status data, this application realizes the construction of an iteratively optimized shared knowledge base through the accumulation of frequently reused knowledge. Effective feedback generated by users during the interaction process can be updated to the iteratively shared knowledge base in real time, improving the dynamic adaptation to user needs.

[0036] Specifically, based on user interaction reuse status data, an iteratively shareable knowledge base is obtained through high-frequency reuse knowledge accumulation. This includes: screening high-value answers from user interaction reuse status data to obtain candidate high-quality question-answer pairs to be added to the database; analyzing feedback differences from user interaction reuse status data to obtain new knowledge question-answer pairs to be added to the database; determining multi-level indexed knowledge question-answer pairs based on the candidate high-quality question-answer pairs and the new knowledge question-answer pairs to be added to the database through multi-level classification indexing; and obtaining an iteratively shareable knowledge base based on the multi-level indexed knowledge question-answer pairs through performance monitoring.

[0037] In one embodiment, the long-term memory module focuses on knowledge accumulation and sharing across the entire platform. Its core objective is to build a self-optimizing general knowledge base that supports reuse by all intelligent agents across users and scenarios.

[0038] The stored objects include general domain basic knowledge, industry-specific rules, question-and-answer pairs optimized by user feedback, and structured knowledge that is frequently reused across scenarios. General domain basic knowledge includes common sense questions and answers, standard process descriptions, etc. Industry-specific rules include financial compliance clauses, medical industry standards, etc. Question-and-answer pairs include valid questions and answers confirmed by likes and accurate questions and answers corrected by dislikes. Structured knowledge involves formula templates, report formats, etc.

[0039] The storage medium adopts a collaborative solution of distributed database and knowledge base system. The distributed database is used to store massive amounts of structured knowledge data and support high-concurrency access, while the knowledge base system is used to build a full-text search index and classification system to ensure the accuracy and efficiency of knowledge retrieval.

[0040] In terms of processing logic, the knowledge database combines user feedback-driven and manual review-assisted mechanisms. When a user likes a response from an AI agent, the system automatically extracts the "question-response" pair of that interaction, marks it as high-confidence data, and directly stores it in the database. When a user dislikes a response and submits text feedback, the large model analyzes the differences between the feedback content and the original response. If it is for expression optimization, the original question-and-answer pair is corrected; if it is for new knowledge, a new question-and-answer pair is constructed. Sensitive areas require manual review, while general areas are automatically verified before being stored in the database. The knowledge database constructs a three-level classification index of domain, category, and tag. Each knowledge entry is associated with attributes such as keywords, applicable scenarios, and confidence levels, supporting multi-dimensional retrieval by scenario, keyword, semantic relevance, and other dimensions. When any AI agent initiates a knowledge call request, the system matches relevant entries in the knowledge database through the retrieval interface, sorts the results by confidence level and usage frequency, and returns the results. At the same time, it records the number of knowledge calls and user feedback, providing data support for subsequent knowledge optimization, such as increasing the weight of high-value knowledge and eliminating low-reusability knowledge.

[0041] Step 104: Perform data interaction control on the interaction state variables, user interaction reuse state data, and iterable shared knowledge base to determine the targeted call state of memory data.

[0042] Specifically, data interaction control is performed on interaction state variables, user interaction reuse state data, and iterative shared knowledge base to determine the directed call state of memory data. This includes: performing data linkage control on interaction state variables, user interaction reuse state data, and iterative shared knowledge base to determine the data linkage state; obtaining the user data scheduling state through data flow scheduling based on the data linkage state; and determining the directed call state of memory data through rule priority determination based on the user data scheduling state.

[0043] For example, the memory collaborative scheduling mechanism is the core of the system to realize the linkage of three-level memory modules and the persistence of state. The core is to dynamically trigger module calls, standardize data flow, and clarify usage priorities based on interaction scenarios, user needs and data attributes, so as to ensure accurate reuse of memory resources and consistency of state.

