Session message processing method, device, storage medium and program product

Abstract

Embodiments of the present application provide a kind of conversation message processing method, equipment, storage medium and program product.In the present application embodiment, during the offline of client, server actively extracts the message digest of the offline session message of each session;After the re-online of client, server sends the lightweight message digest list to client, both can efficiently overview the content of each session, and also can reduce network transmission load.According to the selected message digest of user, the associated offline session message is immediately acquired, which avoids waiting for pulling other offline session messages, and improves the viewing efficiency.In the meantime of acquiring the offline session message associated with the selected message digest, other session message digest can also be recommended according to the currently selected message digest, so that the user can view the related discussion in other sessions while viewing the current session details, and the efficiency and breadth of information discovery are improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a session message processing method, device, storage medium, and program product. Background Technology

[0002] Some internet applications support users opening multiple conversations. For example, in some customer service applications, customer service representatives can simultaneously engage in conversations with multiple customers. Similarly, in some instant messaging applications, users can converse with multiple friends and even create various groups for group chats.

[0003] In practical applications, due to network connection interruptions, program shutdowns, or user logouts, the application may enter an offline state. While offline, it cannot receive messages from other online users in real time. During this period, the server stores messages sent by other online users, and once the user logs back in, it sends all messages accumulated during the offline period to the application.

[0004] Faced with numerous unread messages received after reconnecting to the network, users typically navigate to each conversation interface one by one to view them. If users browse quickly, they may miss useful messages; however, if they carefully examine each message, the process is inefficient and time-consuming. Summary of the Invention

[0005] This application provides a method, apparatus, storage medium, and program product for processing conversational messages to improve message viewing efficiency.

[0006] This application provides a session message processing method applied to a client. The method includes: after the client re-enters the network, in response to an access operation to a first session, displaying the first session interface corresponding to the first session, and sending a first message request to the server to request message summaries corresponding to the real-time session messages and offline session messages of the first session; receiving and displaying a list of first message summaries returned by the server according to the first message request, the list of first message summaries including message summaries obtained by the server using a first large model to extract content from offline session messages of the first session during the client's offline period; in response to a selection operation of the first message summaries in the list of first message summaries, displaying offline session messages associated with the first message summaries, and sending a second message request to the server, so that the server can call a second large model to recommend and return a list of second message summaries based on the first message summaries; the list of second message summaries includes at least one second message summary with the same or similar message topic as the first message summary, the second message summaries being obtained by the server using a first large model to extract content from offline session messages of other sessions during the client's offline period; and displaying the list of second message summaries returned by the server, so that the user can view other offline session messages with the same or similar message topics in other sessions through the second message summaries in the list of second message summaries.

[0007] This application embodiment also provides a session message processing method applied to a server. The method includes: after re-establishing a network connection with a client, receiving a first message request sent by the client, the first message request including an identifier of a first session; obtaining a first message summary list corresponding to real-time session messages and offline session messages of the first session based on the identifier of the first session, and sending it to the client for display in the first session interface corresponding to the first session; the first message summary list includes message summaries obtained by extracting content from offline session messages of the first session using a first large model during the client's offline period; receiving a second message request sent by the client, the second message request including a first message summary selected from the message summaries in the first message summary list; calling a second large model to recommend a second message summary list based on the first message summary, the second message summary list including at least one second message summary with a message topic that is the same as or similar to the first message summary, the second message summary being obtained by extracting content from offline session messages of other sessions using the first large model during the client's offline period; and sending the second message summary list to the client so that the user can view other offline session messages with the same or similar message topics in other sessions through the second message summaries in the second message summary list.

[0008] This application also provides an electronic device, which includes a memory and a processor. The memory is used to store a computer program, and the processor is coupled to the memory to execute the computer program for implementing the steps in any of the session message processing methods.

[0009] This application also provides a computer-readable storage medium storing a computer program / instructions, which, when executed by a processor, enables the processor to implement the steps of any of the methods in the session message processing method.

[0010] This application also provides a computer program product, which includes a computer program / instructions that, when executed by a processor, cause the processor to implement the steps in any of the methods in the session message processing method.

[0011] This application provides a method, device, storage medium, and program product for processing session messages. When a client is offline, the server actively extracts message digests from offline session messages for each session. After the client re-enters the network, the server sends a lightweight list of message digests to the client, efficiently providing an overview of the content of each session while reducing network transmission load. The server instantly retrieves associated offline session messages based on the user-selected message digest, avoiding the time spent waiting to retrieve other offline session messages and improving viewing efficiency. While retrieving offline session messages associated with the selected message digest, the server can also recommend message digests from other sessions based on the currently selected message digest, allowing users to view related discussions in other sessions while viewing the details of the current session, improving the efficiency and breadth of information discovery. Attached Figure Description

[0012] 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:

[0013] Figure 1 A schematic diagram of the architecture and a schematic diagram of the working principle of the session message processing method provided for an exemplary embodiment of this application;

[0014] Figure 2 A flowchart illustrating a session message processing method provided as another exemplary embodiment of this application;

[0015] Figure 3 A flowchart illustrating a session message processing method provided as another exemplary embodiment of this application;

[0016] Figure 4 A schematic diagram of the structure of a session message processing apparatus provided as another exemplary 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] Internet applications refer to computer programs that interact with servers via a network to provide various functions and services. These applications can take many forms, including native mobile applications, lightweight mini-programs, cross-platform web applications, and centralized cloud services. Functionally, internet applications can be categorized into instant messaging applications, customer service applications, email services, audio / video applications, and e-commerce applications, among others.

[0019] Some internet applications support conversational functionality, where a conversation refers to a sustainable message exchange entity established between two or more users. It is not only a message transmission channel but also a core unit for maintaining conversation context, storing historical records, and managing user state. Different internet applications may employ different conversational models due to varying business scenarios. For example, in some customer service applications, customer service representatives can simultaneously engage in conversations with multiple consumers. In other instant messaging applications, users can not only establish one-on-one private conversations with multiple friends but also flexibly create or join various groups for many-to-many group chats.

[0020] In practical applications, internet applications may enter an offline state due to network connection interruptions, program closure, or user logout. While offline, they cannot receive messages from other online users in real time. During this period, the server stores messages sent by other online users, and sends all accumulated messages to the application once the user logs back in. However, this method has drawbacks: viewing messages too quickly can lead to missing important messages, and viewing them one by one is inefficient and time-consuming.

[0021] To improve the efficiency of viewing offline session messages, improvements can be made in three technical directions: storage and transmission mechanisms, message presentation methods, and message retrieval mechanisms. However, these three technical directions belong to different technical fields and employ significantly different technical methods. Combining them presents technical obstacles and can create new technical problems. For example, in terms of storage and transmission mechanisms, binary compression and chunked transmission technologies can be used; in terms of message presentation, semantic analysis can be used to generate tags for messages; and in terms of message retrieval mechanisms, tags can be generated after the user re-logs back online, and messages corresponding to the tags can be retrieved on demand according to the user's intent. Combining these methods will create the following contradictions: First, there is a conflict between compression and semantic analysis. Offline message data after binary compression cannot be directly semantically analyzed to generate tags. To generate tags, the compressed offline message data must first be decompressed, making tag generation inefficient and affecting the efficiency of users retrieving messages corresponding to the tags on demand. Therefore, generating tags after the user re-logs back online takes a long time, and the viewing efficiency remains low. Secondly, even if tags can be generated for a single session, the compressed binary data cannot be directly used for semantic analysis in artificial intelligence (AI) models, and this tagging information based on local context is insufficient to support cross-session recommendations. In summary, the above methods lack the ability to integrate AI models, efficient data transmission, and cross-session information interaction.

[0022] Based on this, this application provides a session message processing method. When a client is offline, the server actively extracts message digests from offline session messages for each session. After the client re-enters the network, the server sends a lightweight list of message digests to the client, which efficiently provides an overview of the content of each session while reducing network transmission load. The associated offline session messages are retrieved instantly based on the message digest selected by the user, avoiding the time spent waiting to retrieve other offline session messages and improving viewing efficiency. While retrieving the offline session messages associated with the selected message digest, message digests from other sessions can also be recommended based on the currently selected message digest. This allows users to view related discussions in other sessions while viewing the details of the current session, improving the efficiency and breadth of information discovery.

