Group Interaction Message Scheduling and Context Management Method, Apparatus, Equipment and Medium

Abstract

This invention provides a method, apparatus, device, and medium for group interaction message scheduling and context management, relating to the field of multi-agent collaboration technology. It receives raw group chat messages, categorizes and tags them, stores them in a buffer window for dynamic sorting, and generates a set of messages to be scheduled. Message tags are matched against a rule base; if a match is successful, deterministic scheduling is performed according to a preset level; if a match fails, relevant context is obtained and submitted to a large language model for processing. The scheduling results are associated and stored in the corresponding nodes of a hierarchical context tree, and reverse pruning is performed when pruning conditions are met. User interaction data is collected to calculate a comprehensive participation score; if the score is below a threshold, an intervention instruction is generated. Different data tables are created for the same scheduling event, and associated identifiers are established. This invention achieves high real-time and low-cost response; overcomes the defects of message disorder, ambiguous context, and lack of proactive intervention; enables cross-scenario reuse of scheduling capabilities; and supports bidirectional evaluation evidence archives.

Description

Technical Field

[0001] This invention relates to the field of multi-agent collaboration technology, and in particular to a method, apparatus, device, and medium for group interaction message scheduling and context management. Background Technology

[0002] For online collaborative training scenarios involving multiple users and agents (such as group engineering thinking training, cross-group collaborative tasks, and step-by-step skill training), the underlying communication and scheduling mechanisms mainly follow two technical paths: First, a flat, "first-come, first-served" message distribution mechanism based on a general instant messaging (IM) protocol. The core logic of this approach is to maintain a global or group-level message queue, where messages sent by all participants (including human users and agents) are distributed to all members of the group chat in strict order of arrival. Second, the linear collaboration module in traditional learning management systems (LMS). This approach pre-defines the collaborative training process as a fixed, linear workflow, such as the "post-reply" mode in a forum module or the "submit-review" mode in a group task module. Its essence is serial asynchronous interaction, rather than real-time group chat-style collaboration. The "first-come, first-served" flat message distribution mechanism adopts an indiscriminate timeline stacking method. Although simple to implement, it exposes some problems in high-concurrency, highly interactive collaborative training scenarios. For example, the system lacks the ability to understand the semantics and priority of message content, resulting in delayed or even completely ignored responses to key instructions. This causes the behavior of the agent generated by the LLM (Large Language Model) to deviate from the preset engineering logic constraints. Secondly, in terms of context management, in order to maintain the coherence of multi-turn dialogues so that the agent can understand the current situation, the existing system backtracks the entire historical record and directly splices the complete log of the entire group chat into the context window of the LLM. This causes the number of tokens in the context window to grow exponentially with the training time, which not only causes high API call costs but also increases the processing latency of the model. The linear collaboration module in traditional learning management systems (LMS) pre-defines the collaborative training process as a fixed, linear workflow. The system can only record surface behavioral data such as the number of times a user speaks, the length of a post, and the task completion status. It is completely unable to deeply analyze the semantic contribution of each user's speech during the collaboration process. Therefore, the system cannot identify the "free-rider" phenomenon (i.e., members who do not actively participate in substantive discussions, but only passively listen or send low-quality content) in real time during the training process, and it cannot automatically trigger "cognitive intervention" actions to dynamically adjust the group's collaborative status and participation structure. Summary of the Invention

[0003] This invention provides a group interaction message scheduling and context management method, apparatus, device and medium to solve the defects of traditional multi-agent scheduling methods, such as the lack of semantic priority judgment leading to key instruction blocking and context memory degradation, the inability to perceive individual semantic contribution in real time to trigger active intervention, and the low degree of data structuring leading to difficulties in subsequent evaluation.

[0004] This invention provides a method for scheduling and managing group interaction messages and context, comprising: Receive raw group chat messages, classify and label the raw group chat messages, store the labeled messages in a buffer window for dynamic sorting, and generate a message set to be scheduled. Outputting scheduling execution results based on the set of messages to be scheduled includes: matching the message tag of each message in the set of messages to be scheduled with a preset scheduling rule base; if the match is successful, the message is determined to be a deterministic instruction and the scheduling is performed according to a preset scheduling level; if the match fails, the message is determined to be a non-deterministic instruction, the context information associated with the current message is retrieved from the hierarchical context tree, the context information and the non-deterministic instruction are submitted to the large language model for processing, and the scheduling is performed based on the processing result. The scheduling execution result is associated and stored in the corresponding hierarchical node in the hierarchical context tree. When the preset pruning conditions are met, reverse pruning is performed on the hierarchical context tree. Collect user interaction behavior data during the training process, quantify the features of the interaction behavior data, and calculate a comprehensive participation score based on the quantified features; when the comprehensive participation score is lower than a preset threshold, generate an intervention scheduling instruction, and update the message set to be scheduled based on the intervention scheduling instruction; Create different data tables for the same scheduling event, and establish the same association identifier for the data in the different data tables.

[0005] According to the group interaction message scheduling and context management method provided by the present invention, the steps of receiving original group chat messages, classifying and tagging the original group chat messages, storing the tagged messages in a buffer window for dynamic sorting, and generating a message set to be scheduled include: Receive raw group chat messages from the group chat client, parse each raw group chat message and extract message attributes; Based on the message attributes, the original group chat messages are classified and labeled as at least one of the following types: training instructions, interactive feedback, system notifications, or other types. The tagged messages are stored in a buffer window and sorted according to priority weight and timestamp. The priority weight is pre-configured according to the message type, and the priority weight of training instructions is higher than that of interactive feedback. The queue length parameter of the buffer window is dynamically adjusted according to a preset adjustment coefficient based on the current queue length of the buffer window and a preset queue length threshold. Messages are extracted from the dynamically sorted buffer window to generate a message set to be scheduled.