[0044] After receiving a user request, the collaborative scheduling layer first parses the request type, user identifier, task attributes, and context, and then triggers the corresponding module according to preset logic. For single-session real-time interactions, short-term memory is prioritized to support immediate continuity; when crossing session tasks or querying preferences, medium-term memory is associated with the user identifier to extract profiles and historical states, and integrate them with the current context; when general knowledge or industry rules are involved, long-term memory retrieval is triggered synchronously. Key information such as task states or user-tagged information that have completed a phase loop in short-term memory is automatically synchronized to medium-term memory for persistence; knowledge frequently accessed or receiving multiple positive feedbacks in medium-term memory is filtered and deposited into long-term memory, forming a complete data flow loop.

[0045] The linkage follows the logic of context priority, user association, and knowledge supplementation. It first obtains the current context from short-term memory to build a basic context, then retrieves user-specific data from medium-term memory, and finally calls on long-term memory to supplement professional knowledge.

[0046] In multi-round tasks, the scheduling layer synchronizes the task status to the intermediate memory in real time, supporting recovery after interruption; when users provide feedback, the corresponding modules are linked to update data, positive feedback optimizes the long-term memory, and personalized feedback is synchronized to the intermediate memory user profile.

[0047] Data prioritization is based on three dimensions: scenario adaptability, data timeliness, and confidence level. Within a single session, short-term memory context has the highest priority; across sessions, user-specific data from medium-term memory takes precedence; and during knowledge retrieval, high-confidence, frequently accessed items from long-term memory are prioritized. In case of data conflicts, the logic is as follows: latest input takes precedence, manually reviewed data takes precedence, and high-confidence feedback data takes precedence, ensuring reasonable and accurate data usage.

[0048] Step 105: Based on the memory data, the state is called in a targeted manner. Through multi-level cache matching of agent demand analysis, the agent interaction memory collaborative storage data is obtained.

[0049] Specifically, based on the memory data directional call state, multi-level cache matching through agent demand analysis is used to obtain agent interactive memory collaborative storage data, including: user-driven feedback on the memory data directional call state to obtain knowledge accumulation data; accumulating the knowledge accumulation data into an iteratively shared knowledge base and storing it through a distributed database to determine the agent interactive memory collaborative storage data.

[0050] Furthermore, distributed database storage is used to determine the collaborative storage data for intelligent agent interaction memory. Specifically, this includes: storing the temporary caching needs of short-term memory through an in-memory database; storing the user-bound persistence needs of medium-term memory through a combination of relational and vector databases; and storing the knowledge accumulation needs of long-term memory through a combination of distributed databases and knowledge base systems.

[0051] Furthermore, after obtaining the agent interaction memory collaborative storage data by using multi-level cache matching based on agent demand analysis to call the state based on memory data, the method also includes: preloading the agent interaction memory collaborative storage data to obtain pre-push knowledge data.

[0052] In one embodiment, the storage layer serves as the foundation for system state persistence. Based on the varying storage requirements of the three memory modules, a "hybrid storage architecture" is employed, configuring different storage media to ensure both storage performance and persistence of the stored data. To address the temporary caching requirements of the short-term memory module, an in-memory database is used as the primary storage medium to ensure high-speed read / write and real-time access to data within a session. To address the user-bound persistence requirements of the mid-term memory module, a storage method combining relational databases and vector databases is adopted. This not only enables stable storage of structured data such as user profiles and task statuses, but also supports vectorized retrieval of high-frequency knowledge. To address the large-scale knowledge accumulation requirements of long-term memory modules, a storage solution combining distributed databases and knowledge base systems is adopted to support the persistent storage, efficient retrieval, and iterative updates of massive amounts of general knowledge, industry rules, and question-and-answer pairs. The storage layer interfaces with the collaborative scheduling layer through a unified data interface to achieve standardized execution of operations such as writing, reading, updating, and deleting memory data, ensuring the consistency and reliability of state persistence.

[0053] Furthermore, a knowledge preheating data flow is established from long-term memory to medium-term and short-term memory. High-frequency, high-confidence general knowledge or historical knowledge strongly related to specific users in the long-term memory module is actively pushed to the user's medium-term memory module when a new user session is created through personalized knowledge preloading processing, and further loaded into the context of the short-term memory module of the current session.

[0054] High-value knowledge entries in the long-term memory module are processed through user profiling and knowledge association analysis to obtain a pre-loaded knowledge package for the user. Through session initialization warm-up processing, the package is pushed and stored in the user's vector database and relational database in the medium-term memory module when the user logs in or initiates a new session. When the user's session officially begins, the specific knowledge pre-loaded in the medium-term memory module is directly loaded into the initial context data queue of the short-term memory module through context active injection processing.

