Information processing method and system in AI dialogue system and electronic equipment

Abstract

The embodiment of the invention discloses an information processing method and system in an AI dialogue system and electronic equipment, and the method comprises the steps: obtaining user archive information through a memory system database associated with the dialogue system after dialogue content inputted by a user is received; adding the user archive information to prompt information, and providing the prompt information to an AI model, so that the AI model generates personalized reply content by referring to the user archive information; the personalized reply content generated by the AI model is provided for a client side so as to be output through a dialogue interface of the client side, a reference information card is displayed, and the reference information card is used for displaying user archive information referenced by the AI model in the process of generating the reply content; and if the dialogue content which is input by the user and used for updating the user archive information is received, updating the user archive information in the database. According to the embodiment of the invention, the conversation effect can be improved, and the trust of the user to the AI system is enhanced.

Description

Technical Field

[0001] This application relates to the field of information processing technology, and in particular to information processing methods, systems and electronic devices in AI dialogue systems. Background Technology

[0002] In recent years, AI (Artificial Intelligence) models have made significant progress in dialogue systems, especially "text-to-text" language models. As powerful "text-chaining" models, AI models can play specific roles (e.g., customer service, technical support) through pre-designed prompts and generate reasonable and coherent responses given a dialogue (including the currently received dialogue content and the dialogue history as context).

[0003] However, applying AI models to the aforementioned dialogue systems still presents some challenges. For example, dialogue models treat all users equally, lacking long-term memory of specific user identities, preferences, and histories, thus failing to provide truly personalized services. Each user is seen as indistinguishable by the model, resulting in untargeted responses and difficulty in meeting users' expectations for a customized experience.

[0004] To address the aforementioned issues, one technological approach involves pre-establishing user profile data, i.e., creating profiles for users in advance (such as through completed questionnaires or imported data), which the dialogue model then reads during the generation of responses. This method achieves personalization to some extent; however, this user profile data is typically maintained manually and is difficult to update in real time, resulting in insufficient personalization. Summary of the Invention

[0005] This application provides information processing methods, systems, and electronic devices in AI dialogue systems, which can improve dialogue effectiveness while enhancing user trust in AI systems.

[0006] This application provides the following solution: An information processing method in an AI dialogue system includes: After receiving the dialogue content input by the user, the system retrieves user profile information through the memory system database associated with the dialogue system. The user profile information is extracted from the dialogue content and dynamically updated into the database by the AI ​​model during the historical dialogue with the user. The user profile information is added to the prompt information and provided to the AI ​​model so that the AI ​​model can generate personalized response content by referring to the user profile information; The personalized response content generated by the AI ​​model is provided to the client for output through the client's dialogue interface, and a reference information card is displayed. The reference information card is used to display the user profile information referenced by the AI ​​model in the process of generating the response content. If the dialogue content input by the user for updating the user profile information is received, the user profile information in the database is updated.

[0007] During the dialogue between the AI ​​model and the user, the historical dialogue content is saved to the database of the memory system. The memory system includes a short-term memory cache and a long-term memory bank. The short-term memory cache is used to cache the dialogue content of the most recent preset number of rounds, and the long-term memory bank is used to persistently store the dialogue content before the preset number of rounds. The long-term memory bank is also used to persistently store the user profile information. The method further includes: After receiving the dialogue content input by the user, the intent is identified based on the dialogue content; If the identified user intent is related to historical dialogue content, the relevant historical dialogue content is retrieved from the memory system and input into the AI ​​model through the prompt information, so that the AI ​​model can generate response content by referring to the relevant historical dialogue content.

[0008] This also includes: The reference information card also includes historical dialogue information referenced by the AI ​​model when generating the response content.

[0009] Specifically, when saving historical dialogue content to the long-term memory, the original text of the historical dialogue content is saved to the original text data table in the long-term memory, so that the AI ​​model can generate response content by referring to the original text of the relevant historical dialogue content.

[0010] In the process of saving historical dialogue content to the long-term memory, summary content is also generated for topics that may be used repeatedly and saved to the summary data table in the long-term memory. The summary content is generated by summarizing or summarizing the original text of multiple dialogue contents related to the same topic. The retrieval of relevant historical dialogue content from the memory system includes: Retrieve summaries of relevant historical dialogue content from the summary data table and input them into the AI ​​model.

[0011] This also includes: After retrieving the summary content, it is determined whether it is necessary to refer to the original text of the historical dialogue. If so, the relevant original text of the historical dialogue is retrieved from the original text data table for the AI ​​model to refer to when generating the response content.

[0012] Specifically, the orchestration module retrieves user profile information from the long-term memory, performs intent recognition on the user's input dialogue content, determines decision logic based on the intent recognition results, and performs process control; the decision logic is used to determine the retrieval method for multiple types of data tables in the memory system.

[0013] The decision logic includes: whether to retrieve historical dialogue content from the short-term cache or long-term memory; if it is necessary to retrieve the long-term memory, then further determine whether to obtain a summary or the original text of the relevant historical dialogue content; so as to perform process control according to the decision logic. During the process control, the original text data table and / or summary data table in the short-term cache and long-term memory are retrieved to obtain the dialogue content within the most recent preset number of rounds, the summary and / or the original text of the dialogue content before the preset number of rounds. Then, a prompt message is generated in combination with the user profile information and input into the AI ​​model.

[0014] This also includes: The orchestration module determines whether additional information outside the dialogue system is needed based on the identified user intent. If so, it obtains the additional information by calling relevant tools and provides it to the AI ​​model for reference.

[0015] An information processing method in an AI dialogue system includes: The system receives dialogue content input by the user and submits it to the server. The server then retrieves user profile information from the memory system database associated with the dialogue system, adds it to the prompt information, and provides it to the AI ​​model. The user profile information is extracted from the dialogue content and dynamically updated to the database by the AI ​​model during the historical dialogue between the AI ​​model and the user. The AI ​​model is used to generate personalized response content by referring to the user profile information. The personalized response content is output through the client's dialogue interface, and a reference information card is displayed. The reference information card is used to display the user profile information referenced by the AI ​​model in the process of generating the response content. If a user input is received requesting an update to their user profile information, it is submitted to the server for updating the user profile information in the database.

[0016] An information processing method in an AI dialogue system includes: During the process of engaging in historical dialogues with users through an AI model, the dialogue content of the most recent preset number of rounds is added to the short-term memory cache of the AI ​​model, the dialogue content before the preset number of rounds is persistently stored in the long-term memory outside the AI ​​model, and user profile information is extracted from the dialogue content, persistently stored in the long-term memory and dynamically updated. After receiving the current dialogue content input by the user, the user profile information is obtained from the long-term memory, and intent is identified based on the current dialogue content. If the identified user intent is related to historical dialogue content, the relevant historical dialogue content is retrieved from the short-term memory cache and / or the long-term memory. Based on the current conversation content, user profile information, and relevant historical conversation content, a prompt message is generated, and the AI ​​model is invoked based on the prompt message so that the AI ​​model can generate personalized response content.