[0023] The following is in conjunction with the appendix Figure 1 This application provides detailed explanations of the technical solutions provided in each embodiment, such as... Figure 1The diagram shows the architecture and working principle of the session message processing method provided in this application embodiment. The technical solution provided in this application embodiment involves a client 100 and a server 101. The client can refer to a terminal device through which the user 102 interacts with the server, or an application running on that terminal device. This terminal device can include, but is not limited to, smartphones, tablets, computers, smart wearable devices, etc., and can be understood as the aforementioned internet applications. In this application embodiment, the client is mainly responsible for presenting the user interface, receiving user operation instructions, sending requests to the server, and receiving and displaying data from the server. The server can refer to a computing device or software system that provides computing, storage, and business logic support to the client. It can be deployed on a physical server, server cluster, or cloud platform and maintain continuous operation to handle requests from one or more clients at any time. In this application embodiment, the server is mainly responsible for receiving and processing requests sent by the client, executing corresponding business logic, storing and managing data, and returning the processing results to the client.

[0024] In this embodiment, the client manages one or more sessions. Each session can refer to an independent communication context, which can be used to organize and carry a series of ordered messages between two or more participants. For example, a session can be a point-to-point session between two users or a group session containing multiple members. Each session is associated with an interactive session interface, which can refer to a graphical user interface used to carry and display session message content and receive user input. To ensure data consistency and availability, the client's data state needs to be synchronized with the server, and a local data cache can be maintained. Based on the network connection status with the server, the client can be in one of the following two states:

[0025] Online status: The client and server maintain a network connection and can exchange data in real time. A client in this state can be called an online client. Session messages can be pushed from the server to the online client in real time, or sent from the online client to the server and updated in the corresponding session interface.

[0026] Offline state: The network connection between the client and the server is interrupted. A client in this state is called an offline client. At this time, the offline client cannot exchange data in real time; that is, it cannot receive information sent by the server. Messages sent by online clients in a session associated with the offline client cannot be received by the offline client. These messages are considered offline session messages to the offline client, and the server can store them. Additionally, the offline client can access locally cached data, for example, by viewing and browsing historical messages in the session interface.

[0027] When a client returns from an offline state to an online state (which can be understood as the client coming back online), the client will re-establish a connection with the server and trigger a data synchronization process. During this process, the server can send the client offline session messages received and stored during the client's offline period, as well as message digests obtained by extracting the content of those offline session messages. Simultaneously, the client regains its ability to receive real-time session messages. The client can then integrate and display the acquired offline session messages, message digests, and real-time session messages in the corresponding session interface for user viewing.

[0028] The following is combined with Figure 2 The illustrated method embodiments provide a detailed description of the session message processing method provided in this application. For example... Figure 2 As shown, the method includes:

[0029] S200: After the client comes back online, in response to the access operation of the first session, the first session interface corresponding to the first session is displayed, and the first message request is sent to the server to request the message digests corresponding to the real-time session message and the offline session message corresponding to the first session.

[0030] S210. Receive and display the first message digest list returned by the server according to the first message request. The first message digest list includes message digests obtained by the server extracting content from the offline session messages of the first session using the first big model during the client's offline period.

[0031] S220. In response to the selection operation of the first message digest in the first message digest list, display the offline session messages associated with the first message digest, and send a second message request to the server so that the server can call the second major model to recommend and return a second message digest list based on the first message digest; the second message digest list includes at least one second message digest with a message topic that is the same as or similar to the first message digest, and the second message digest is obtained by the server using the first major model to extract the content of offline session messages of other sessions during the client's offline period.

[0032] S230. Display the second message digest list returned by the server so that users can view other offline session messages with the same or similar message topics in other sessions through the second message digest in the second message digest list.

[0033] After the client reconnects, the user can select one session from multiple sessions to view the details of that session. In this embodiment, the first session may refer to the session currently selected by the user. The access operation to the first session may refer to the operation performed to access the first session. This embodiment does not limit the implementation method of the access operation, for example, it may be a click, tap, voice command, gesture recognition, or any other form of user input that can be recognized by the terminal device.

[0034] After the client responds to the access operation of the first session, the following two processes can be executed simultaneously or sequentially: display the first session interface corresponding to the first session; send the first message request to the server.

[0035] The first session interface can refer to a graphical user interface that uniquely corresponds to the first session and is used to display the content of that first session. The data displayed on the first session interface may come from local cache or from the server. Therefore, the content displayed on the first session interface may change depending on the local cache and / or the content requested from the server.

[0036] The first message request refers to the initial session synchronization request sent to the server when the client re-enters the network. This first message request includes the real-time session message corresponding to the first session and message summaries corresponding to the offline session messages. The message summary refers to the summary content generated by the server during the client's offline period by calling the first major model to extract content from one or more offline session messages. This message summary is only generated for sessions involving offline session messages. The message summary of the first session can be obtained by extracting content from the offline session messages of the first session using the first major model.

[0037] This application does not limit the implementation of the message digest. In one implementation, the server obtains messages sent by online clients in sessions associated with offline clients in real time, i.e., offline session messages; it monitors in real time whether the difference between the current time and the previous update time of each session exceeds a preset time period. If it does not exceed the preset time period, the server calls the first main model to extract the content of the offline session messages received within the preset time period to obtain a message digest. The previous update time can refer to the time when the content of the offline session messages within the previous preset time period is extracted. The previous update time for each session can be different. For example, the preset time period is 2 minutes. For session e1 in the offline client, the server obtains offline session messages m1, m2, and m3 sent by online clients E1 and E2 in session e1 from the previous update time to the current time; for session e2 in the offline client, the server obtains offline session message m4 sent by online client E3 in session e2 from the previous update time to the current time. If the difference between the current time and the previous update time of session e1 exceeds 2 minutes, and the difference between the current time and the previous update time of session e2 does not exceed 2 minutes, then the first major model is called to extract the content of m1, m2, and m3 corresponding to session e1 to obtain the message digest a1 of e1.

[0038] The message digest may include sender information, overview content, message sending time, the time of the first offline session message received within a preset time period, and the time of the last offline session message, etc. This application embodiment does not limit this. For example, the message digest may be: [{start_time:2024-06-21 00:22:34,end_time:2024-06-21 00:24:22,desc:“Discussed the itinerary for the Shanghai trip, and discussed the food plan during the trip”},{…}], where start_time may represent the time of the first offline session message received within the preset time period, end_time may represent the time of the last offline session message received within the preset time period, and desc represents the overview content.

[0039] The first major model refers to an AI model with a relatively large parameter scale that can understand natural language. This model can recognize various information formats such as text, speech, and images, and can also understand the contextual relationships, deep semantics, sentiment, and implicit intentions of input offline conversation messages. Based on this, the first major model can perform tasks such as content analysis and logical reasoning, thereby generating representative and logically coherent message summaries. Therefore, the message summaries generated by the first major model should not be regarded as a simple aggregation of keywords, but rather as a precise summary of the offline conversation content, which can serve as a basis for users to make decisions, such as deciding whether to click to view the details of the message summary.

[0040] The server can generate a corresponding message digest list for each session involving message digests. This message digest list can refer to a set of permutations and combinations of at least one message digest corresponding to that session, for example, it can be arranged according to the priority of the message digests. Based on this, the server can send a first message digest list to the client after receiving a first message request; the first message digest list can refer to a set of permutations and combinations of message digests corresponding to the first session.

[0041] The client receives and displays the first message digest list. This application embodiment does not limit the display method of the first message digest list. For example, the first message digest list and the first session interface can adopt a parent-child hierarchical layout, with the first session interface as the parent container and the first message digest list as the child container nested within the first session interface.

[0042] Furthermore, the server can also send the real-time session message corresponding to the first session to the client based on the received first message request. The real-time session message and the first message summary list can be displayed on the first session interface at the same time.