[0055] The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide an intelligent agent interactive memory collaboration device based on memory hierarchy, the structure of which is as follows: Figure 2 As shown.

[0056] Figure 2 This is a schematic diagram of the internal structure of an intelligent agent interactive memory collaboration device based on memory hierarchy, provided as an embodiment of this application. Figure 2 As shown, the device includes: At least one processor 201; And a memory 202 that is communicatively connected to at least one processor; The memory 202 stores instructions executable by at least one processor, which are executed by at least one processor 201 to enable at least one processor 201 to: The system acquires raw user interaction data and performs contextual interaction analysis to obtain interaction state variables. Based on these variables, it determines user interaction reuse state data through cross-session reuse of unique user identifiers. Based on this data, it obtains an iteratively shared knowledge base through high-frequency reuse knowledge accumulation. It then performs data interaction control on the interaction state variables, user interaction reuse state data, and the iteratively shared knowledge base to determine the targeted call state of memory data. Finally, based on this targeted call state, it obtains agent interaction memory collaborative storage data through multi-level cache matching based on agent demand analysis.

[0057] Some embodiments of this application provide corresponding to Figure 1 A non-volatile computer storage medium based on memory hierarchy and intelligent agent interactive memory collaboration, storing computer-executable instructions, wherein the computer-executable instructions are configured as follows: The system acquires raw user interaction data and performs contextual interaction analysis to obtain interaction state variables. Based on these variables, it determines user interaction reuse state data through cross-session reuse of unique user identifiers. Based on this data, it obtains an iteratively shared knowledge base through high-frequency reuse knowledge accumulation. It then performs data interaction control on the interaction state variables, user interaction reuse state data, and the iteratively shared knowledge base to determine the targeted call state of memory data. Finally, based on this targeted call state, it obtains agent interaction memory collaborative storage data through multi-level cache matching based on agent demand analysis.

[0058] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for IoT devices and media are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0059] The systems, media, and methods provided in this application are one-to-one correspondences. Therefore, the systems and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0060] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0061] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0062] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0063] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0064] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0065] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0066] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0067] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0068] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. An agent interaction memory coordination method based on memory hierarchy, characterized in that, The method comprises: obtaining user original interaction data, and performing context interaction analysis on the original interaction data to obtain interaction state variables; based on the interaction state variables, through the cross-session reuse of the user unique identifier, determine the user interaction reuse state data; According to the user interaction reuse state data, through the high-frequency reuse knowledge precipitation, obtain the iteratively shared knowledge base; Data interaction control is performed on the interaction state variables, the user interaction reuse state data and the iteratively shared knowledge base to determine the memory data directional calling state; Based on the memory data directional calling state, through the multi-level cache matching of agent demand analysis, obtain the agent interaction memory collaborative storage data.

2. The memory hierarchy-based agent interaction memory coordination method according to claim 1, characterized in that, The context interaction analysis on the original interaction data to obtain the interaction state variables, specifically includes: The FIFO queue sorting is performed on the original interaction data to obtain the context data queue; The context window threshold monitoring is performed on the context data queue to obtain the semantic summary processing data: wherein, the semantic summary processing data includes: refined summary fragments and removed original data; The local cache dump is performed on the removed original data to obtain the historical data backup; Based on the refined summary fragments, through the key-value pair state update of multiple rounds of tasks, to determine the task state variables; According to the historical data backup and the task state variables, through the session continuation instruction filtering, to obtain the interaction state variables.

3. The memory hierarchy-based agent interaction memory coordination method according to claim 1, wherein, Based on the interaction state variables, through the cross-session reuse of the user unique identifier, determine the user interaction reuse state data, specifically includes: Based on the interaction state variables, through real-time information analysis, determine the user portrait parameters; Wherein, the user portrait parameters include: user preference, user attribute; Obtain user historical portrait parameters, and according to the user portrait parameters and the user historical portrait parameters, through the confidence fusion of portrait data, obtain user portrait update data; The high-frequency knowledge recognition is performed on the user portrait update data to determine the user dimension knowledge vector; Bind the user dimension knowledge vector and user ID, and in the case of new session under the user ID, through the vector similarity retrieval of stage state persistence, determine the user interaction reuse state data.