[0006] According to the group interaction message scheduling and context management method provided by the present invention, the step of storing the scheduling execution result in association with the corresponding hierarchical node in the hierarchical context tree, and performing reverse pruning on the hierarchical context tree when the preset pruning conditions are met, includes: The training level information is extracted from the scheduling execution result, and the training level information includes task identifier, stage identifier, group identifier and user identifier; Construct or update a hierarchical context tree based on four dimensions: task layer, stage layer, team layer, and user layer; The message content in the scheduling execution result is associated and stored in the corresponding hierarchical node of the hierarchical context tree; Real-time monitoring of the validity period of nodes at each level in the hierarchical context tree; When any node at any level exceeds the preset validity period threshold, the node is marked as a node to be pruned. When the pruning trigger condition is met, reverse pruning is performed on the marked nodes to be pruned, prioritizing the deletion of context information from the user layer and group layer, while retaining the core context information from the task layer and stage layer, and outputting the updated hierarchical context tree.

[0007] According to the group interaction message scheduling and context management method provided by the present invention, the pruning triggering condition includes at least one of the following: The total amount of context information stored in the hierarchical context tree exceeds the preset token threshold; The validity period of any node at any level exceeds the corresponding validity period threshold; A context compression request was received from the large language model.

[0008] According to the group interaction message scheduling and context management method provided by the present invention, the step of quantifying the interaction behavior data and calculating a comprehensive participation score based on the quantified features includes: The collected interactive behavior data are mapped to a unified quantitative scoring range to obtain various quantitative feature values; Based on the preset importance of each interactive behavior data, a corresponding weight coefficient is assigned to each quantitative feature value, and the sum of the weight coefficients is 1; The overall participation score is calculated by summing the products of each quantitative feature value and its corresponding weight coefficient.

[0009] According to the group interaction message scheduling and context management method provided by the present invention, when the comprehensive participation score is lower than a preset threshold, an intervention scheduling instruction is generated, and the message set to be scheduled is updated based on the intervention scheduling instruction, including: The calculated overall engagement score is compared with a preset engagement threshold. When the overall participation score is lower than the preset threshold, the intervention level is determined based on the degree of score deviation. According to the intervention level, a corresponding intervention scheduling instruction is generated, and the intervention scheduling instruction includes the target user identifier, intervention type, and intervention content; The generated intervention scheduling instructions are encapsulated into a standard message format and assigned scheduling priorities; The set of messages to be scheduled is updated based on the encapsulated intervention scheduling instructions, so that the intervention scheduling instructions participate in subsequent scheduling execution.

[0010] According to the group interaction message scheduling and context management method provided by the present invention, the step of creating different data tables for the same scheduling event and establishing the same association identifier for the data in the different data tables includes: A first data table is created for the scheduling logs generated during the scheduling execution process. The scheduling logs include at least message identifier, scheduling timestamp, source sender, target receiver, scheduling level, and scheduling status. A second data table is created for the hierarchical context tree generated by the hierarchical context management step, the intervention records generated by the participation measurement and intervention steps, and the training content; Generate the same association identifier for the scheduling log, hierarchical context tree, intervention record, and training content corresponding to the same scheduling event; Write the association identifier into the first data table and the second data table respectively to establish a mapping relationship between the first data table and the second data table; The data in the first data table and the second data table are synchronously written to the persistent storage medium.

[0011] This invention provides a group interaction message scheduling and context management device, comprising: The receiving module is used to receive raw group chat messages, classify and label the raw group chat messages, store the labeled messages in a buffer window for dynamic sorting, and generate a message set to be scheduled. The output module is used to output the scheduling execution result based on the set of messages to be scheduled, including: matching the message tag of each message in the set of messages to be scheduled with a preset scheduling rule base; if the match is successful, the message is determined to be a deterministic instruction and the scheduling is performed according to a preset scheduling level; if the match fails, the message is determined to be a non-deterministic instruction, the context information associated with the current message is retrieved from the hierarchical context tree, the context information and the non-deterministic instruction are submitted to the large language model for processing, and the scheduling is performed based on the processing result. The pruning module is used to associate and store the scheduling execution result with the corresponding hierarchical node in the hierarchical context tree, and to perform reverse pruning on the hierarchical context tree when the preset pruning conditions are met. The intervention module is used to collect user interaction behavior data during the training process, quantify the features of the interaction behavior data, and calculate a comprehensive participation score based on the quantified features; when the comprehensive participation score is lower than a preset threshold, an intervention scheduling instruction is generated, and the message set to be scheduled is updated based on the intervention scheduling instruction. A creation module is used to create different data tables for the same scheduling event and establish the same association identifier for the data in the different data tables.

[0012] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the group interaction message scheduling and context management method as described in any of the preceding claims.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the group interaction message scheduling and context management method described in any of the preceding claims.