[0017] This also includes: The personalized response content generated by the AI ​​model is provided to the client for output through the client's dialogue interface, and a reference information card is displayed. The reference information card is used to display the user profile information referenced by the AI ​​model in the process of generating the response content. If the dialogue content input by the user for updating the user profile information is received, the user profile information in the database is updated.

[0018] An information processing system for an AI dialogue system includes: The memory module includes a short-term memory cache and a long-term memory database. The short-term memory cache is used to cache the dialogue content of the most recent preset number of rounds. The long-term memory database includes a raw text data table, a summary data table, and a user profile data table. The raw text data table is used to persistently store the raw text of the dialogue content up to the preset number of rounds. The summary data table is used to persistently store the summary content generated by summarizing or summarizing multiple dialogue contents related to topics that may be used repeatedly. The user profile data table is used to persistently store user profile information. The user profile information is extracted from the dialogue content and dynamically updated into the user profile data table by the AI ​​model during historical dialogues with the user. The orchestration module, upon receiving dialogue input from the user, retrieves user profile information from the user profile data table in the long-term memory and performs intent recognition on the user's input dialogue content. Based on the intent recognition result, it determines decision logic, which includes: determining whether it is necessary to search the short-term cache or long-term memory to obtain historical dialogue content; if it is necessary to search the long-term memory, it further determines whether it is necessary to obtain a summary or original text of the relevant historical dialogue content. Based on the decision logic, it searches the original text data table and / or summary data table in the short-term cache and long-term memory to obtain the dialogue content within the most recent preset number of rounds, and the summary and / or original text of the dialogue content before the preset number of rounds. Then, it generates prompt information in conjunction with the user profile information and inputs it into the AI ​​model, enabling the AI ​​model to generate personalized response content by referring to the user profile information and the relevant historical dialogue content.

[0019] The orchestration layer is further configured to output the personalized response content through the client's dialogue interface and display reference information cards, which are used to display the user profile information referenced by the AI ​​model in generating the response content; wherein, if a dialogue content input by the user is received to update the user profile information, the user profile information in the user profile data table is updated.

[0020] The orchestration module is also used to determine whether additional information outside the dialogue system is needed based on the identified user intent. If so, it obtains the additional information by calling relevant tools and provides it to the AI ​​model for reference.

[0021] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps of any of the preceding methods.

[0022] An electronic device, comprising: One or more processors; and A memory associated with the one or more processors, the memory being used to store program instructions that, when read and executed by the one or more processors, perform the steps of any of the preceding methods.

[0023] A computer program product includes a computer program / computer executable instructions that, when executed by a processor in an electronic device, implement the steps of any of the preceding methods.

[0024] According to the specific embodiments provided in this application, the following technical effects are disclosed: Through the embodiments of this application, during historical dialogues between an AI model and a user, the AI ​​model can extract user profile information from the dialogue content and dynamically update it in the memory system database. Thus, upon receiving dialogue input from the user, the AI ​​model can retrieve user profile information from the memory system database, add it to the prompt information, and provide it to the AI ​​model. The AI ​​model can then generate personalized responses by referring to the user profile information. The personalized responses generated by the AI ​​model can then be provided to the client for output through the client's dialogue interface. Reference information cards can also be displayed, showing the user profile information referenced by the AI ​​model during the generation of the responses. Furthermore, if dialogue input from the user is received to update the user profile information, the user profile information in the database can also be updated. In this way, the construction and updating of user profile information can be jointly completed by the AI ​​model and the user. By automatically extracting and saving user profile information during the dialogue process, the reliance on manual construction and updating of user profiles can be reduced, and dynamic updates of user profile information can be achieved. Furthermore, direct user participation can improve the accuracy of user profile information in the memory system and prevent the model from making inappropriate responses based on erroneous memories. This human-machine collaborative memory management not only enhances the effectiveness of dialogue but also strengthens users' trust in the AI ​​system.

[0025] In other embodiments, historical dialogue content can be stored through a memory system, employing a hierarchical design of "short-term caching + long-term indexing." User profile data can also be stored in a long-term memory repository. Furthermore, an orchestration layer can be used for overall scheduling, combining the long and short-term memory support provided by the memory system. The AI ​​model handles language processing, supplemented by a user-interactive memory interface. Optionally, external capabilities can be extended through a toolset, forming a complete memory-enhanced dialogue agent. This approach combines a hierarchical memory architecture of dynamic user profiles, reference information cards, and historical dialogue content, not only solving the problem of long-conversation amnesia but also improving the credibility of AI responses and user satisfaction through a user-verifiable memory mechanism. This synergistic effect is not simply a superposition of various technical means. Moreover, this approach provides a comprehensive memory-enhanced dialogue model design that can simultaneously address the personalization of response content and the "amnesia" problem of long conversations through a single mechanism. This allows the AI ​​model to possess a human-like memory system, maintaining contextual coherence in long conversations, "recognizing" each user, providing personalized services, and ensuring the accuracy and controllability of memory.

[0026] Of course, any product implementing this application does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a schematic diagram of the system architecture provided in the embodiments of this application; Figure 2 This is a flowchart of the first method provided in the embodiments of this application; Figure 3 This is a schematic diagram of the first interface provided in an embodiment of this application; Figure 4 This is a schematic diagram of the second interface provided in an embodiment of this application; Figure 5 This is a flowchart of the second method provided in the embodiments of this application; Figure 6 This is a flowchart of the third method provided in the embodiments of this application; Figure 7 This is a schematic diagram of the system provided in the embodiments of this application; Figure 8 This is a schematic diagram of the electronic device provided in the embodiments of this application. Detailed Implementation

[0029] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0030] In this embodiment, a solution is provided to better reflect the personalization of different user profiles in the AI ​​model's responses. This solution utilizes Dialogue State Tracking (DST) technology to dynamically build and update user profiles. Specifically, during each conversation, prompts can be used to tell the AI ​​model which information to extract and store separately in the user profile data table. This allows the AI ​​model to analyze user speech, automatically extract key information (such as the user's occupation, hometown, preferences, or the type of product being inquired about), and add this information to the corresponding user's profile information in a structured manner. The specific user profile information can be stored in a database in vector form for rapid retrieval. When the same user starts a new conversation, the system first retrieves the user's latest profile information from the profile database based on the user's identifier, providing this as background knowledge to the AI ​​model. The AI ​​model references the user profile when generating responses, adjusting tone and content to match user preferences. For example, if the profile indicates the user is a technical expert, the model will use more professional terminology; if the user prefers concise answers, the model's response will be more succinct. Through user profiles, the system achieves "long-term memory" of user information, truly enabling personalized dialogue for each user.