[0043] After receiving the first message digest list, the user can select a desired message digest from the list to view the associated offline session messages. Based on this, the client can respond to the selection of a first message digest in the first message digest list by displaying the associated offline session messages. The first message digest may refer to the message digest currently selected by the user in the first message digest list. The associated offline session messages can be obtained from the client's local cache or by sending a request to the server. In this embodiment, similar to the aforementioned access operation to the first session interface, this application does not limit the specific implementation of the selection operation, including but not limited to clicks, taps, or voice commands.

[0044] Furthermore, after the client responds to the selection operation of the first message digest in the first message digest list, it can also send a second message request to the server. This second message request is used to request a message digest with the same or similar message subject as the first message digest.

[0045] Upon receiving a second message request, the server can invoke a second large model to recommend at least one second message summary based on the first message summary. Similar to the first large model, the second large model can refer to an AI model with a relatively large parameter scale capable of understanding natural language. This second large model possesses deep semantic understanding capabilities, enabling it to deeply analyze the contextual relationships, deep semantics, sentiment, and implicit intent of multiple input message summaries. Based on this understanding, the second large model can perform content analysis and logical reasoning, identifying deep, non-explicit semantic relationships between different message summaries. This allows it to accurately recommend second message summaries whose message topics are the same as or similar to the selected first message summary, achieving cross-session recommendation.

[0046] The first and second main models can be integrated within the same main model, with different functional modules implementing their respective functions; alternatively, they can be two completely independent main models, differing in architecture, parameter count, or training data. The second message digest can refer to a message digest from other sessions whose topic is the same as or similar to the first message digest. It can be obtained by the server extracting content from offline session messages from other sessions using the first main model while the client is offline. The list of second message digests can refer to a set formed by permutations and combinations of at least one second message digest, with different second message digests originating from different other sessions.

[0047] After the server generates the second message digest list, it returns it to the client. The client receives and displays this list, allowing the user to view other offline conversation messages with the same or similar topics from other conversations through the second message digests in the list. In this embodiment, by implementing intelligent cross-conversation recommendations, dynamic associations based on semantic content are built between originally independent conversation messages. This mechanism can proactively present relevant content from other conversations to the user, thereby breaking down traditional information silos. This not only significantly improves the efficiency of users discovering related information but also effectively expands the breadth of information acquisition by displaying potential knowledge connections.

[0048] The embodiments of this application do not limit the implementation of the display of the second message summary list and the first message summary list. For example, the second message summary list and the first message summary list can be displayed side by side on the upper layer of the first session interface in the form of pop-up windows, floating cards or sidebars; or, the second message summary list can be embedded inside the first message summary list.

[0049] This application provides a session message processing method. When a client is offline, the server actively extracts message digests from offline session messages for each session. After the client re-enters the network, the server sends a lightweight list of message digests to the client, which efficiently provides an overview of the content of each session while reducing network transmission load. The method instantly retrieves associated offline session messages based on the user-selected message digest, avoiding the time spent waiting to retrieve other offline session messages and improving viewing efficiency. While retrieving offline session messages associated with the selected message digest, the method can also recommend message digests from other sessions based on the currently selected message digest. This allows users to view related discussions in other sessions while viewing the details of the current session, improving the efficiency and breadth of information discovery.

[0050] Current offline session message processing methods struggle to simultaneously optimize network, time, and cognitive costs. For example, full-data retrieval sacrifices bandwidth and time, while manual browsing sacrifices time and effort. This application's embodiments address these issues by integrating deep content understanding using AI models (first and second major models), efficient data transmission mechanisms (such as prioritizing user-selected information), and innovative interaction methods (recommending content from other sessions). This integration goes beyond simply improving viewing efficiency; it creates a positive feedback loop. The first major model's deep understanding of offline session messages ensures accurate message summaries, providing a foundation for precise user decision-making and enabling reliable selection of essential content. This accurate user selection facilitates on-demand retrieval, ensuring precise loading of user-required data. The second major model's deep understanding of message summaries ensures accurate content recommendations, while innovative interaction methods provide valuable guidance. This combination transforms offline session message processing from an inefficient message retrieval process into a precise, efficient, and guided knowledge discovery workflow.

[0051] In one optional embodiment, the client may further include a task scheduler and a background service. The task scheduler can act as a coordination and decision-making module within the client, responding to various triggering events and creating corresponding tasks accordingly. Triggering events may include user behavior triggering events and client status triggering events. Based on user behavior triggering events, session or content creation tasks can be created for user actions. For example, user behavior triggering events may include the aforementioned "access to the first session" and "selection of the first message digest." For the "access to the first session" operation, the task scheduler can create a task to retrieve data related to the first session; for the "selection of the first message digest" operation, the task scheduler can create a task to retrieve offline session messages associated with the first message digest. Client status triggering events may include client re-online triggering events. For this event, the task scheduler can create a task to retrieve all offline session messages, corresponding message digests, and real-time session messages generated during the client's offline period to achieve synchronization with the server. The background service may refer to an execution unit running independently in the client's background, responsible for receiving and executing the retrieval tasks created and dispatched by the task scheduler.

[0052] To improve data synchronization efficiency in scenarios triggered by client re-entry, each existing session on the client can be assigned a priority. This priority is dynamically determined based on session attribute information and is uniformly maintained by the task scheduler. This attribute information may include session type, session creation time, and the session's importance to the user. In response to a client re-entry event, the task scheduler creates a retrieval task sequence for each session according to its maintained priority from highest to lowest, and then sends this sequence to the background service. The background service executes the retrieval operations in the order of the retrieval task sequence, ensuring that offline session messages for higher-priority sessions are retrieved from the server first.

[0053] Based on this, in one optional embodiment, after the client reconnects, the method further includes: according to the session priority maintained by the task scheduler, using the task scheduler to schedule a background service to obtain message digests corresponding to each session and offline session messages corresponding to each message digest from the server, and caching them locally on the client for display in the session interface corresponding to each session. In this embodiment, sending the first message request to the server can occur before the access operation in response to the first session. If it occurs in response to the access operation of the first session, the client can cache the real-time session messages corresponding to the first session and the message digests corresponding to the offline session messages locally.

[0054] To ensure immediate response to user actions, the task scheduler employs both a cache-first strategy and a priority scheduling mechanism when handling user-triggered events. Specifically, when a user action triggers an event, the task scheduler first queries the client's local cache. If relevant data is found in the local cache, it is retrieved directly from the cache to respond to the user action. If no relevant data is found in the local cache, the task scheduler creates a fetch task and assigns it the highest priority. Based on this, if a background service is executing a synchronous task triggered by a client re-entry event, the task scheduler can issue an interrupt command, suspending the current synchronous task and prioritizing the task triggered by the user action. After the user action task is completed, the background service can resume execution of the previously suspended synchronous task according to the task scheduler's instructions.

[0055] Based on this, in an optional embodiment, in response to the selection operation of the first message digest in the first message digest list, displaying the offline session message associated with the first message digest may include: in response to the selection operation of the first message digest in the first message digest list, querying the offline session message corresponding to the first message digest in the local cache of the client; if the first message digest is not found, using the task scheduler to create a first task with the highest priority based on the first message digest, and using the first task scheduling background service to obtain and display the offline session message associated with the first message digest from the server.

[0056] By replacing the initial synchronization of all messages with a lightweight message digest, unnecessary network transmission and client data processing are significantly reduced at the source. Furthermore, a cache-first retrieval and priority scheduling mechanism prioritizes the retrieval of messages the user wants to view. Combining lightweight message digests, cache-first retrieval, and priority scheduling improves user viewing efficiency while ensuring low perceived latency.

[0057] In one optional embodiment, the first message digest includes multi-level message topics, and the list of first message digests is displayed in a display container in the first session interface. Responding to a selection operation on a first message digest in the first message digest list, displaying offline session messages associated with the first message digest may include: responding to a selection operation on a first message digest in the first message digest list, displaying the first-level message topic in the area associated with the first message digest within the display container; responding to a selection operation on any message topic in the current level message topics, displaying the next-level message topic in the area associated with the first message digest, until the last-level message topic is displayed; wherein the current-level message topic may be the first-level message topic or a non-first-level message topic; responding to a selection operation on any message topic in the last-level message topics, obtaining and displaying the offline session messages associated with the currently selected last-level message topic; wherein offline session messages corresponding to unselected message topics are not obtained from the client's local cache or the server.