[0014] The group interaction message scheduling and context management method, apparatus, device, and medium provided by this invention receive original group chat messages, classify and label the original group chat messages, store the labeled messages in a buffer window for dynamic sorting, and generate a message set to be scheduled; output the scheduling execution result according to the message set to be scheduled, including: matching the message tag of each message in the message set to be scheduled with a preset scheduling rule base; if the match is successful, the message is determined to be a deterministic instruction and the scheduling is executed according to a preset scheduling level; if the match fails, the message is determined to be a non-deterministic instruction, the context information associated with the current message is retrieved from the hierarchical context tree, the context information and the non-deterministic instruction are submitted to a large language model for processing, and the scheduling is executed according to the processing result; The scheduling execution results are associated and stored in the corresponding hierarchical nodes in the hierarchical context tree. When the preset pruning conditions are met, the hierarchical context tree is reverse-pruned. User interaction behavior data during the training process is collected, and the interaction behavior data is quantified. A comprehensive participation score is calculated based on the quantified features. When the comprehensive participation score is lower than a preset threshold, an intervention scheduling instruction is generated, and the message set to be scheduled is updated based on the intervention scheduling instruction. Different data tables are created for the same scheduling event, and the same association identifier is established for the data in the different data tables. Through concurrent buffer merging and two-layer asynchronous scheduling decision-making, high real-time and low-cost response of the agent in a high-concurrency environment is achieved. A hierarchical context compression and participation feature quantification intervention mechanism is adopted to overcome the defects of disordered message processing, ambiguous context management, and lack of proactive intervention in the existing technology. Cross-scenario reuse of the underlying scheduling capabilities is realized through standardized interfaces, and an evaluation evidence archive that supports bidirectional penetration is generated. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0016] Figure 1 This is a flowchart of the group interaction message scheduling and context management method provided in the embodiments of the present invention; Figure 2 This is an architecture diagram of the group chat collaboration scheduling engine component provided in an embodiment of the present invention; Figure 3 This is a flowchart of a two-layer asynchronous scheduling decision-making process provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the hierarchical context clipping model provided in an embodiment of the present invention; Figure 5This is a schematic diagram of the functional structure of the group interaction message scheduling and context management device provided in an embodiment of the present invention; Figure 6 This is a functional structure diagram of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0018] Figure 1 A flowchart of the group interaction message scheduling and context management method provided in the embodiments of the present invention is shown below. Figure 1 As shown, the group interaction message scheduling and context management method provided in this embodiment of the invention includes: Step 101: Receive the original group chat messages, classify and label the original group chat messages, store the labeled messages in a buffer window for dynamic sorting, and generate a message set to be scheduled. Step 102: Output the scheduling execution result based on the set of messages to be scheduled, including: matching the message tag of each message in the set of messages to be scheduled with a preset scheduling rule base; if the match is successful, the message is determined to be a deterministic instruction and the scheduling is performed according to the preset scheduling level; if the match fails, the message is determined to be a non-deterministic instruction, the context information associated with the current message is retrieved from the hierarchical context tree, the context information and the non-deterministic instruction are submitted to the large language model for processing, and the scheduling is performed based on the processing result; Step 103: The scheduling execution result is associated and stored in the corresponding hierarchical node in the hierarchical context tree. When the preset pruning conditions are met, reverse pruning is performed on the hierarchical context tree. Step 104: Collect user interaction behavior data during the training process, quantify the interaction behavior data, and calculate the comprehensive participation score based on the quantified features; when the comprehensive participation score is lower than a preset threshold, generate an intervention scheduling instruction, and update the message set to be scheduled based on the intervention scheduling instruction; Step 105: Create different data tables for the same scheduling event, and establish the same association identifier for the data in the different data tables.

[0019] In this embodiment of the invention, the above method can be deployed and executed in an engine, which is an independently deployed software architecture, such as... Figure 2As shown, in the engine, MessageBuffer acts as a message buffer, used to receive messages to be processed and perform temporary storage and preprocessing; Dispatcher acts as a scheduler, retrieving messages to be scheduled from MessageBuffer and then executing scheduling decisions and message distribution; ContextManager acts as a context manager, engaging in bidirectional data interaction with Dispatcher. Dispatcher reads context information from ContextManager and writes updated context to it. At the external interaction level, user input enters the system and is processed by the above components, ultimately outputting an Agent response and synchronously storing the state. These three components together form a complete closed loop for message reception and distribution, and context reading and updating. This engine is compatible with mainstream server operating systems and is decoupled from the front-end interaction, user management, and task management modules of the education and training platform.

[0020] Existing group chat collaboration scheduling systems suffer from numerous technical flaws, such as timing disorder in high-concurrency scenarios; when multiple agents and users interact simultaneously, key instructions generated by LLM are easily blocked by fragmented student speech, leading to the failure of engineering constraints; token explosion and memory degradation, where flattened context management leads to an exponential increase in the number of tokens during long-term training, while simple truncation causes the AI ​​to lose the core engineering logic set in the early stages; the inability to detect "free-riding" in real time, as existing systems lack in-depth monitoring of individual semantic contributions and cannot automatically trigger cognitive intervention actions during training; and strong coupling between evaluation data and training logic, with data storage lacking a structured evidence chain, making it difficult for subsequent evaluation modules to extract key engineering capability indicators from massive amounts of messy logs.