4. The memory hierarchy-based agent interaction memory coordination method according to claim 1, wherein, According to the user interaction reuse state data, through the high-frequency reuse knowledge precipitation, obtain the iteratively shared knowledge base, specifically includes: The high-value answer screening is performed on the user interaction reuse state data to obtain the candidate high-quality question and answer pair to be warehoused; The feedback difference analysis is performed on the user interaction reuse state data to obtain the new knowledge question and answer pair to be warehoused; Based on the candidate high-quality question and answer pair to be warehoused and the new knowledge question and answer pair to be warehoused, through multi-level classification index, determine the multi-level index knowledge question and answer pair; According to the multi-level index knowledge question and answer pair, through the use effect monitoring, obtain the iteratively shared knowledge base.

5. The memory hierarchy-based agent interaction memory coordination method according to claim 1, wherein, Data interaction control is performed on the interaction state variables, the user interaction reuse state data and the iteratively shared knowledge base to determine the memory data directional calling state, specifically includes: Data linkage control is performed on the interaction state variable, the user interaction multiplexing state data and the iterable shared knowledge base to determine a data linkage state; Based on the data linkage state, a user data scheduling state is obtained through data flow scheduling; According to the user data scheduling state, a memory data directional calling state is determined through rule priority determination.

6. The memory hierarchy-based agent interaction memory coordination method according to claim 1, wherein, Based on the memory data directional calling state, an agent interaction memory collaborative storage data is obtained through multi-level cache matching of agent demand analysis, specifically including: User-driven feedback is performed on the memory data directional calling state to obtain knowledge sedimentation data; The knowledge sedimentation data is deposited into the iterable shared knowledge base and stored through a distributed database to determine the agent interaction memory collaborative storage data.

7. The memory hierarchy-based agent interaction memory coordination method according to claim 1, wherein, The agent interaction memory collaborative storage data is determined through distributed database storage, specifically including: Temporary cache requirements for short-term memory are stored through an in-memory database; User-bound persistent requirements for medium-term memory are stored through a combination of a relational database and a vector database; Knowledge sedimentation requirements for long-term memory are stored through a combination of a distributed database and a knowledge base system.

8. The memory hierarchy-based agent interaction memory coordination method according to claim 1, wherein, After obtaining the agent interaction memory collaborative storage data through multi-level cache matching of agent demand analysis based on the memory data directional calling state, the method further includes: The agent interaction memory collaborative storage data is knowledge preloaded to obtain pre-pushed knowledge data.

9. An agent interaction memory coordination device based on memory hierarchy, characterized in that, The device includes: At least one processor; and a memory connected in communication with the at least one processor; Wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: Obtain user original interaction data and perform context interaction analysis on the original interaction data to obtain an interaction state variable; Based on the interaction state variable, user interaction multiplexing state data is determined through cross-session multiplexing of a user unique identifier; According to the user interaction multiplexing state data, an iterable shared knowledge base is obtained through high-frequency multiplexing knowledge sedimentation; Data interaction control is performed on the interaction state variable, the user interaction multiplexing state data and the iterable shared knowledge base to determine a memory data directional calling state; Based on the memory data directional calling state, an agent interaction memory collaborative storage data is obtained through multi-level cache matching of agent demand analysis.

10. A non-transitory computer storage medium storing computer-executable instructions for memory hierarchy based agent interaction memory coordination, the computer-executable instructions comprising: The computer executable instructions are set to: Obtain user original interaction data and perform context interaction analysis on the original interaction data to obtain an interaction state variable; Based on the interaction state variable, user interaction multiplexing state data is determined through cross-session multiplexing of a user unique identifier; According to the user interaction multiplexing state data, an iterable shared knowledge base is obtained through high-frequency multiplexing knowledge sedimentation; Data interaction control is performed on the interaction state variable, the user interaction multiplexing state data and the iterable shared knowledge base to determine a memory data directional calling state; Based on the memory data oriented call state, through multi-level cache matching of agent demand analysis, the agent interactive memory collaborative storage data is obtained.

Images