[0031] In addition, during the display of responses generated by the AI ​​model, reference information cards can be shown. These cards display user profile information referenced by the AI ​​model in generating the responses. By displaying these cards, users can more intuitively understand which information the AI ​​model consulted during response generation, thus increasing the transparency and explainability of the dialogue and enhancing user trust in the model's responses. Furthermore, by visually viewing this user profile information, users can assess or monitor its accuracy. If inaccuracies are found, users can interact with the AI ​​model based on the card, for example, by directly conversing with the AI ​​model to modify the profile information. However, since the specific user profile information is extracted and saved by the AI ​​model during the conversation with the user, there may be errors in information extraction, or some information may not be extracted in a timely manner. For example, a user's occupation may have recently changed and not yet mentioned in the conversation with the AI ​​model, in which case the AI ​​model may not be able to update the occupation information in the profile in a timely manner, and so on. Therefore, by displaying the user profile information currently referenced by the AI ​​model to the user, the user can promptly identify specific problems and make timely modifications through dialogue with the AI ​​model. The system will update the user's corrected information in the database to ensure more accurate subsequent responses.

[0032] As can be seen, the above methods enable the construction and updating of user profile information to be jointly completed by the AI ​​model and the user, for use in subsequent conversations. By automatically extracting and saving user profile information during the conversation, the reliance on manual construction and updating of user profiles can be reduced, and dynamic updates of user profile information can be achieved. Furthermore, direct user participation can improve the accuracy of user profile information in the memory system and prevent the model from making inappropriate responses based on incorrect memories. This human-machine collaborative memory management not only improves the effectiveness of the conversation but also enhances the user's trust in the AI ​​system.

[0033] In addition to the issue of personalized responses, dialogue systems using AI models may also suffer from "amnesia" problems in long conversations. While existing AI models employ the concept of a context window—where user-provided information is converted into tokens and populated into a window, and the model uses attention mechanisms to focus on all information within that window and generate responses—this context window is typically limited to a fixed length. When there are many dialogue rounds or the content is too long, exceeding the window's length, earlier information may be pushed out, causing the model to "forget" earlier dialogue content. This can result in incoherent responses, broken communication experiences, and difficulty maintaining consistency across multiple dialogue rounds, hindering in-depth interaction. While requiring users to re-enter previously input information can be helpful, this negatively impacts the user experience.

[0034] To address the aforementioned issues, Retrieval-augmented Generation (RAG) technology can be used to store dialogue history in an external knowledge base. Relevant fragments can be retrieved and provided to the model when needed, thereby expanding the model's access to context to some extent. RAG is a technical architecture that combines information retrieval with AI models. Its core idea is that when an AI model needs to answer a question or generate text, it first retrieves information related to the question from an external, trusted knowledge base. This information, along with the original question, is then input into the AI ​​model as context, allowing the model to generate a more accurate answer. Based on these characteristics, historical dialogue content can be saved in this external knowledge base, enabling the AI ​​model to retrieve the necessary historical dialogue content during the dialogue process, thus assisting in generating a more accurate answer. In this embodiment, the aforementioned RAG technology can also be used to save historical dialogue content in the form of an external knowledge base for the AI ​​model, for subsequent use in the answer generation process. When saving historical dialogue content, it can be segmented into smaller, semantically coherent fragments (also called "chunks," and retrieval is typically performed at the "chunk" level). Then, an embedding model can be used to convert each text chunk into a high-dimensional numerical vector, which captures the semantic information of the text chunk. These vectors and their corresponding original text chunks can then be stored in a vector database. During retrieval, the user's question can be converted into a vector, and a similarity search can be performed in the vector database to find several text chunks similar to the question vector. These relevant text chunks are then combined with the original question according to a pre-designed prompt word template to generate prompt words, which are then input into the AI ​​model.

[0035] The above solution can resolve the "amnesia" problem in long conversations. Without special handling, the personalization of the aforementioned response content and the "amnesia" problem in long conversations are two independent issues, requiring separate solutions. However, in the preferred embodiment of this application, a unified framework can be provided to simultaneously address both issues. In other words, a comprehensive memory-enhanced dialogue model design can be offered, enabling the AI ​​model to possess a human-like memory system, maintain contextual coherence in long conversations, "recognize" each user, provide personalized services, and simultaneously ensure the accuracy and controllability of memory.

[0036] For details, see Figure 1 The dialogue system provided in this application embodiment can incorporate a dynamically hierarchical memory module and a user-interactive memory interface. Specifically, the dynamically hierarchical memory module can provide both short-term and long-term memory capabilities. The memory system can be constructed based on the vector database in the aforementioned RAG technology, and the long-term memory portion can be divided into two sub-modules: "conversation memory" and "user profile."

[0037] The conversation memory component stores the history of dialogues, including multi-turn interactions between the user and the AI ​​model. To balance efficiency and effectiveness, conversation memory employs a hierarchical design of "short-term cache + long-term index." The short-term cache stores the most recent few rounds of dialogue as working memory (e.g., the most recent one or two rounds), allowing the model to directly incorporate them into the context window. While this caching method has limited capacity, it offers fast read and write speeds; however, data may be lost after the dialogue ends. The long-term index encodes the complete dialogue history as a vector and stores it in a database as a long-term memory. As the number of dialogue rounds increases, new content is continuously added to the cache, while older content is represented by vectors and stored in the long-term database. This mechanism is similar to how humans retrieve long-term memories using the hippocampus, enabling the model to accurately reference previous information even in long dialogues.

[0038] In long-term memory, on the one hand, the original text of historical dialogues can be directly stored to avoid loss of details; on the other hand, for topics that may be reused, summaries can be generated and stored in long-term memory. Specifically, summaries can be generated by summarizing or abstracting multiple dialogues related to the same topic. For example, if historical dialogues mention a particular order, without summarizing or abstracting, information about the order's amount, requests, and problems might be scattered across multiple dialogues. However, by summarizing and abstracting, the information from these multiple dialogues related to the same order can be summarized, perhaps generating a single record for that order. During retrieval, this summary can be directly retrieved, providing quick access to comprehensive information about the order and improving efficiency.

[0039] In this embodiment, the generation of summary content is typically only necessary when required. That is, only important information that may be repeatedly used later is summarized, while other dialogue content is simply saved in its original form. Of course, even if summaries and conclusions are generated for certain dialogue content, the original text of these dialogues is also stored separately in long-term memory. During the AI ​​model's response generation process, if the summary content is provided to the AI ​​model for reference, but the AI ​​model deems it necessary to refer to details of the dialogue content, the corresponding original dialogue content can be further retrieved and provided to the AI ​​model. The original text and summary content of historical dialogues can be stored in different data tables within the long-term memory.

[0040] Additionally, user profile information can also be stored in another data table in the aforementioned long-term memory. In other words, during dialogues with users via AI models, on the one hand, the most recent rounds of conversation can be added to a short-term cache, while the original text of earlier conversations is saved in the original text table of the long-term memory. For more important conversation content that may be repeatedly used later, summaries or conclusions can be generated and saved in a summary database. On the other hand, the AI ​​model can also extract user profile information from the user's conversation content and save or update it in the user profile data table of the long-term memory.