[0021] The group interaction message scheduling and context management method provided in this invention receives original group chat messages, classifies and tags the original group chat messages, stores the tagged messages in a buffer window for dynamic sorting, and generates a message set to be scheduled. It outputs scheduling execution results based on the message set to be scheduled, including: matching the message tag of each message in the message set to be scheduled with a preset scheduling rule base; if the match is successful, the message is determined to be a deterministic instruction and scheduling is performed according to a preset scheduling level; if the match fails, the message is determined to be a non-deterministic instruction, context information associated with the current message is retrieved from a hierarchical context tree, the context information and the non-deterministic instruction are submitted to a large language model for processing, and scheduling is performed based on the processing result; the scheduling execution result is associated and stored in the corresponding hierarchical node in the hierarchical context tree, satisfying the preset... When the pruning conditions are set, reverse pruning is performed on the hierarchical context tree; user interaction behavior data during training is collected, the interaction behavior data is quantified, and a comprehensive participation score is calculated based on the quantified features; when the comprehensive participation score is lower than a preset threshold, an intervention scheduling instruction is generated, and the message set to be scheduled is updated based on the intervention scheduling instruction; different data tables are created for the same scheduling event, and the same association identifier is established for the data in the different data tables. Through concurrent buffer merging and two-layer asynchronous scheduling decision, high real-time and low-cost response of the agent in a high-concurrency environment is achieved; a hierarchical context compression and participation feature quantification intervention mechanism is adopted to overcome the defects of disordered message processing, ambiguous context management, and lack of proactive intervention in the existing technology; cross-scenario reuse of the underlying scheduling capabilities is realized through standardized interfaces, and an evaluation evidence archive that supports bidirectional penetration is generated.

[0022] Based on any of the above embodiments, the step of receiving original group chat messages, classifying and tagging the original group chat messages, storing the tagged messages in a buffer window for dynamic sorting, and generating a message set to be scheduled includes: Step 201: Receive raw group chat messages from the group chat client, parse each raw group chat message and extract message attributes; Step 202: Based on the message attributes, classify and label the original group chat message as at least one of the following types: training instructions, interactive feedback, system notification, or others; Step 203: Store the tagged messages in a buffer window and sort them according to priority weight and timestamp. The priority weight is pre-configured according to the message type, and the priority weight of training instructions is higher than that of interactive feedback. Step 204: Based on the current queue length of the buffer window and the preset queue length threshold, dynamically adjust the queue length parameter of the buffer window according to the preset adjustment coefficient. Step 205: Extract messages from the dynamically sorted buffer window to generate a message set to be scheduled.

[0023] In this embodiment of the invention, the MessageBuffer sliding window mechanism is adopted to solve the technical problem of "agent response storm" caused by concurrent input in multi-person collaboration.

[0024] Based on any of the above embodiments, the scheduling execution result is output according to the set of messages to be scheduled, including: matching the message tag of each message in the set of messages to be scheduled with a preset scheduling rule base; if the match is successful, the message is determined to be a deterministic instruction, and scheduling is performed according to a preset scheduling level; if the match fails, the message is determined to be a non-deterministic instruction, the context information associated with the current message is retrieved from the hierarchical context tree, the context information and the non-deterministic instruction are submitted together to the large language model for processing, and scheduling is performed according to the processing result. This embodiment of the invention distinguishes between code deterministic rules and LLM semantic decisions, significantly reducing the frequency of expensive large model computational power calls while ensuring real-time response. Figure 3 As shown, the two-layer asynchronous scheduling decision-making process is specifically divided into two stages: code-level rule matching and model-level decision-making. Upon arrival of a user message, code-level rule matching is initiated first. If a rule is matched, the code-level decision is executed directly, generating a scheduling instruction and executing the scheduling action, then returning a response. If the rule is not matched, the process proceeds to model-level decision-making. In the model-level decision-making stage, the system calls the large language model for analysis and model judgment: if the intent is clear, a scheduling instruction is generated and the scheduling action is executed; if the intent is ambiguous, a fallback strategy is triggered, using either polling or the default scheduling method to execute the scheduling action, ultimately returning a response. The entire process embodies the decision-making logic of "code-level rule priority, model-level fallback." By bypassing the LLM and directly executing deterministic instructions when a rule is matched, the frequency of large model calls is significantly reduced. Simultaneously, the LLM and fallback strategy handle non-deterministic instructions, ensuring complete coverage of scheduling decisions.

[0025] Based on any of the above embodiments, the step of associating and storing the scheduling execution result with the corresponding hierarchical node in the hierarchical context tree, and performing reverse pruning on the hierarchical context tree when the preset pruning conditions are met, includes: Step 301: Extract training level information from the scheduling execution result. The training level information includes task identifier, stage identifier, group identifier, and user identifier. Step 302: Construct or update the hierarchical context tree according to the four dimensions of task layer, stage layer, group layer and user layer; Step 303: Storing the message content in the scheduling execution result in association with the corresponding hierarchical node in the hierarchical context tree; Step 304: Monitor the validity period of each level node in the hierarchical context tree in real time; Step 305: When any level node exceeds the preset validity period threshold, mark the node as a node to be pruned; Step 306: When the pruning trigger condition is met, perform reverse pruning on the marked nodes to be pruned, prioritize deleting the context information of the user layer and group layer, retain the core context information of the task layer and stage layer, and output the updated hierarchical context tree.

[0026] In this embodiment of the invention, the clipping trigger condition includes at least one of the following: The total amount of context information stored in the hierarchical context tree exceeds the preset token threshold; The validity period of any node at any level exceeds the corresponding validity period threshold; A context compression request was received from the large language model.