[0058] A message topic can refer to a structured unit that summarizes the core content of an offline session message corresponding to a first message digest, extracted through semantic understanding and content summarization. A multi-level message topic refers to a predefined parent-child hierarchy among multiple message topics, forming a hierarchical tree-like data structure. In this structure, a higher-level message topic represents a relatively macroscopic and generalized content, while a lower-level topic refers to a more detailed description of the higher-level message topic. In an optional embodiment, in response to a client re-entry trigger event, the task scheduler can retrieve the corresponding message topic in addition to the message digest. Therefore, in response to a selection operation of any message topic in the current level, the task scheduler can first check its local cache to see if the next-level message topic is cached. If not, it can create a task to prioritize retrieving the next-level message topic and schedule a background service to execute the task.

[0059] A display container refers to a visually independent view component in a graphical user interface (GUI) used to hold a list of first message summaries. Its form can be a floating layer, a card panel, or an expandable window. This display container provides an independent interactive space for the first message summary list, allowing users to expand, collapse, or navigate within it without interfering with other elements in the first session interface. When a user selects a message summary within the first message summary list, multiple levels of message topics associated with that first message summary can be dynamically loaded as subviews within the display container and displayed in the area associated with the first message summary.

[0060] In one optional implementation, the display container may include multiple independent sub-containers at the view hierarchy. Each sub-container may correspond to a message digest and is responsible for carrying and managing the display content of that message digest. This approach achieves operational isolation between sub-containers, ensuring that selecting the first message digest does not affect the display status of other message digests in the first message digest list, guaranteeing the stability and reliability of the interaction, and improving the user experience. In this implementation, the area associated with the first message digest may refer to the sub-container corresponding to the first message digest. When message topics at the current level are displayed, the sub-container can dynamically expand or shrink the size of its visible area according to the number and content of message topics at the current level. During this process, the view style of the first message digest or the message topic at the previous level can be flexibly changed, such as being hidden, remaining unchanged, or changing to a fixed title style. If multiple message topics exist at the same level, they can be displayed side by side within the sub-container.

[0061] In this embodiment, multi-level message topics can convert unstructured offline session messages into a data format with logical hierarchical relationships, thereby supporting users to perform progressive information queries. Users can first quickly overview the information through the first-level message topic, and then freely select the depth of information retrieval based on their network conditions, interests, and time constraints, improving personalized efficiency. Furthermore, by setting multi-level message topics, offline session messages from the selected final-level message topic can be retrieved, while offline session messages from intermediate-level message topics can be omitted, reducing the amount of offline session message data retrieved on demand and further improving viewing efficiency.

[0062] In one optional embodiment, the second message digest may also include message topics at multiple levels. In response to a selection operation on any message topic within the current level, the server may invoke a second major model to recommend at least one message topic from other sessions that is the same as or similar to the currently selected message topic from the second message digest list. For ease of description, the recommended message topics may be simply referred to as candidate message topics. The server may organize multiple candidate message topics into a candidate message topic list. Preferably, when recommending candidate message topics for any level of message topic, a global search of the second message digest list may not be performed. Instead, the candidate message topic list corresponding to the current level of message topic is obtained based on the candidate message topic list corresponding to the message topic at the previous level.

[0063] The server can organize multiple candidate message topics into a candidate message topic list and send this list to the client. The client receives the candidate message topic list and displays it on the first session interface. In response to a selection of any candidate message topic in the list, the client displays the offline session messages associated with that candidate message topic in either the first session interface or the session interface associated with that candidate message topic. During the display of the offline session messages associated with that candidate message topic, if the candidate message topic is not the lowest-level message topic, it can be displayed layer by layer according to the user's selection of candidate message topics.

[0064] Furthermore, after responding to the selection operation of any message topic in the last-level message topics, the method further includes: sending a third message request to the server, so that the server can match message topics that are the same as or similar to the currently selected last-level message topic from the second message digest to obtain a first candidate message topic list; obtaining offline session messages associated with the first candidate message topic list and returning them to the client; receiving offline session messages associated with the first candidate message topic list and displaying them in the first session interface in association with the offline session messages associated with the currently selected last-level message topic.

[0065] The third message request can be used to request offline session messages associated with the same or similar message topics as the currently selected last-level message topic. The first candidate message topic list can refer to the list of candidate message topics corresponding to the selection operation of any message topic in the last-level message topics.

[0066] In this embodiment, by recommending the selected first message topic in real time, a deeper connection can be established between the first session and other sessions. This presents the content from other sessions in a more hierarchical and logical manner within the first session, shortening the information discovery path and further improving viewing efficiency. The second message summary itself has a similar or identical message topic to the first message summary. Therefore, the recommended message topics further filtered from the second message summary are also related to the overall context of the first message summary. Based on this approach, the recommendation scope can be reduced from message summaries of all sessions to the second message summary, significantly reducing the computational complexity of recommendations, improving recommendation efficiency, and providing users with a smoother interactive experience. Furthermore, this progressively focused filtering mechanism prevents recommended content from deviating from the initial intent as the user explores, thus ensuring that the recommended message topics always focus on the user's core areas of interest.

[0067] In one optional embodiment, the first session interface may include operation controls associated with the displayed offline session message. These operation controls provide users with an interactive entry point to directly perform subsequent operations on the offline session message. The operation types include, but are not limited to, forwarding, favorited, saving to local storage, deleting, marking as read / unread, and returning. In response to triggering the operation control, the client performs a preset corresponding operation on the offline session message associated with that operation control. In one implementation, if the operation control includes a delete control, after displaying the offline session message associated with the first message summary, the method further includes: in response to a selection operation on the delete control associated with the offline session message associated with the first message summary, deleting the offline session message associated with the first message summary from the first session interface, and simultaneously deleting the offline session message associated with the first message summary cached locally on the client.

[0068] In one alternative embodiment, the message topic, message summary, and message summary list can also be associated with operation controls, enabling users to perform corresponding operations at different information organization granularities, such as batch processing the entire summary list or collecting and sharing a specific message topic.

[0069] In one optional embodiment, to achieve effective management of client-side local storage resources and enhance user awareness of important information, a lifecycle management mechanism for message digests can be configured, the details of which may include:

[0070] Data cleanup strategies: An immediate cleanup strategy can be adopted, which immediately deletes all message digests associated with a session when the user actively exits or closes the session. Alternatively, a time-based periodic cleanup strategy can be used, which sets a preset retention period for locally stored message digests; once the preset period is reached, deletion will be automatically performed to free up storage space.

[0071] Pending message reminder strategy: For any session with a generated message digest, if it is detected that the message digest remains unviewed by the user within a preset time window after its generation, a pending message reminder notification will be proactively sent to the user to guide the user to pay attention to the session.

[0072] Data expiration notification policy: To prevent accidental data deletion before implementing data cleanup policies, an expiration reminder mechanism can be configured. This involves sending a message summary to the user before the data reaches its set retention period, notifying them of its impending expiration, and providing options to extend the retention period or back up the data.

[0073] In an optional embodiment, after the client reconnects, in order to enable users to grasp the relevant content during the offline period without entering a specific session, the number of message digests and the latest message digest content of the session can be displayed in each session entry in the session list.

[0074] In this embodiment, the mechanism that links the content presented in the first session interface with the client's local cache for updates can prevent cache bloat by allowing users to actively release storage space, thus optimizing the long-term performance of the client. Combining this update mechanism with the recommendation mechanism allows for adjustments to the recommendation strategy of the second major model. For example, it can reduce the priority of message topics similar to deleted content, avoid repeatedly recommending content that users are not interested in, and improve the personalization of recommended content.