[0041] Building upon the aforementioned layered design of "short-term caching + long-term indexing," when generating a response based on the user's current input, the orchestration layer first parses the user's input intent and determines which components to invoke to handle the current request. Specifically, user profile information can be retrieved from the long-term memory. Furthermore, based on the intent identified from the user's input, it can be determined whether historical dialogue content needs to be retrieved. If so, the short-term cache, the original text data table, and the summary data table in the long-term memory can be searched to obtain the relevant historical dialogue content's summary and / or original text. This summary, along with the user profile information, generates a prompt message to be input into the AI ​​model. It should be noted that in some cases, additional information outside the current dialogue system, such as weather conditions, may also be needed. This additional information can be obtained by calling relevant tools and added to the aforementioned prompt message before being input into the AI ​​model.

[0042] Then, the AI ​​model generates the response by referencing multiple sources of information. When displaying the generated response to the user, a reference information card can be provided, showing the information the AI ​​model referenced in generating the response. This information may include user profile information, or, optionally, referenced historical conversations, etc. If the user believes that any part of the user profile information displayed in the card needs modification, they can initiate the modification by directly interacting with the AI ​​model. Afterwards, the user's profile information in the long-term memory is updated to generate more accurate responses in subsequent conversations.

[0043] The underlying AI model is provided by the capability provider, and different AI models can be selected as the backend based on application requirements. The AI ​​model service can be accessed through API or local deployment, and is responsible for the actual language understanding and generation tasks. In this embodiment, the AI ​​model receives comprehensive prompts from the orchestration layer (including the user's current question, short dialogue context, retrieved historical dialogue content, user profile information, etc.) and generates response content accordingly. The AI ​​service can be a pluggable component, supporting model replacement or upgrades without affecting the overall architecture. The AI ​​model service can also be fine-tuned for specific tasks or have plug-in tools added to enhance its capabilities.

[0044] A toolkit is used to provide external functional interfaces, extending the capabilities of a dialogue system. The toolkit can connect to the orchestration layer via an API gateway. When the AI ​​model or the orchestration layer determines that an external function needs to be invoked, the toolkit can perform the corresponding operation and return the result. The toolkit can contain various types of tools, such as: search engine interfaces for real-time information queries, database query interfaces for accessing internal enterprise knowledge bases, calculator or function execution interfaces for numerical calculations or code execution, and application function interfaces such as weather queries, scheduling, and email sending. In this embodiment, the introduction of the toolkit enables the AI ​​to "act," allowing it to complete tasks beyond dialogue. The orchestration layer can determine whether to invoke a tool based on user requests. For example, when a user asks "What will the weather be like tomorrow?", the orchestration layer will invoke a weather query tool to obtain the result, which will then be passed to the AI ​​model to generate a natural language response.

[0045] In summary, the orchestration layer primarily relies on user intent recognition to determine which components to invoke to handle the current request based on the specific intent recognition results. Specifically, after receiving a user's question, the orchestration layer can decide whether to directly pass the question to the AI ​​model for an answer, first query the internal memory system for relevant context, or call external tools to obtain additional information, and so on. From an architectural perspective, the orchestration layer can implement the above decision-making logic and process control through workflow engines and other methods, ensuring that the dialogue process executes according to preset strategies. For example, when it detects that a user's question involves previously discussed content, the orchestration layer can trigger a memory retrieval process; when a user inquires about real-time information, it can invoke the corresponding tool interface. Furthermore, the orchestration layer can also be responsible for integrating the outputs of various modules to generate a final response and send it to the user.

[0046] As can be seen, in the above system architecture, through the overall scheduling of the orchestration layer, the memory system provides long and short term memory support, the AI ​​model is responsible for language processing, the toolset expands external capabilities, and is supplemented by a user-interactive memory interface, forming a complete memory-enhanced conversational intelligent agent.

[0047] The specific implementation schemes provided in the embodiments of this application will be described in detail below.

[0048] Example 1 First, from the server's perspective, this first embodiment provides an information processing method in a dialogue system, see [link to previous document]. Figure 2 The method may include: S201: After receiving the dialogue content input by the user, the user profile information is obtained through the memory system database associated with the dialogue system. The user profile information is extracted from the dialogue content and dynamically updated into the database by the AI ​​model during the historical dialogue with the user through the artificial intelligence AI model.

[0049] Specifically, the dialogue content input by the user can vary depending on the application of the dialogue system. For example, in a customer service system within a product information service system, the user's input is typically related to products, orders, and after-sales service. After receiving the specific user dialogue content, the system's orchestration layer can retrieve user profile information from the memory system database. As mentioned earlier, in this embodiment, the specific user profile information can be extracted from the dialogue content and dynamically updated to the database by the AI ​​model during historical dialogues with the user. In other words, the specific user profile information does not require manual maintenance but can be automatically extracted and dynamically updated by the AI ​​model during dialogues with the user. For example, if the user inputs "I am an engineer," the user's profession can be extracted as "engineer"; or, if the user expresses "please keep it simple," the user's preference can be extracted as "prefers concise answers," and so on. The user profile information is dynamic; as the AI ​​model engages in more dialogues with the user, it may extract richer user profile information and modify previously saved information.

[0050] S202: Add the user profile information to the prompt information and provide it to the AI ​​model so that the AI ​​model can generate personalized response content by referring to the user profile information.

[0051] After obtaining user profile information, this information can be added to the prompts so that the AI ​​model can generate personalized responses by referring to the user profile. In other words, when generating responses, the AI ​​model can refer to the user profile information to adjust the tone and content to match the user's preferences. For example, if the profile shows the user is a technical expert, the AI ​​model can use more professional terminology in its response; if the user prefers concise answers, the AI ​​model will respond more succinctly, and so on. In this way, "long-term memory" of user information can be achieved, truly realizing personalized dialogue tailored to each individual.

[0052] Furthermore, as mentioned earlier, during the dialogue between the AI ​​model and the user, historical dialogue content can be saved to the database of the memory system. This memory system includes a short-term memory cache and a long-term memory. The short-term memory cache caches dialogue content from the most recent preset number of rounds, while the long-term memory persistently stores dialogue content from rounds prior to the preset number of rounds. User profile information can also be persistently stored in this long-term memory. In this case, after receiving dialogue input from the user, intent recognition can be performed based on the dialogue content. If the recognized user intent is related to historical dialogue content, relevant historical dialogue content can be retrieved from the memory system and input into the AI ​​model via the prompt information, allowing the AI ​​model to generate response content by referring to the relevant historical dialogue content. For example, if the user's current dialogue input mentions "an order consulted yesterday," it can be recognized that the user's intent is to further inquire about an order consulted yesterday, and dialogue content related to that order can be retrieved from the memory system. Specifically, a search can first be performed in the short-term memory cache; if no dialogue content related to that order exists, relevant dialogue content can be retrieved from the long-term memory.