[0027] like Figure 4 As shown, the context pruning model arranges pruning priorities from high to low. The first layer is system messages, with the highest priority, and are the first to be pruned; the second layer is historical dialogues, which are subject to secondary pruning; the third layer is tool call results, which require careful pruning; the fourth layer is the agent persona, which should be avoided as much as possible; and the fifth layer is the current message, with the lowest priority, which is retained as the core and not pruned. When the token limit is exceeded, the system performs reverse pruning according to this priority order: starting with the highest priority first-layer system messages, pruning proceeds downwards until the total token amount drops below the threshold, while the fifth-layer current message is always retained to ensure that the core context of the current interaction is not lost. This model, through structured layering and explicit pruning priorities, controls token consumption while preserving the most critical information for agent decision-making to the greatest extent.

[0028] This invention abandons full log input and adopts a five-layer dynamic pruning model to reduce the agent input token by more than 60%, thereby achieving efficient token utilization and improving the "role accuracy" of the response.

[0029] Based on any of the above embodiments, the step of quantifying the interactive behavior data and calculating a comprehensive participation score based on the quantified features includes: Step 401: Map the collected interactive behavior data to a unified quantitative scoring range to obtain the quantitative feature values ​​of each item; Step 402: Based on the preset importance of each interactive behavior data, assign corresponding weight coefficients to each quantitative feature value, and the sum of the weight coefficients is 1; Step 403: Sum the products of each quantitative feature value and its corresponding weight coefficient to calculate the overall participation score.

[0030] Based on any of the above embodiments, the step of generating an intervention scheduling instruction when the overall participation score is lower than a preset threshold, and updating the message set to be scheduled based on the intervention scheduling instruction, includes: Step 501: Compare the calculated overall participation score with the preset participation threshold; Step 502: When the overall participation score is lower than the preset threshold, the intervention level is determined based on the degree of score deviation; Step 503: Generate a corresponding intervention scheduling instruction according to the intervention level. The intervention scheduling instruction includes the target user identifier, intervention type, and intervention content. Step 504: Encapsulate the generated intervention scheduling instructions into a standard message format and assign scheduling priorities; Step 505: Update the set of messages to be scheduled based on the encapsulated intervention scheduling instructions, so that the intervention scheduling instructions can participate in subsequent scheduling execution.

[0031] In this embodiment of the invention, scheduling actions are triggered based on numerical participation scores, transforming "teaching management" into underlying "feedback loop control" to achieve feature-driven proactive intervention.

[0032] Based on any of the above embodiments, the step of creating different data tables for the same scheduling event and establishing the same association identifier for the data in the different data tables includes: Step 601: Create a first data table for the scheduling logs generated during the scheduling execution process. The scheduling logs include at least message identifier, scheduling timestamp, source sender, target receiver, scheduling level, and scheduling status. Step 602: Create a second data table for the hierarchical context tree generated by the hierarchical context management step, the intervention records generated by the participation in the metric and intervention steps, and the training content; Step 603: Generate the same association identifier for the scheduling log, hierarchical context tree, intervention record and training content corresponding to the same scheduling event; Step 604: Write the association identifier into the first data table and the second data table respectively to establish a mapping relationship between the first data table and the second data table; Step 605: Synchronously write the data from the first data table and the second data table to the persistent storage medium.

[0033] The embodiments of the present invention utilize cognitive conflict matrix design and pre-commitment mechanism to ensure the consistency of agent viewpoints and prevent viewpoint degradation in multi-agent collaboration from the bottom layer.

[0034] The group interaction message scheduling and context management method provided in this embodiment of the invention is based on the engine and the steps are executed sequentially combined with cyclic feedback, including: The execution conditions are that the engine deployment is complete, the front-end interface is connected, and the training task initialization configuration is complete; during the execution cycle, message processing is performed in real time, data verification is performed periodically, and participation monitoring is performed cyclically. The specific steps include: (1) Message reception and buffering: The front end collects group chat messages and passes them to the buffer window module to complete tagging, buffering and sorting; (2) Priority rule matching: The set of messages to be scheduled is passed to the scheduling module, and the execution sequence is generated by matching the rule base; (3) Scheduling and distribution: Complete message push according to the execution sequence, generate scheduling result set and pass it to subsequent modules; (4) Hierarchical context construction and update: Construct and update the context tree based on hierarchical information to achieve accurate maintenance; (5) Participation measurement and intervention: Collect user data to calculate participation level, trigger tiered interventions and generate intervention records; (6) Dual write archiving and verification: Write the scheduling log and training content into two tables and perform consistency verification periodically; (7) Extended interface call (optional): Derivative scenarios call the core scheduling capabilities through extended interfaces and record the call log.

[0035] Based on any of the above embodiments, the method provided in the embodiments of the present invention was tested. The test environment was a Linux cloud server (4 cores and 8G), and the training platform was a multi-agent engineering thinking training platform. 30 people were divided into 6 groups of 5 people each to verify the scheduling effect of the engine in different scenarios.

[0036] Example 1: Collaborative Training in Group Engineering Thinking (1) Experimental parameters: Buffer queue 200 / 50, adjustment coefficient 0.5; scheduling level 1 instant / level 2 1s / level 3 3s / level 4 5s; context phase 8h / group 2h / user 0.5h; participation threshold 40 / 20 points; verification cycle 10min; training 2h, complete the engineering design scheme discussion.

[0037] (2) Experimental steps: Initialize task and engine parameters → User sends various messages → Engine executes buffering, scheduling, context update, participation intervention, and double write archive in sequence → Statistical data after training.

[0038] (3) Experimental results: 862 messages were received without loss or error; 5 out of 6 low-participation users reached the target after intervention, with an effective participation rate of 96.7%; 862 data were written twice and all were verified to be consistent; all 6 groups completed the plan on time with an average score of 85 points.