[0075] In one optional embodiment, this application does not limit the implementation of "using the first major model to extract content from the offline session messages of the first session to obtain each message summary in the first message summary list during client offline time". In one implementation, the server can construct profile information for each session, which can be used to characterize the user's level of attention to the session. Then, during client offline time, the server can generate personalized hierarchical message summaries based on the profile information. For example, for profile information representing a high level of attention, a deeper and more detailed message summary can be generated; for profile information representing a low level of attention, a less hierarchical and more general message summary can be generated. In this way, this application embodiment can transform the abstract user attention into specific differences in the content structure and depth of the message summary, thereby generating personalized hierarchical message summaries. Based on this, during the client's offline period, the first major model is used to extract the content of the offline session messages of the first session to obtain each message summary in the first message summary list, including: constructing the profile information of the first session, which includes personalized attribute information obtained based on the historical interaction data of the first session; according to the profile information of the first session, the first major model is called to extract the content of the offline session messages to obtain each message summary in the first message summary list as well as the corresponding level and message topic.

[0076] The profile information of the first conversation reflects the user's level of attention to that conversation. Personalized attribute information within this profile refers to metrics derived from the first major model based on historical interaction data between the user and the first conversation, which quantify the user's level of attention to that conversation. These personalized attributes may include interest preference metrics and activity metrics. For example, interest preference metrics may include whether the user has performed actions that explicitly express personal preferences, such as pinning the conversation, setting specific message notifications, or adding it to favorites. Activity metrics may include the frequency with which the user accesses and views the first conversation interface within a preset time period, and the average time spent viewing the conversation, reflecting the user's level of engagement with the conversation.

[0077] In an optional embodiment, to further enhance the adaptability of the generated message digest, the profile information of the first session may also include global attribute information. Global attribute information describes the user's state from a global perspective and may include environmental information and usage context information. Environmental information may refer to the physical and network environment status of the user's device, such as the network type and signal strength. Usage context information may refer to the specific scenario in which the user interacts, such as the time the user uses the client and the user's activity status.

[0078] Global attribute information is the key basis for the server to dynamically adjust the message digest generation strategy. For example, when the network condition of the user's device is detected to be poor, a message digest with fewer levels and more concise content can be generated. When the user is identified to be in a suitable reading context, a multi-level and richer message digest can be presented to them.

[0079] This application does not limit the method of obtaining global attribute information, and different methods will affect the generation of message digests. In one implementation, the global attribute information can be obtained based on historical data statistics when the client is online. In this case, personalized attribute information and global attribute information are considered together to generate a personalized message digest while the client is offline. In another implementation, the global attribute information can be obtained by real-time monitoring of the client after the client comes back online. In this case, a preliminary message digest is generated based on personalized attribute information while the client is offline; then, after the client comes back online, the preliminary message digest is fine-tuned based on the real-time monitored global attribute information.

[0080] In this embodiment, the first major model can adopt a scalable message digest generation mechanism. By integrating personalized attribute information and global attribute information, the first major model can generate different message digests for different users, different sessions, and different scenarios. The number of levels and content details of the message digest are adapted to the user's own network conditions, interests, and time costs, thereby enabling the user to obtain offline session messages in an efficient manner, avoiding the generation of redundant information from the source, reducing unnecessary computational consumption, and improving viewing efficiency.

[0081] This application does not limit the implementation of "using the first large model to extract content from the offline session messages based on the profile information of the first session to obtain each message summary in the first message summary list and its corresponding level and message topic". In an optional embodiment, using the first large model to extract content from the offline session messages based on the profile information of the first session to obtain each message summary in the first message summary list and its corresponding level and message topic includes: using the first large model to perform semantic clustering and content analysis on the offline session messages of the first session to obtain at least one message summary; for each message summary, assigning weight information to each offline session message in the current message summary based on the profile information of the first session, the weight information representing the degree of relevance between the corresponding offline session message and the profile information; calculating the user input score of the current message summary based on the weight information; determining the number of levels of the message topic of the current message summary based on the user input score; and extracting content from the offline session messages corresponding to the current message summary based on the number of levels to generate message topics for each level.

[0082] In this embodiment of the application, the message digest of the first session can be understood as the result obtained by semantic clustering and content extraction of all offline session messages of the first session.

[0083] After obtaining the message digest of the first session, weight information is assigned to each offline session message in each message digest to map the macroscopic user preferences represented by the profile information to each offline session message constituting each message digest. This application does not limit the implementation of "assigning weight information to each offline session message in the current message digest based on the profile information of the first session." In one implementation, assigning weight information to each offline session message in the current message digest based on the profile information of the first session may include: the server calling a first large model to vectorize the profile information, the current message digest of the first session, and the corresponding offline session message, and performing feature fusion on the vectorized data through a cross-attention mechanism; predicting the weight information based on the fusion result. Preferably, a higher weight indicates a higher relevance of the offline session message to the user in the current message digest; conversely, a lower weight indicates a lower relevance.

[0084] After obtaining the weight information, a user engagement score for the current message digest can be calculated based on that weight information. The user engagement score can refer to an assessment of the attractiveness of the current message digest to the user. This application does not limit the implementation method of "calculating the user engagement score for the current message digest based on weight information." In one implementation, the user engagement score can be obtained by calculating the average of the weight information of all offline session messages associated with the current message digest. In another implementation, a machine learning model can be trained to learn the relationship between the distribution patterns of various sample weight information and the sample user engagement scores, and this relationship can be optimized based on the user's actual engagement with the message digest, thereby accurately predicting the user engagement score for the current message digest.

[0085] In one optional embodiment, the priority of each message digest retrieval and its display order in the first session interface can be determined based on the user's input rating. Generally, the higher the user's input rating, the more levels may be divided, the higher the retrieval priority, and the earlier it is displayed in the first session interface.

[0086] After obtaining the user engagement score, the number of message topic levels in the current message digest can be determined. This application does not limit the implementation method. In one implementation, a set of segmented intervals for user engagement scores can be predefined, and a unique corresponding number of message topic levels can be preset for each interval. During runtime, after calculating the user engagement score corresponding to the current message digest, it can be determined which preset interval the score falls into. Subsequently, the number of levels corresponding to that interval can be determined as the number of levels the current message digest should have. In an optional embodiment, this decision-making process can also introduce constraints, ensuring that the final determined number of levels does not exceed the maximum reasonable level depth that the current message digest can support, as evaluated by the first major model, to ensure the effectiveness of the generated hierarchical structure.

[0087] After calculating the number of levels, content extraction can be performed on the offline session messages corresponding to the current message digest based on the calculated number of levels to generate message topics for each level. This application embodiment does not limit the implementation method. In one implementation, the server calls the first major model to perform content understanding and clustering on all offline session messages corresponding to the current message digest to obtain the message topic for the first level. Assuming the current number of levels is 1, if the current number of levels is less than the total number of levels, the server calls the first major model to perform content understanding and clustering on all offline session messages corresponding to each message topic in the current level to obtain the message topic for the next level, and increments the current number of levels by 1. It then checks if the current number of levels is less than the total number of levels; if it is, the message topic for the next level is used as the message topic for the current level, and the above process is repeated until the current number of levels matches the calculated number of levels. Wherein, when the current number of levels is 1, the current level is the first level.

[0088] In an optional embodiment, content understanding and clustering of all offline session messages corresponding to the message digest can be omitted. Instead, offline session messages with weights greater than a preset weight threshold can be selected based on the weight information of each offline session message calculated in the above process. The first large model can then be used to process the selected offline session messages to obtain the message topic. In this embodiment, the amount of data processed by the first large model can be reduced, improving the efficiency of generating message topics. Furthermore, the message topic reflects the user's core interests and is representative, avoiding wasting time on irrelevant information. By combining the message digest generation mechanism with a client-interruptible hybrid retrieval strategy, the improved decision-making efficiency brought by message digests can be instantly transformed into improved data transmission efficiency and reduced user experience latency, thereby improving the overall performance of offline message processing.

[0089] In one optional embodiment, the server can deploy multiple second-large models, each associated with a session and responsible for recommendations for that session. Each session's second-large model can be trained on historical session messages within its assigned session, thereby gaining a deeper understanding of the session's context and semantics, and generating personalized recommendations based on this understanding, thus improving recommendation efficiency. Calling the second-large model to recommend a second message summary list based on the first message summary includes: calling the second-large model corresponding to the first session to calculate the similarity between the message topics of the first message summary and the message topics of other sessions; and recommending message topics from other sessions with similarity greater than a similarity threshold as second message summaries to obtain the second message summary list.