[0053] When saving historical dialogue content to the long-term memory, the original text of the dialogue should be saved as much as possible in the original text data table of the long-term memory. This allows the AI ​​model to generate responses by referring to the original text of relevant historical dialogues, avoiding the loss of details. Additionally, when saving historical dialogue content to the long-term memory, summaries of frequently used topics can be generated and saved in the summary data table of the long-term memory. These summaries can be generated by summarizing or abstracting multiple original texts of dialogues related to the same topic. Thus, when searching the long-term memory, relevant summaries of historical dialogues can be retrieved from the summary data table, and if found, can be input into the AI ​​model. Furthermore, if the AI ​​model needs to refer to the original text of historical dialogues, it can retrieve the relevant original text from the original text data table for reference when generating responses.

[0054] Based on the aforementioned layered design of "short-term cache + long-term index," as mentioned earlier, the orchestration module can also retrieve user profile information from the long-term memory and perform intent recognition on the user's input dialogue content. Based on the intent recognition results, the system can coordinate and schedule the process to determine decision logic and control the flow. In one implementation, the decision logic specifically includes: whether to retrieve historical dialogue content from the short-term cache or long-term memory; if a retrieval of the long-term memory is required, further determining whether to obtain a summary or the original text of the relevant historical dialogue content. This allows for flow control based on the decision logic. During flow control, the system retrieves the original text data table and / or summary data table in the short-term cache and long-term memory to obtain the dialogue content within the most recent preset number of rounds, and the summary and / or original text of the dialogue content before the preset number of rounds. Then, it combines this with the user profile information to generate prompt information for input into the AI ​​model.

[0055] As described in the steps above, in the preferred implementation, relevant historical dialogue content can also be retrieved, and additional information can be obtained through tools. This information can also be added to the prompt information and provided to the AI ​​model so that the AI ​​model can generate response content by integrating this information.

[0056] S203: Provide the personalized response content generated by the AI ​​model to the client so that it can be output through the client's dialogue interface and a reference information card can be displayed. The reference information card is used to display the user profile information referenced by the AI ​​model in the process of generating the response content. If a dialogue content input by the user is received to update the user profile information, the user profile information in the database is updated.

[0057] After the AI ​​model generates the response content, a client application can be provided for display on the client interface. Additionally, in this embodiment, the user profile information referenced by the AI ​​model can be provided to the client, so that when displaying the response content, the client can also display a reference information card, which can show the user profile information referenced by the AI ​​model in generating the response content. For example, such as... Figure 3 As shown, assuming a user is having a conversation with AI, and the AI ​​model generates a response, then the 31 reference information cards shown below can be displayed after the response. Figure 3The cards displayed only show user profile information and can therefore be called "user identity information cards." These cards can include user identity information and other relevant profile items based on the specific identity. For example, if the current user is an internal staff member (e.g., an "operations assistant"), the profile items could include "business type," "responsible business entity," "country," etc. If the current user is a regular buyer, the specific profile items will differ, such as including the user's occupation, gender, etc. This approach increases the transparency and explainability of the conversation, enhancing user trust in the model-generated responses.

[0058] In addition, users can interact with the AI ​​model based on the user profile information displayed in the specific reference information cards. For example, if they find an error in a piece of information in the profile, or if they need to supplement certain information, they can directly interact with the AI ​​model in the dialog window. For instance, if a user finds that the identity information displayed in the reference information card is incorrect, they can... Figure 4 As shown, a user can enter: "Hello, I'm the operations manager responsible for overseas hosting - cross-border entities." The AI ​​model can then generate a response, such as, "Based on the information you entered, we have detected the following changes: User information change detected. The changes are as follows: Business Module: Overseas Hosting; Business Entity: Cross-border Entity (Non-EU, Non-Payment Term Country)." Additionally, the system can display the changed information to the user and ask for confirmation. For example, if the user clicks... Figure 4 The "Confirm Save" option shown will update and save the user's profile data in the database. In summary, through Embodiment 1 of this application, during historical dialogues between an AI model and a user, the AI ​​model can extract user profile information from the dialogue content and dynamically update it in the memory system database. Thus, upon receiving dialogue input from the user, the AI ​​model can retrieve user profile information from the memory system database, add it to the prompt information, and provide it to the AI ​​model. The AI ​​model can then generate personalized responses by referring to the user profile information. The personalized responses generated by the AI ​​model can then be provided to the client for output through the client's dialogue interface. Furthermore, a reference information card can be displayed, showing the user profile information referenced by the AI ​​model during the generation of the responses. Additionally, if dialogue input from the user is received to update the user profile information, the user profile information in the database can also be updated. In this way, the construction and updating of user profile information can be jointly completed by the AI ​​model and the user. By automatically extracting and saving user profile information during the dialogue process, the reliance on manual construction and updating of user profiles can be reduced, and dynamic updates of user profile information can be achieved. Furthermore, direct user participation can improve the accuracy of user profile information in the memory system and prevent the model from making inappropriate responses based on erroneous memories. This human-machine collaborative memory management not only enhances the effectiveness of dialogue but also strengthens users' trust in the AI ​​system.

[0059] In other embodiments, historical dialogue content can be stored through a memory system, employing a hierarchical design of "short-term caching + long-term indexing." User profile data can also be stored in a long-term memory repository. Furthermore, an orchestration layer can be used for overall scheduling, combining the long and short-term memory support provided by the memory system. The AI ​​model handles language processing, supplemented by a user-interactive memory interface. Optionally, external capabilities can be extended through a toolset, forming a complete memory-enhanced dialogue agent. This approach combines a hierarchical memory architecture of dynamic user profiles, reference information cards, and historical dialogue content, not only solving the problem of long-conversation amnesia but also improving the credibility of AI responses and user satisfaction through a user-verifiable memory mechanism. This synergistic effect is not simply a superposition of various technical means. Moreover, this approach provides a comprehensive memory-enhanced dialogue model design that can simultaneously address the personalization of response content and the "amnesia" problem of long conversations through a single mechanism. This allows the AI ​​model to possess a human-like memory system, maintaining contextual coherence in long conversations, "recognizing" each user, providing personalized services, and ensuring the accuracy and controllability of memory.

[0060] Example 2 This second embodiment corresponds to the first embodiment and, from the client's perspective, provides an information processing method in an AI dialogue system. (See [link to previous embodiment]). Figure 5 The method may include: S501: Receive the dialogue content input by the user and submit it to the server so that the server can obtain the user profile information through the memory system database associated with the dialogue system and add it to the prompt information, and then provide it to the AI ​​model. The user profile information is extracted from the dialogue content and dynamically updated to the database by the AI ​​model during the process of conducting historical dialogues with the user. The AI ​​model is used to generate personalized response content by referring to the user profile information. S502: Output the personalized reply content through the client's dialogue interface and display a reference information card, which is used to display the user profile information referenced by the AI ​​model in the process of generating the reply content; S503: If a user input is received to update user profile information, it is submitted to the server so that the user profile information in the database can be updated.