[0039] This invention's embodiments can verify the effectiveness of basic scheduling capabilities and core functions in typical group collaborative training scenarios. By setting medium-sized buffer queues (200 / 50 messages), regular scheduling levels (1s / 3s / 5s), and tiered participation thresholds (40 / 20 points), it is demonstrated that the method can achieve zero message loss and zero errors within a standard 2-hour training cycle, and by automatically intervening in low-participation users, the effective participation rate is increased to 96.7%. This embodiment mainly illustrates the technical advantages of this invention compared to existing technologies in terms of message processing reliability and proactive intervention in participation.

[0040] Example 2: Cross-group task collaborative training (1) Experimental parameters: Buffer queue 300 / 80, adjustment coefficient 0.6; scheduling level 1 instant / level 2 0.5s / level 3 2s / level 4 4s; context phase 12h / group 3h / user 0.8h; participation threshold 35 / 25 points; verification period 15min; training 4h, complete cross-group division of labor and collaboration.

[0041] (2) Experimental steps: Same as Example 1, but with the addition of cross-group message classification and labeling step.

[0042] (3) Experimental results: 1586 messages were received, and the average response time for cross-group instruction scheduling was 0.3s; all 8 low-participation users met the standards, with an effective participation rate of 100%; there were no abnormalities in the dual-write data, and the cross-group scheduling behavior could be traced throughout the entire chain; the task completion rate was 100%, and the average collaboration efficiency was 90 points.

[0043] This invention's embodiments can verify scheduling performance and traceability in scenarios with higher concurrency and more complex collaboration relationships. Compared to Example 1, this embodiment adds a cross-group message classification and tagging step, and adopts a larger buffer capacity (300 / message) and stricter scheduling latency requirements (0.5s / 2s / 4s). Experimental results show that the average response time for cross-group instruction scheduling is only 0.3 seconds, and the cross-group scheduling behavior can achieve full-link traceability. This embodiment is mainly used to illustrate the low-latency scheduling capability of this invention in high-concurrency cross-group collaboration scenarios, and the full-link traceability of complex scheduling behavior through the dual-write archiving mechanism.

[0044] Example 3: Multi-person step-by-step skills training (1) Experimental parameters: Buffer queue 250 / 60, adjustment coefficient 0.4; scheduling level 1 instant / level 2 0.8s / level 3 2.5s / level 4 5s; context phase 10h / group 2.5h / user 0.6h; participation threshold 45 / 30 points; verification cycle 8min; training 3h, complete 5-step skill step training.

[0045] (2) Experimental steps: Same as Example 1, but with the addition of step node context management.

[0046] (3) Experimental results: 1253 messages were received and the step switching instructions were executed immediately; all 5 low-participation users met the standards, with an effective participation rate of 100%; there were no abnormalities in the dual-write data, and the data in each step could be accurately retrieved; the step completion accuracy rate was 100%, and the average process smoothness score was 92.

[0047] This invention verifies the accuracy and smoothness of context management in procedural and step-by-step training scenarios. This embodiment adds step node context management to ensure that step switching commands during training can be executed instantly. Experimental results show that data for each step can be accurately retrieved, and the step completion accuracy reaches 100%. This embodiment mainly illustrates the technical effectiveness of the hierarchical context management module of this invention in ensuring that core logic is not lost and step nodes are not confused during long-process training, i.e., solving the "memory degradation" problem in existing technologies.

[0048] The group interaction message scheduling and context management method provided in this invention achieves zero message loss and no timing disorder in a high-concurrency environment with multiple users through concurrent buffer merging, two-layer asynchronous scheduling decision-making, hierarchical context compression, and quantitative intervention of participation features. This improves scheduling response efficiency by over 80%, with core instruction execution latency ≤1s. Hierarchical context improves the semantic matching degree of collaborative interaction by 75%, completely solving the comprehension bias problem of flat context and effectively preventing memory degradation during multi-agent collaboration. Active intervention improves the trainee's overall participation score by 60%, reduces the proportion of invalid interactions to below 10%, and ensures agent consistency through cognitive conflict matrix design and pre-commitment mechanism. Dual-write archiving and consistency verification achieve full-link traceability of scheduling behavior, with 100% data storage reliability and a zero inconsistency rate. Standardized interfaces improve the development efficiency of subsequent debate engines and scenario orchestration tools by 70%, reducing redundant development. Completely decoupled from front-end business and evaluation modules, it is compatible with mainstream systems and development languages, and can directly connect to existing education and training platforms without large-scale modifications. Priority scheduling ensures the orderly progress of the training process, unblocked core instructions, and a 90% improvement in the smoothness of the collaborative training process.

[0049] By combining deterministic scheduling through code rules with creative decision-making through LLM, this solution addresses issues such as low scheduling accuracy, poor training results, low data traceability, and inability to support derivative collaborative scenarios in educational collaborative training. It provides an independent and decoupled general group chat collaborative scheduling platform, enabling efficient and accurate message scheduling in training scenarios. At the same time, it reserves standardized scheduling interfaces for dynamic debate engines in seminar halls and AI-assisted scenario orchestration tools, enabling cross-scenario reuse of underlying scheduling capabilities.