[0090] This application does not limit the implementation of "calling the second major model corresponding to the first session to calculate the similarity between the message topic of the first message digest and the message topic of other sessions". In one implementation, the second major model corresponding to the first session is called to extract features from the message topic of the first message digest and the message topics of other sessions, generating semantic feature vectors corresponding to the message topic of the first message digest and the message topics of other sessions; based on the semantic feature vectors, the similarity between the message topic of the first message digest and the message topics of other sessions is calculated.

[0091] In this embodiment, by combining the second major model with the first major model employing a scalable message digest generation mechanism, the server can adjust the generated second message digest based on the personalized and global attribute information of the first message digest. For example, the server can invoke the second major model to adjust the recommendation priority of message digests in other sessions that match the personalized attributes based on the personalized attributes; it can also invoke the second major model to adjust the recommended content based on global attribute information, such as filtering unsuitable content formats based on network status or adjusting the level of detail of the recommended content based on the interaction period, to ensure that the recommendation results are adapted to the user's current context, achieving the effect of adaptive optimization of the second major model. Furthermore, by configuring one second major model for each session, the recommendation and adjustment processes of the second major model in different sessions are independent and do not affect each other, which can improve the efficiency of adaptive optimization of the second major model and the accuracy of recommendations.

[0092] In one optional embodiment, the second major model includes one or more logical relationship classifiers. The second major model corresponding to the first session is invoked to calculate the similarity between the message topics of the first message digest and the message topics of other sessions. This includes: invoking the second major model corresponding to the first session to extract features from the message topics of the first message digest and the message topics of other sessions, obtaining a first encoded representation and a second encoded representation; inputting the first encoded representation and the second encoded representation into the logical classifier to obtain the probability distribution corresponding to the logical relationship associated with each logical classifier; and fusing the various probability distributions to obtain the similarity between the message topics of the first message digest and the message topics of other sessions.

[0093] The first encoding refers to the semantic feature vector obtained by extracting features from the message topics of the first message digest, and the second encoding refers to the semantic feature vector obtained by extracting features from the message topics of other sessions. The logical relation classifier can refer to a neural network module within the second model, used to receive the semantic feature vectors corresponding to the message topics of the first message digest and the semantic feature vectors corresponding to the message topics of any other session. Through its internal neural network analysis, it determines whether a predefined logical relationship exists between the two input message topics and outputs a probability distribution. For example, the logical relation classifier can be a problem-solution relationship classifier, an opinion comparison relationship classifier, etc. The probability distribution can refer to the numerical confidence level of the logical relation classifier in whether the logical relationship it is responsible for is valid. For example, a binary logical relation classifier might output [0.85, 0.15], which indicates that it has an 85% confidence level that the logical relationship is valid.

[0094] To obtain a comprehensive similarity score, the probability distributions from multiple different logical relation classifiers can be fused. This application does not limit the implementation method of fusing the various probability distributions. In one implementation, an importance weight can be preset for each logical relation, and then the probability values ​​of the relation being true output by each classifier can be weighted and summed to obtain the similarity score. In another implementation, a logical relation learning model can be trained, taking the probability distributions of multiple logical relation classifiers as input, and the logical relation learning model automatically learns the non-linear dependencies between the probabilities of different logical relations to obtain the similarity score. In another implementation, a cross-attention mechanism can be used to fuse the various probability distributions. Specifically, the first and second encoding representations can be aggregated to obtain a context query vector. This aggregation operation can include concatenation, element-wise addition or multiplication, pooling, etc. Logical relationship embedding vectors are generated for each logical classifier's associated logical relationships. The probability values ​​of the logical relationships output by each logical classifier are obtained. The cross-attention mechanism is used to process the context query vector and each logical relationship embedding vector to obtain fusion weights. The probability values ​​of the logical relationships are weighted and summed according to the fusion weights to obtain a similarity score. In this implementation, the fusion weights are dynamic, reflecting the importance of different logical relationships for the two currently input message topics. During the weighted summation of the probability values ​​of the logical relationships based on the fusion weights, higher weights can be assigned to logical relationships with higher fusion weights, and lower weights to logical relationships with lower fusion weights. This dynamic, context-aware weighting mechanism can adapt to various different input message topics and can deeply analyze the association between the two currently input message topics, improving the accuracy of similarity.

[0095] In an optional embodiment, the first message digest includes multiple message topics. After sending the second message digest list to the client, the method further includes: receiving a fourth message request sent by the client, the fourth message request including the currently selected message topic at the current level; obtaining corresponding context information based on the currently selected message topic at the current level, the context information including message topics at each level selected historically; calling the second major model corresponding to the first session, and selecting a message topic that is the same as or similar to the currently selected message topic at the current level from the second message digest based on the context information to obtain the second message topic; and sending the second message topic to the client so that the user can view other offline session messages with the same or similar message topics in other sessions through the second message topic.

[0096] The fourth message request can refer to a request for a message topic that is the same as or similar to the currently selected message topic at the current level. Contextual information may include message topics selected by the user at the previous level or several previous levels, as well as recommended message topics. In this embodiment, through contextual information, a comprehensive understanding of the user's intent can be achieved. When the second major model is invoked to execute the recommendation process, the resulting recommendation not only matches the currently selected message topic at the current level but also ensures that the recommendation result remains highly relevant to the preceding information, providing a logically coherent information exploration path.

[0097] In this application embodiment, the first and second major models are not merely tools, but core technologies that give this application embodiment its non-obvious characteristics. They can be understood as the core engine of this application embodiment. The first and second major models reconstruct the offline session message processing method from two dimensions: message description and message association, respectively. The first major model solves the technical problem of how to present messages in an efficient and personalized way, moving beyond passively compressing or summarizing the original text to converting complex offline session messages into adaptive message summaries for each session. The second major model solves the technical problem of how to make accurate recommendations based on the understood offline session messages. It receives the message summaries constructed by the first major model and uses them as a basis to transform the recommendation task from traditional content matching based on surface similarity to recommendation based on deep semantics and logical relationships. Through the combination of the first and second major models, passive data retrieval and presentation can be transformed into an information service process that can predict user intent and provide guided message discovery.

[0098] In one alternative embodiment, a first large model and a second large model can be pre-trained. In one implementation, the first large model can be pre-trained on offline conversation message sample data using a large language model to optimize its content understanding and message summarization capabilities. The second large model can be pre-trained on message summarization sample data using a large language model to optimize its message summarization recommendation capabilities, where each message summarization sample data includes at least two message summaries and labels representing the relationships between them. In another implementation, the first and second large models can be jointly trained, i.e., message summarization generation and recommendation can be optimized simultaneously under a unified large model. During training, the output of the first large model can be used as the input of the second large model. The second large model processes the input data to obtain recommendation results; calculates the difference between the message summaries generated by the first large model and a first true value to obtain a summary generation loss value; calculates the difference between the recommendation results output by the second large model and a second true value to obtain a recommendation loss value; and updates the parameters of the first and second large models based on the summary generation loss value and the recommendation loss value. Through joint optimization, the optimization of the recommended message digest of the second model can guide the message digest generation process of the first model, enabling the first model to generate message digests that are not only accurate in content but also beneficial to the recommendation of the second model, thereby improving overall performance.

[0099] In one optional embodiment, a mini-batch stochastic gradient descent method can be used to divide offline session message sample data and / or message digest sample data into multiple batches for iterative training. For each batch of sample data, the loss values ​​of the first and / or second largest models are calculated, and the gradient is calculated using the backpropagation algorithm based on the loss values. This gradient represents the direction of parameter adjustment. The parameters of the first and second largest models are updated based on the gradient. By repeating the above process multiple times, the parameters of the first and / or second largest models can be gradually optimized towards minimizing the loss function, thereby improving the accuracy of model training.