[0061] Example 3 In this third embodiment, a method for information processing in an AI dialogue system is provided, primarily addressing the "comprehensive memory-enhanced dialogue model" mentioned earlier. (See [link to previous document]). Figure 6 The method may specifically include: S601: During the process of engaging in historical dialogues with users through an AI model, the dialogue content of the most recent preset number of rounds is added to the short-term memory cache of the AI ​​model, the dialogue content before the preset number of rounds is persistently stored in the long-term memory outside the AI ​​model, and user profile information is extracted from the dialogue content, persistently stored in the long-term memory and dynamically updated. S602: After receiving the current dialogue content input by the user, the user profile information is obtained from the long-term memory, and intent is identified based on the current dialogue content. If the identified user intent is related to historical dialogue content, the relevant historical dialogue content is retrieved from the short-term memory cache and / or the long-term memory. S603: Generate prompt information based on the current dialogue content, user profile information, and relevant historical dialogue content, and invoke the AI ​​model based on the prompt information so that the AI ​​model generates personalized response content.

[0062] In a preferred implementation, the personalized response content generated by the AI ​​model can also be provided to the client for output through the client's dialogue interface, and a reference information card can be displayed. The reference information card is used to display the user profile information referenced by the AI ​​model in the process of generating the response content. If a dialogue content input by the user is received to update the user profile information, the user profile information in the database is updated.

[0063] Through this third embodiment, overall scheduling can be achieved through an orchestration layer, with the memory system providing long and short-term memory support, and the AI ​​model handling language processing. In a preferred embodiment, a user-interactive memory interface can be added, and external capabilities can be expanded through a toolset, forming a complete memory-enhanced conversational AI agent.

[0064] Example 4 This fourth embodiment also provides an information processing system for an AI dialogue system, see [link to documentation]. Figure 7 The system may specifically include: The memory module 701 includes a short-term memory cache and a long-term memory database. The short-term memory cache is used to cache the dialogue content of the most recent preset number of rounds. The long-term memory database includes a raw text data table, a summary data table, and a user profile data table. The raw text data table is used to persistently store the raw text of the dialogue content before the preset number of rounds. The summary data table is used to persistently store the summary content generated by summarizing or summarizing multiple dialogue contents related to topics that may be used repeatedly. The user profile data table is used to persistently store user profile information. The user profile information is extracted from the dialogue content and dynamically updated into the user profile data table by the AI ​​model during historical dialogues with the user. The orchestration module 702, upon receiving dialogue content input by the user, retrieves user profile information from the user profile data table in the long-term memory, performs intent recognition on the user-input dialogue content, and determines decision logic based on the intent recognition result. The decision logic includes: determining whether it is necessary to search the short-term cache or long-term memory to obtain historical dialogue content; if it is necessary to search the long-term memory, further determining whether it is necessary to obtain a summary or original text of the relevant historical dialogue content, so as to search the original text data table and / or summary data table in the short-term cache and long-term memory according to the decision logic, and obtain the dialogue content within the most recent preset number of rounds, the summary and / or original text of the dialogue content before the preset number of rounds, and then generating prompt information in combination with the user profile information to input into the AI ​​model, so that the AI ​​model can generate personalized response content by referring to the user profile information and the relevant historical dialogue content.

[0065] In a specific implementation, the orchestration layer can also be used to output the personalized reply content through the client's dialogue interface and display reference information cards. The reference information cards are used to display the user profile information referenced by the AI ​​model in the process of generating the reply content. If a dialogue content input by the user is received to update the user profile information, the user profile information in the user profile data table is updated.

[0066] In addition, the orchestration module is also used to determine whether additional information outside the dialogue system is needed based on the identified user intent. If so, the additional information is obtained by calling relevant tools and provided to the AI ​​model for reference.

[0067] Through the above embodiment four, a unified framework can be provided to simultaneously address the two issues mentioned above: the "amnesia" problem in long conversations and the personalization of response content. In other words, a comprehensive memory-enhanced dialogue model design is provided, which enables the AI ​​model to have a memory system similar to that of humans, maintain contextual coherence in long conversations, "recognize" each user, provide personalized services, and at the same time ensure the accuracy and controllability of memory.

[0068] Of course, in the preferred implementation, the orchestration layer can also display reference information cards through the client's dialog interface to show the user profile information referenced by the AI ​​model in the process of generating the response content. The cards are interactive, and users can modify the user profile information in them, so that the above unified framework can also have the effect described in Embodiment 1.

[0069] For the parts not described in detail in Embodiments 2 to 4 above, please refer to Embodiment 1 and other parts of this specification, which will not be repeated here.

[0070] It should be noted that the embodiments of this application may involve the use of user data. In practical applications, user-specific personal data may be used in the scheme described herein within the scope permitted by applicable laws and regulations, provided that it complies with the applicable laws and regulations of the country (e.g., with the user's explicit consent, with the user being properly notified, etc.).

[0071] Corresponding to Embodiment 1, this application also provides an information processing device in an AI dialogue system, which may include: The user profile information acquisition unit is used to acquire user profile information through the memory system database associated with the dialogue system after receiving the dialogue content input by the user. The user profile information is extracted from the dialogue content by the AI ​​model and dynamically updated into the database during the process of conducting historical dialogues with the user through the artificial intelligence AI model. The prompt information generation unit is used to add the user profile information to the prompt information and provide it to the AI ​​model so that the AI ​​model can generate personalized response content by referring to the user profile information; The response content providing unit is used to provide personalized response content generated by the AI ​​model to the client so that it can be output through the client's dialogue interface and a reference information card can be displayed. The reference information card is used to display the user profile information referenced by the AI ​​model in the process of generating the response content. If the dialogue content input by the user for updating the user profile information is received, the user profile information in the database is updated.

[0072] During the dialogue between the AI ​​model and the user, the historical dialogue content is saved to the database of the memory system. The memory system includes a short-term memory cache and a long-term memory bank. The short-term memory cache is used to cache the dialogue content of the most recent preset number of rounds, and the long-term memory bank is used to persistently store the dialogue content before the preset number of rounds. The long-term memory bank is also used to persistently store the user profile information. The device may further include: The user intent recognition unit is used to recognize intent based on the dialogue content input by the user after receiving the dialogue content. The historical dialogue content acquisition unit is used to retrieve relevant historical dialogue content from the memory system if the identified user intent is related to historical dialogue content, and input it into the AI ​​model through the prompt information so that the AI ​​model can generate response content by referring to the relevant historical dialogue content.

[0073] In addition, the reference information card may also include historical dialogue content information referenced by the AI ​​model when generating the response content.

[0074] In addition to saving historical dialogue content to the long-term memory, the original text of the historical dialogue content can also be saved to the original text data table in the long-term memory, so that the AI ​​model can generate response content by referring to the original text of the relevant historical dialogue content.