[0050] Figure 5 This is a schematic diagram of the structure of the group interaction message scheduling and context management device provided in an embodiment of the present invention, as shown below. Figure 5 As shown, the group interaction message scheduling and context management device provided in this embodiment of the invention includes: The receiving module 501 is used to receive the original group chat messages, classify and label the original group chat messages, store the labeled messages in a buffer window for dynamic sorting, and generate a message set to be scheduled. Output module 502 is used to output scheduling execution results according to the set of messages to be scheduled, including: matching the message tag of each message in the set of messages to be scheduled with a preset scheduling rule base; if the match is successful, the message is determined to be a deterministic instruction and the scheduling is performed according to a preset scheduling level; if the match fails, the message is determined to be a non-deterministic instruction, the context information associated with the current message is retrieved from the hierarchical context tree, the context information and the non-deterministic instruction are submitted to the large language model for processing, and the scheduling is performed according to the processing result; The pruning module 503 is used to associate and store the scheduling execution result with the corresponding hierarchical node in the hierarchical context tree, and to perform reverse pruning on the hierarchical context tree when the preset pruning conditions are met. Intervention module 504 is used to collect user interaction behavior data during the training process, quantify the interaction behavior data, calculate a comprehensive participation score based on the quantified features, and generate an intervention scheduling instruction when the comprehensive participation score is lower than a preset threshold, and update the message set to be scheduled based on the intervention scheduling instruction. The creation module 505 is used to create different data tables for the same scheduling event and establish the same association identifier for the data in the different data tables.

[0051] In this embodiment of the invention, the engine further includes the following modules, each of which achieves data interaction through a standardized interface, as specifically designed below: 1. Message buffer window module (1) Data objects: message buffer queue, message tag library, buffer window threshold parameter set (queue length 50-500, adjustment coefficient 0.1-0.9); (2) Processing steps: Receive raw group chat messages → Classify and label (training instructions / interactive feedback / system notifications / others) → Queue status monitoring → Dynamic threshold adjustment → Sort by "priority + timestamp" → Generate a set of messages to be scheduled; (3) Input: Original group chat messages (JSON format), threshold parameter set; (4) Output: Set of messages to be scheduled, queue status log.

[0052] 2. Rule-priority scheduling module (1) Data objects: scheduling rule base, scheduling execution queue, scheduling result set; (2) Processing steps: Receive the set of messages to be scheduled → Parse message tags → Match scheduling rules → Generate scheduling execution sequence (Level 1 executes immediately, Level 2 ≤ 1s, Level 3 ≤ 3s, Level 4 executes in batches) → Schedule execution and distribution → Generate scheduling result set; (3) Input: Set of messages to be scheduled, scheduling rule base; (4) Output: scheduling execution sequence, scheduling result set.

[0053] 3. Hierarchical Context Management Module (1) Data objects: hierarchical context tree (task-phase-group-user), context validity period parameters (phase 1-24h, group 0.5-8h, user 0.1-2h), invalid context removal rules; (2) Processing steps: Receive scheduling result set → parse training level information → update context tree association → monitor validity period → reverse pruning of invalid context → multi-dimensional retrieval as needed; (3) Input: scheduling result set, hierarchical context tree, validity period parameter; (4) Output: Updated hierarchical context tree and context retrieval result set.

[0054] 4. Participation and Anti-Conformity Intervention Module (1) Data objects: Participation feature set (speaking frequency, message validity, interaction response duration, task completion, 0-100 points), participation threshold set (low participation 30-50 points, invalid interaction 0-30 points), intervention strategy library; (2) Processing steps: Collect user data → Feature quantification → Calculate the comprehensive score by weight (weight 0.2 / 0.4 / 0.2 / 0.2) → Threshold comparison → Match the hierarchical intervention strategy → Execute the intervention and monitor the effect → Generate the intervention record; (3) Inputs: user-related message set, task completion data, threshold set, intervention strategy library; (4) Output: Overall participation score, intervention execution instructions, intervention record.

[0055] 5. Dual-write archive module (1) Data objects: scheduling log table, training content table, data consistency verification rules, and verification result set; (2) Processing steps: Receive log / content data from each module → Simultaneous writing to both tables (within 1 second) → Generate association identifier → Periodic verification (5-30 minutes) → Data repair if inconsistent → Generate verification / repair record; (3) Input: Logs / content data of each module, and verification rules; (4) Output: scheduling log table, training content table, verification result set, and repair record.

[0056] 6. Extended Interface Module (1) Data objects: standardized interface protocol (RESTful), interface permission library, interface call log; (2) Processing steps: Encapsulate core scheduling capabilities → Provide basic interfaces for message scheduling / context management / data archiving → Interface permission management → Record call logs → Support dynamic interface expansion; (3) Input: Derivative scenario interface call request, permission configuration; (4) Output: Interface call response results and call log.