[0100] In one alternative embodiment, the training process of the first and second largest models can be carried out in a multi-GPU parallel computing environment. Each GPU can process a batch of data in parallel and independently compute gradients. At the end of an iteration step, the gradients computed by all GPUs can be aggregated synchronously or asynchronously, and the aggregated gradients can be used to uniformly update the parameters of the first and / or second largest models, thereby shortening the model training time.

[0101] In an optional embodiment, the first and second main models can employ an encoder-decoder architecture. The encoder can refer to a neural network module that performs deep feature extraction and context encoding on the input data, and may include an attention module and a feedforward neural network. In this embodiment, the encoder is responsible for receiving the input offline session message or message summary, and performing word segmentation to obtain a labeled sequence; the attention module captures the relationship between the labeled sequences, and the feedforward neural network processes this relationship to obtain a vector representation. The decoder may also include an attention module and a feedforward neural network, and is responsible for generating a message summary or recommended message summary based on the encoder's output. In this decoder, a cross-attention mechanism can be used to generate the message summary and recommended message summary, as described above, and will not be elaborated further here.

[0102] In one alternative embodiment, residual connections and layer normalization can be added after each attention module and feedforward neural network. This design helps alleviate the vanishing gradient problem in deep neural networks and accelerates model convergence. In another alternative embodiment, the feedforward neural network may include a multilayer perceptron, which performs a nonlinear transformation on the features output by the attention module to enhance the model's representational power.

[0103] In one optional embodiment, the first major model may be associated with a first prompt word, and the second major model may be associated with a second prompt word; the server guides the first major model to generate a message digest based on the first prompt word, and guides the second major model to recommend a message digest based on the second prompt word.

[0104] Figure 3 This is a flowchart illustrating a session message processing method, provided as another exemplary embodiment of this application. Figure 3 As shown, this method is applied to a server, and the method includes:

[0105] S300: After re-establishing the network connection with the client, receive the first message request sent by the client. The first message request includes the identifier of the first session.

[0106] S310. Based on the identifier of the first session, obtain the first message summary list corresponding to the real-time session message and the offline session message corresponding to the first session, and send it to the client for display in the first session interface corresponding to the first session; the first message summary list includes the message summary obtained by extracting the content of the offline session message of the first session using the first big model during the client's offline period.

[0107] S320: Receive a second message request sent by the client, the second message request including a first message digest obtained by selecting message digests from the first message digest list.

[0108] S330. The second major model is invoked to recommend a list of second message summaries based on the first message summary. The list of second message summaries includes at least one second message summary with a message subject that is the same as or similar to the first message summary. The second message summaries are obtained by extracting content from offline session messages of other sessions using the first major model during client offline time.

[0109] S340. Send a second message digest list to the client so that the user can view other offline session messages with the same or similar message topics in other sessions through the second message digest in the second message digest list.

[0110] In one optional embodiment, each message digest in the first message digest list includes multiple message topics. During client offline periods, the first major model is used to extract content from the offline session messages of the first session to obtain each message digest in the first message digest list. This includes: constructing profile information for the first session, which includes personalized attribute information obtained based on historical interaction data with the first session; and, based on the profile information of the first session, calling the first major model to extract content from the offline session messages to obtain each message digest in the first message digest list, as well as the corresponding level and message topic.

[0111] In one optional embodiment, based on the profile information of the first session, a first large model is invoked to extract content from offline session messages to obtain the hierarchy of each message summary in the first message summary list and each message topic in each message summary. This includes: invoking the first large model to perform semantic clustering and content analysis on the offline session messages of the first session to obtain at least one message summary; for each message summary, based on the profile information of the first session, assigning weight information to each offline session message in the current message summary, where the weight information represents the degree of relevance between the corresponding offline session message and the profile information; calculating the user engagement score of the current message summary based on the weight information; determining the number of levels of the message topics in the current message summary based on the user engagement score; and extracting content from the offline session messages corresponding to the current message summary based on the number of levels to generate message topics at each level.

[0112] In one optional embodiment, the server deploys multiple second-large models, each associated with a session and responsible for recommendations for the corresponding session. Calling the second-large model to recommend a list of second message summaries based on the first message summary includes: calling the second-large model corresponding to the first session to calculate the similarity between the message topic of the first message summary and the message topics of other sessions; recommending the message topics of other sessions with similarity greater than a similarity threshold as second message summaries to obtain a list of second message summaries.

[0113] In one optional embodiment, the second major model includes one or more logical relationship classifiers. The second major model corresponding to the first session is invoked to calculate the similarity between the message topics of the first message digest and the message topics of other sessions. This includes: invoking the second major model corresponding to the first session to extract features from the message topics of the first message digest and the message topics of other sessions, obtaining a first encoded representation and a second encoded representation; inputting the first encoded representation and the second encoded representation into the logical classifier to obtain the probability distribution corresponding to the logical relationship associated with each logical classifier; and fusing the various probability distributions to obtain the similarity between the message topics of the first message digest and the message topics of other sessions.

[0114] In an optional embodiment, the first message digest includes multiple message topics. After sending the second message digest list to the client, the method further includes: receiving a fourth message request sent by the client, the fourth message request including the currently selected message topic at the current level; obtaining corresponding context information based on the currently selected message topic at the current level, the context information including message topics at each level selected historically; calling the second major model corresponding to the first session, and selecting a message topic that is the same as or similar to the currently selected message topic at the current level from the second message digest based on the context information to obtain the second message topic; and sending the second message topic to the client so that the user can view other offline session messages with the same or similar message topics in other sessions through the second message topic.

[0115] In some of the processes described in the above embodiments and accompanying drawings, multiple operations are included that appear in a specific order. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or they may be executed in parallel. The operation numbers, such as S1, S2, etc., are merely used to distinguish different operations and do not represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first" and "second" in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0116] Figure 4 This is a schematic diagram of the structure of a session message processing apparatus provided as another exemplary embodiment of this application. For example... Figure 4 As shown, the device includes a memory 54, a processor 55, and a communication component 56.

[0117] Memory 54 is used to store computer programs and can be configured to store various other data to support operation on the processing device. Examples of this data include instructions for any application or method operating on the processing device, contact data, phone book data, messages, pictures, videos, etc.

[0118] The memory 54 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0119] Processor 55, coupled to memory 54, executes a computer program in memory 54 for: upon client re-entry, responding to an access operation to the first session via communication component 56, displaying the first session interface corresponding to the first session, and sending a first message request to the server to request message summaries corresponding to the real-time session messages and offline session messages of the first session; receiving and displaying a list of first message summaries returned by the server based on the first message request, the list including message summaries obtained by the server extracting content from offline session messages of the first session using a first major model during client offline time; responding to a selection operation of the first message summaries in the list, displaying offline session messages associated with the first message summaries, and sending a second message request to the server for the server to call a second major model to recommend and return a list of second message summaries based on the first message summaries; the list of second message summaries includes at least one second message summary with a message topic that is the same as or similar to the first message summaries, the second message summaries being obtained by the server extracting content from offline session messages of other sessions using the first major model during client offline time; and displaying the list of second message summaries returned by the server for the user to view other offline session messages with the same or similar message topics in other sessions through the second message summaries in the list.

[0120] The processor 55 can also perform other operations, which can be found in the descriptions in the foregoing embodiments and will not be repeated here.

[0121] Furthermore, such as Figure 4 As shown, the processing device also includes a display 57, a power supply unit 58, an audio unit 59, and other components. Figure 4 The diagram only shows some components and does not imply that the processing device includes only these components. Figure 4 The components shown.

[0122] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed, can perform the steps that can be executed by a processing device in the above method embodiments.

[0123] The above Figure 4The communication component is configured to facilitate wired or wireless communication between the device containing the communication component and other devices. The device containing the communication component can access wireless networks based on communication standards, such as WiFi, 2G, 3G, 4G / LTE, 5G, or combinations thereof. In one exemplary embodiment, the communication component receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, the communication component further includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID), Infrared Data Association (IrDA) technology, Ultra-Wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0124] The above Figure 4 The display includes a screen, which may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen can be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors can sense not only the boundaries of the touch or swipe action, but also the duration and pressure associated with the touch or swipe operation.