[0075] In addition, when saving historical dialogue content to the long-term memory, summary content can also be generated for topics that may be used repeatedly and saved to the summary data table in the long-term memory. The summary content is generated by summarizing or summarizing the original text of multiple dialogue contents related to the same topic. At this point, the summary content of the relevant historical dialogue can be retrieved from the summary data table and input into the AI ​​model.

[0076] Alternatively, after retrieving the summary content, it can be determined whether it is necessary to refer to the original text of the historical dialogue. If so, the relevant original text of the historical dialogue can be retrieved from the original text data table for the AI ​​model to refer to when generating the response content.

[0077] In specific implementation, the orchestration module retrieves user profile information from the long-term memory, performs intent recognition on the user's input dialogue content, determines decision logic based on the intent recognition results, and performs process control; the decision logic is used to determine the retrieval method for multiple types of data tables in the memory system.

[0078] Specifically, the decision logic may include: whether it is necessary to retrieve the short-term cache or long-term memory to obtain historical dialogue content; if it is necessary to retrieve the long-term memory, then it is further determined whether it is necessary to obtain the summary or original text of the relevant historical dialogue content; so as to perform process control according to the decision logic. During the process control, the original text data table and / or summary data table in the short-term cache and long-term memory are retrieved to obtain the dialogue content within the most recent preset number of rounds, the summary and / or original text of the dialogue content before the preset number of rounds, and then a prompt message is generated in combination with the user profile information and input into the AI ​​model.

[0079] In addition, the orchestration module can determine whether additional information outside the dialogue system is needed based on the identified user intent. If so, the additional information can be obtained by calling relevant tools and provided to the AI ​​model for reference.

[0080] Corresponding to Embodiment 2, this application also provides an information processing device in an AI dialogue system, which may include: The dialogue content receiving unit is used to receive dialogue content input by the user and submit it to the server so that the server can obtain user profile information through the memory system database associated with the dialogue system and add it to the prompt information, and then provide it to the AI ​​model. The user profile information is extracted from the dialogue content and dynamically updated to the database by the AI ​​model during the process of conducting historical dialogues with the user. The AI ​​model is used to generate personalized response content by referring to the user profile information. The reply content display unit is used to output the personalized reply content through the client's dialogue interface and display a reference information card. The reference information card is used to display the user profile information referenced by the AI ​​model in the process of generating the reply content. The interaction unit is used to submit dialogue content input by the user for updating user profile information to the server so as to update the user profile information in the database.

[0081] Corresponding to Embodiment 3, this application also provides an information processing device for an AI dialogue system, which may include: The information storage unit is used to add the dialogue content of the most recent preset number of rounds to the short-term memory cache of the AI ​​model during the process of historical dialogue with the user through the AI ​​model, persist the dialogue content before the preset number of rounds to the long-term memory outside the AI ​​model, and extract user profile information from the dialogue content, persist it to the long-term memory and update it dynamically. The retrieval unit is configured to, upon receiving the current dialogue content input by the user, retrieve the user profile information from the long-term memory, and perform intent recognition based on the current dialogue content. If the recognized user intent is related to historical dialogue content, the relevant historical dialogue content is retrieved from the short-term memory cache and / or the long-term memory. The prompt information generation unit is used to generate prompt information based on the current dialogue content, user profile information, and relevant historical dialogue content, and to call the AI ​​model based on the prompt information so that the AI ​​model can generate personalized response content.

[0082] Additionally, the device may also include: The card display and interaction unit is used to provide the personalized response content generated by the AI ​​model to the client so that it can be output through the client's dialogue interface and to display reference information cards. The reference information cards are used to display the user profile information referenced by the AI ​​model in the process of generating the response content. If the user inputs dialogue content for updating the user profile information is received, the user profile information in the database is updated.

[0083] In addition, embodiments of this application also provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method described in any of the foregoing method embodiments.

[0084] And an electronic device, comprising: One or more processors; and A memory associated with the one or more processors, the memory being used to store program instructions that, when read and executed by the one or more processors, perform the steps of the method described in any of the foregoing method embodiments.

[0085] A computer program product includes a computer program / computer executable instructions that, when executed by a processor in an electronic device, implement the steps of the method described in the foregoing method embodiments.

[0086] in, Figure 8 An exemplary architecture of an electronic device is shown, which may include a processor 810, a video display adapter 811, a disk drive 812, an input / output interface 813, a network interface 814, and a memory 820. The processor 810, video display adapter 811, disk drive 812, input / output interface 813, network interface 814, and memory 820 can communicate with each other via a communication bus 830.

[0087] The processor 810 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solution provided in this application.

[0088] The memory 820 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 820 can store the operating system 821 for controlling the operation of the electronic device 800, and the basic input / output system (BIOS) for controlling the low-level operations of the electronic device 800. Additionally, it can store a web browser 823, a data storage management system 824, and a dialogue information processing system 825, etc. The aforementioned dialogue information processing system 825 can be the application program that specifically implements the aforementioned steps in this embodiment. In summary, when implementing the technical solution provided in this application through software or firmware, the relevant program code is stored in the memory 820 and is called and executed by the processor 810.

[0089] The input / output interface 813 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components within the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touchscreens, microphones, various sensors, etc., while output devices may include displays, speakers, vibrators, indicator lights, etc.

[0090] Network interface 814 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0091] Bus 830 includes a pathway for transmitting information between various components of the device, such as processor 810, video display adapter 811, disk drive 812, input / output interface 813, network interface 814, and memory 820.

[0092] It should be noted that although the above-described device only shows the processor 810, video display adapter 811, disk drive 812, input / output interface 813, network interface 814, memory 820, bus 830, etc., in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the solution of this application, and does not necessarily include all the components shown in the figures.

[0093] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0094] The various embodiments in this specification 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, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0095] The foregoing has provided a detailed description of the information processing method, system, and electronic device in the AI ​​dialogue system provided by this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are merely for the purpose of helping to understand the method and its core ideas. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. An information processing method in an AI dialogue system, characterized in that, include: After receiving the dialogue content input by the user, the system retrieves user profile information through the memory system database associated with the dialogue system. The user profile information is extracted from the dialogue content and dynamically updated into the database by the AI ​​model during the historical dialogue with the user. The user profile information is added to the prompt information and provided to the AI ​​model so that the AI ​​model can generate personalized response content by referring to the user profile information; The personalized response content generated by the AI ​​model is provided to the client for output through the client's dialogue interface, and a reference information card is displayed. The reference information card is used to display the user profile information referenced by the AI ​​model in the process of generating the response content. If the dialogue content input by the user for updating the user profile information is received, the user profile information in the database is updated.

2. The method according to claim 1, characterized in that, During the dialogue with the user through the AI ​​model, the historical dialogue content is also saved to the database of the memory system. The memory system includes a short-term memory cache and a long-term memory bank. The short-term memory cache is used to cache the dialogue content of the most recent preset number of rounds, and the long-term memory bank is used to persistently store the dialogue content before the preset number of rounds. The long-term memory is also used for persistent storage of the user profile information; The method further includes: After receiving the dialogue content input by the user, the intent is identified based on the dialogue content; If the identified user intent is related to historical dialogue content, the relevant historical dialogue content is retrieved from the memory system and input into the AI ​​model through the prompt information, so that the AI ​​model can generate response content by referring to the relevant historical dialogue content.

3. The method according to claim 2, characterized in that, Also includes: The reference information card also includes historical dialogue information referenced by the AI ​​model when generating the response.

4. The method according to claim 2, characterized in that, When saving historical dialogue content to the long-term memory, the original text of the historical dialogue content is saved to the original text data table in the long-term memory so that the AI ​​model can generate response content by referring to the original text of the relevant historical dialogue content.

5. The method according to claim 4, characterized in that, When saving historical dialogue content to the long-term memory, summary content is also generated for topics that may be used repeatedly and saved to the summary data table in the long-term memory. The summary content is generated by summarizing or summarizing the original text of multiple dialogue contents related to the same topic. The retrieval of relevant historical dialogue content from the memory system includes: Retrieve summaries of relevant historical dialogue content from the summary data table and input them into the AI ​​model.

6. The method according to claim 5, characterized in that, Also includes: After retrieving the summary content, it is determined whether it is necessary to refer to the original text of the historical dialogue. If so, the relevant original text of the historical dialogue is retrieved from the original text data table for the AI ​​model to refer to when generating the response content.

7. The method according to claim 6, characterized in that, The orchestration module retrieves user profile information from the long-term memory and performs intent recognition on the user's input dialogue content. Based on the intent recognition results, it determines decision logic and performs process control. The decision logic is used to determine the retrieval method for multiple types of data tables in the memory system.

8. The method according to claim 7, characterized in that, The decision logic includes: whether it is necessary to retrieve the short-term cache or long-term memory to obtain historical dialogue content; if it is necessary to retrieve the long-term memory, then it is further determined whether it is necessary to obtain the summary or original text of the relevant historical dialogue content; so as to perform process control according to the decision logic. During the process control, the original text data table and / or summary data table in the short-term cache and long-term memory are retrieved to obtain the dialogue content within the most recent preset number of rounds, the summary and / or original text of the dialogue content before the preset number of rounds, and then a prompt message is generated in combination with the user profile information and input into the AI ​​model.

9. The method according to claim 7, characterized in that, Also includes: The orchestration module determines whether additional information outside the dialogue system is needed based on the identified user intent. If so, it obtains the additional information by calling relevant tools and provides it to the AI ​​model for reference.

10. An information processing method in an AI dialogue system, characterized in that, include: The system receives dialogue content input by the user and submits it to the server. The server then retrieves user profile information from the memory system database associated with the dialogue system, adds it to the prompt information, and provides it to the AI ​​model. The user profile information is extracted from the dialogue content and dynamically updated to the database by the AI ​​model during the historical dialogue between the AI ​​model and the user. The AI ​​model is used to generate personalized response content by referring to the user profile information. The personalized response content is output through the client's dialogue interface, and a reference information card is displayed. The reference information card is used to display the user profile information referenced by the AI ​​model in the process of generating the response content. If a user input is received requesting an update to their user profile information, it is submitted to the server for updating the user profile information in the database.

11. An information processing method in an AI dialogue system, characterized in that, include: During the process of engaging in historical dialogues with users through an AI model, the dialogue content of the most recent preset number of rounds is added to the short-term memory cache of the AI ​​model, the dialogue content before the preset number of rounds is persistently stored in the long-term memory outside the AI ​​model, and user profile information is extracted from the dialogue content, persistently stored in the long-term memory and dynamically updated. After receiving the current dialogue content input by the user, the user profile information is obtained from the long-term memory, and intent is identified based on the current dialogue content. If the identified user intent is related to historical dialogue content, the relevant historical dialogue content is retrieved from the short-term memory cache and / or the long-term memory. Based on the current conversation content, user profile information, and relevant historical conversation content, a prompt message is generated, and the AI ​​model is invoked based on the prompt message so that the AI ​​model can generate personalized response content.

12. The method according to claim 11, characterized in that, Also includes: The personalized response content generated by the AI ​​model is provided to the client for output through the client's dialogue interface, and a reference information card is displayed. The reference information card is used to display the user profile information referenced by the AI ​​model in the process of generating the response content. If the dialogue content input by the user for updating the user profile information is received, the user profile information in the database is updated.

13. An information processing system in an AI dialogue system, characterized in that, include: The memory module includes a short-term memory cache and a long-term memory database. The short-term memory cache is used to cache the dialogue content of the most recent preset number of rounds. The long-term memory database includes a raw text data table, a summary data table, and a user profile data table. The raw text data table is used to persistently store the raw text of the dialogue content up to the preset number of rounds. The summary data table is used to persistently store the summary content generated by summarizing or summarizing multiple dialogue contents related to topics that may be used repeatedly. The user profile data table is used to persistently store user profile information. The user profile information is extracted from the dialogue content and dynamically updated into the user profile data table by the AI ​​model during historical dialogues with the user. The orchestration module, upon receiving dialogue input from the user, retrieves user profile information from the user profile data table in the long-term memory and performs intent recognition on the user's input dialogue content. Based on the intent recognition result, it determines decision logic, which includes: determining whether it is necessary to search the short-term cache or long-term memory to obtain historical dialogue content; if it is necessary to search the long-term memory, it further determines whether it is necessary to obtain a summary or original text of the relevant historical dialogue content. Based on the decision logic, it searches the original text data table and / or summary data table in the short-term cache and long-term memory to obtain the dialogue content within the most recent preset number of rounds, and the summary and / or original text of the dialogue content before the preset number of rounds. Then, it generates prompt information in conjunction with the user profile information and inputs it into the AI ​​model, enabling the AI ​​model to generate personalized response content by referring to the user profile information and the relevant historical dialogue content.

14. The system according to claim 13, characterized in that, The orchestration layer is also used to output the personalized reply content through the client's dialogue interface and display reference information cards, which are used to display the user profile information referenced by the AI ​​model in the process of generating the reply content; wherein, if a dialogue content input by the user for updating the user profile information is received, the user profile information in the user profile data table is updated.

15. The system according to claim 13, characterized in that, The orchestration module is also used to determine whether additional information outside the dialogue system is needed based on the identified user intent. If so, the additional information is obtained by calling relevant tools and provided to the AI ​​model for reference.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps of the method described in any one of claims 1 to 12.

17. An electronic device, characterized in that, include: One or more processors; as well as A memory associated with the one or more processors, the memory being used to store program instructions that, when read and executed by the one or more processors, perform the steps of the method according to any one of claims 1 to 12.

18. A computer program product comprising a computer program / computer-executable instructions, characterized in that, When the computer program / computer-executable instructions are executed by a processor in an electronic device, they implement the steps of the method according to any one of claims 1 to 12.

Images