[0125] The above Figure 4 The power supply component provides power to the various components of the device in which it resides. The power supply component may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which it resides.

[0126] The above Figure 4 The audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC) configured to receive external audio signals when the device containing the audio component is in an operating mode, such as call mode, recording mode, or voice recognition mode. The received audio signals can be further stored in memory or transmitted via a communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0127] 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.

[0128] 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.

[0129] 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.

[0130] 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.

[0131] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0132] 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.

[0133] 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.

[0134] 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 a process, method, article, or apparatus. Without further limitation, 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 said element.

[0135] The above description is merely an embodiment of this application and is 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. A method for processing session messages, characterized in that, Applied to a client, the method includes: After the client comes back online, in response to the access operation of the first session, the first session interface corresponding to the first session is displayed, and a first message request is sent to the server to request the message digests corresponding to the real-time session message and the offline session message of the first session. Receive and display a list of first message digests returned by the server in accordance with the first message request. The list of first message digests includes message digests obtained by the server extracting content from offline session messages of the first session using a first large model during the client's offline period. In response to the selection operation of the first message digest in the first message digest list, the offline session messages associated with the first message digest are displayed, and a second message request is sent to the server so that the server can call the second major model to recommend and return a second message digest list based on the first message digest; the second message digest list includes at least one second message digest with a message topic that is the same as or similar to the first message digest, and the second message digest is obtained by the server using the first major model to extract the content of offline session messages of other sessions during the client's offline period; The server returns a list of the second message digests, allowing users to view other offline session messages with the same or similar message topics from other sessions through the second message digests in the list.

2. The method according to claim 1, characterized in that, After the client comes back online, the method further includes: Based on the session priorities maintained by the task scheduler, the task scheduler schedules the background service to obtain message digests corresponding to each session and offline session messages corresponding to each message digest from the server, and caches them locally on the client for display in the session interface corresponding to each session. The step of displaying offline session messages associated with the first message digest in response to a selection operation on the first message digest list includes: In response to the selection operation of the first message digest in the first message digest list, the offline session message corresponding to the first message digest is queried from the local cache of the client. If the first message digest is not found, the task scheduler creates a first task with the highest priority based on the first message digest, and the background service is scheduled to retrieve and display the offline session message associated with the first message digest from the server based on the first task.

3. The method according to claim 1, characterized in that, The first message digest includes multi-level message topics, and the first message digest list is displayed in the first session interface in the form of a display container; The step of displaying offline session messages associated with the first message digest in response to a selection operation on the first message digest list includes: In response to a selection operation of a first message digest in the first message digest list, the first-level message topic is displayed in the area associated with the first message digest within the display container; In response to a selection operation on any message topic in the current level message topics, the next level message topics are displayed in the area associated with the first message digest until the last level message topic is displayed; wherein, the current level message topic is either the first level message topic or a non-first level message topic; In response to the selection operation of any message topic in the last-level message topics, the offline session messages associated with the currently selected last-level message topics are obtained and displayed; wherein, the offline session messages corresponding to the unselected message topics are not obtained from the local cache of the client or the server.

4. The method according to claim 3, characterized in that, Following a selection operation on any message topic in the last-level message topics, the method further includes: A third message request is sent to the server so that the server can match the message topics that are the same as or similar to the message topics of the currently selected last level from the second message digest to obtain a first candidate message topic list, obtain the offline session message associated with the first candidate message topic list and return it to the client; Receive offline session messages associated with the first candidate message topic list, and display them in the first session interface along with the offline session messages associated with the currently selected last-level message topic.

5. The method according to any one of claims 1-4, characterized in that, The first session interface includes operation controls associated with the displayed offline session messages, and the operation controls include a delete control; After displaying the offline session message associated with the first message digest, the method further includes: In response to the selection operation of the delete control associated with the offline session message associated with the first message digest, the offline session message associated with the first message digest in the first session interface is deleted, and the offline session message associated with the first message digest cached locally on the client is also deleted.

6. A method for processing session messages, characterized in that, Applied to a server, the method includes: After re-establishing a network connection with the client, the system receives a first message request sent by the client, the first message request including an identifier of the first session; Based on the identifier of the first session, a first message digest list corresponding to the real-time session message and the offline session message corresponding to the first session is obtained and sent to the client for display in the first session interface corresponding to the first session; the first message digest list includes message digests obtained by extracting content from the offline session message of the first session using the first big model during the client's offline period; Receive a second message request sent by the client, the second message request including a first message digest obtained by selecting message digests from the first message digest list; The second major model is invoked to recommend a second message digest list based on the first message digest. The second message digest list includes at least one second message digest with a message topic that is the same as or similar to the first message digest. The second message digest is obtained by extracting content from offline session messages of other sessions using the first major model during the client's offline period. The second message digest list is sent to the client so that the user can view other offline session messages with the same or similar message topics in other sessions through the second message digest in the second message digest list.

7. The method according to claim 6, characterized in that, Each message digest in the first message digest list includes multiple message topics. During the client's offline period, the first large model is used to extract the content of the offline session messages of the first session to obtain each message digest in the first message digest list, including: Construct profile information for the first session, the profile information including personalized attribute information obtained based on historical interaction data of the first session; Based on the profile information of the first session, the first large model is invoked to extract the content of the offline session messages to obtain each message summary in the first message summary list, as well as the corresponding level and message topic.

8. The method according to claim 7, characterized in that, The step of extracting content from the offline session messages based on the profile information of the first session, calling the first large model to obtain the hierarchy of each message digest in the first message digest list and the message topics in each message digest, includes: The first large model is invoked to perform semantic clustering and content analysis on the offline session messages of the first session to obtain at least one message digest; For each message digest, weight information is assigned to each offline session message in the current message digest based on the profile information of the first session. The weight information represents the degree of relevance between the corresponding offline session message and the profile information. Based on the weight information, a user engagement score for the current message digest is calculated. Based on the user engagement score, the number of message topics in the current message digest is determined. Based on the number of levels, content is extracted from the offline session messages corresponding to the current message digest to generate message topics for each level.

9. The method according to claim 6, characterized in that, The server deploys multiple second-large models, each associated with a session and responsible for recommendations for that session. The process of calling a second-large model to recommend a second message digest list based on the first message digest includes: The second major model corresponding to the first session is invoked to calculate the similarity between the message topic of the first message digest and the message topics of the other sessions; the message topics of the other sessions with similarity greater than the similarity threshold are recommended as the second message digest to obtain the second message digest list.

10. The method according to claim 9, characterized in that, The second major model includes one or more logical relation classifiers. The step of calling the second major model corresponding to the first session to calculate the similarity between the message topic of the first message digest and the message topics of the other sessions includes: The second major model corresponding to the first session is invoked to extract features from the message topic of the first message digest and the message topics of the other sessions to obtain a first encoding representation and a second encoding representation. The first encoding representation and the second encoding representation are input into the logic classifier to obtain the probability distribution corresponding to the logical relationship associated with each logic classifier; By fusing the various probability distributions, the similarity between the message topic of the first message digest and the message topics of the other sessions is obtained.

11. The method according to any one of claims 6-10, characterized in that, The first message digest includes multiple message topics. After sending the second message digest list to the client, the method further includes: Receive a fourth message request sent by the client, the fourth message request including the message topic of the currently selected current level; Based on the currently selected message topic at the current level, obtain the corresponding context information, which includes the message topics at each previously selected level. The second major model corresponding to the first session is invoked, and a message topic that is the same as or similar to the currently selected message topic of the current level is selected from the second message digest to obtain the second message topic, based on the context information. The second message topic is sent to the client so that the user can view other offline session messages with the same or similar message topics in other sessions through the second message topic.

12. An electronic device, characterized in that, include: Memory and processor; The memory is used to store a computer program; the processor is coupled to the memory and is used to execute the computer program to implement the steps of the method according to any one of claims 1-11.

13. A computer-readable storage medium storing a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, the processor is enabled to perform the steps of the method according to any one of claims 1-11.

14. A computer program product, characterized in that, include: A computer program / instruction that, when executed by a processor, causes the processor to perform the steps of the method according to any one of claims 1-11.

Images