[0057] The group interaction message scheduling and context management device provided in this embodiment of the invention receives original group chat messages, classifies and tags the original group chat messages, stores the tagged messages in a buffer window for dynamic sorting, and generates a message set to be scheduled; outputs scheduling execution results based on the message set to be scheduled, including: matching the message tag of each message in the message set to be scheduled with a preset scheduling rule base; if the match is successful, the message is determined to be a deterministic instruction and the scheduling is executed according to a preset scheduling level; if the match fails, the message is determined to be a non-deterministic instruction, the context information associated with the current message is retrieved from the hierarchical context tree, the context information and the non-deterministic instruction are submitted to a large language model for processing, and the scheduling is executed according to the processing result; the scheduling execution result is associated and stored in the corresponding hierarchical node in the hierarchical context tree to meet the preset requirements. When the pruning conditions are set, reverse pruning is performed on the hierarchical context tree; user interaction behavior data during training is collected, the interaction behavior data is quantified, and a comprehensive participation score is calculated based on the quantified features; when the comprehensive participation score is lower than a preset threshold, an intervention scheduling instruction is generated, and the message set to be scheduled is updated based on the intervention scheduling instruction; different data tables are created for the same scheduling event, and the same association identifier is established for the data in the different data tables. Through concurrent buffer merging and two-layer asynchronous scheduling decision, high real-time and low-cost response of the agent in a high-concurrency environment is achieved; a hierarchical context compression and participation feature quantification intervention mechanism is adopted to overcome the defects of disordered message processing, ambiguous context management, and lack of proactive intervention in the existing technology; cross-scenario reuse of the underlying scheduling capabilities is realized through standardized interfaces, and an evaluation evidence archive that supports bidirectional penetration is generated.

[0058] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6As shown, the electronic device may include a processor 610, a communications interface 620, a memory 630, and a communication bus 640. The processor 610, communications interface 620, and memory 630 communicate with each other via the communication bus 640. The memory 630 includes computer programs, an operating system, and acquired data. The processor 610 can call logical instructions stored in the memory 630 to execute a group interaction message scheduling and context management method. This method includes: receiving original group chat messages; classifying and tagging the original group chat messages; storing the tagged messages in a buffer window for dynamic sorting to generate a message set to be scheduled; and outputting scheduling execution results based on the message set to be scheduled, including: matching the message tag of each message in the message set to be scheduled with a preset scheduling rule base; if a match is successful, the message is determined to be a deterministic instruction and scheduled according to a preset scheduling level; if a match fails, the message is determined to be a non-deterministic instruction and retrieved from a hierarchical context tree... The context information associated with the current message is submitted to the large language model for processing along with the non-deterministic instruction, and scheduling is performed based on the processing result. The scheduling execution result is stored in the corresponding hierarchical node in the hierarchical context tree. When a preset pruning condition is met, the hierarchical context tree is reverse-pruned. User interaction behavior data during the training process is collected, and the interaction behavior data is quantized. A comprehensive participation score is calculated based on the quantized features. When the comprehensive participation score is lower than a preset threshold, an intervention scheduling instruction is generated, and the message set to be scheduled is updated based on the intervention scheduling instruction. Different data tables are created for the same scheduling event, and the same association identifier is established for the data in the different data tables.

[0059] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to related technologies, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0060] On the other hand, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the group interaction message scheduling and context management method provided by the above methods. The method includes: receiving original group chat messages, classifying and tagging the original group chat messages, storing the tagged messages in a buffer window for dynamic sorting, and generating a message set to be scheduled; outputting the scheduling execution result according to the message set to be scheduled, including: matching the message tag of each message in the message set to be scheduled with a preset scheduling rule base; if the match is successful, determining that the message is a deterministic instruction and executing the scheduling according to a preset scheduling level; if the match fails, determining that the message is a non-deterministic instruction and executing the scheduling according to the hierarchical context. The system retrieves context information associated with the current message from the tree, submits this context information and the nondeterministic instruction to the large language model for processing, and performs scheduling based on the processing result. The scheduling execution result is stored in the corresponding hierarchical node of the hierarchical context tree. When preset pruning conditions are met, the hierarchical context tree is subjected to reverse pruning. User interaction behavior data during training is collected, and the interaction behavior data is quantized. A comprehensive engagement score is calculated based on the quantized features. When the comprehensive engagement score is lower than a preset threshold, an intervention scheduling instruction is generated, and the message set to be scheduled is updated based on the intervention scheduling instruction. Different data tables are created for the same scheduling event, and the same association identifier is established for the data in the different data tables.

[0061] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0062] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for scheduling and managing group interactive messages, characterized in that, include: Receive raw group chat messages, classify and label the raw group chat messages, store the labeled messages in a buffer window for dynamic sorting, and generate a message set to be scheduled. Outputting scheduling execution results based on the set of messages to be scheduled includes: matching the message tag of each message in the set of messages to be scheduled with a preset scheduling rule base; if the match is successful, the message is determined to be a deterministic instruction and the scheduling is performed according to a preset scheduling level; if the match fails, the message is determined to be a non-deterministic instruction, the context information associated with the current message is retrieved from the hierarchical context tree, the context information and the non-deterministic instruction are submitted to the large language model for processing, and the scheduling is performed based on the processing result. The scheduling execution result is associated and stored in the corresponding hierarchical node in the hierarchical context tree. When the preset pruning conditions are met, reverse pruning is performed on the hierarchical context tree. Collect user interaction behavior data during the training process, quantify the features of the interaction behavior data, and calculate a comprehensive participation score based on the quantified features; when the comprehensive participation score is lower than a preset threshold, generate an intervention scheduling instruction, and update the message set to be scheduled based on the intervention scheduling instruction; Create different data tables for the same scheduling event, and establish the same association identifier for the data in the different data tables; When the overall participation score is lower than a preset threshold, an intervention scheduling instruction is generated, and the set of messages to be scheduled is updated based on the intervention scheduling instruction, including: The calculated overall engagement score is compared with a preset engagement threshold. When the overall participation score is lower than the preset threshold, the intervention level is determined based on the degree of score deviation. According to the intervention level, a corresponding intervention scheduling instruction is generated, and the intervention scheduling instruction includes the target user identifier, intervention type, and intervention content; The generated intervention scheduling instructions are encapsulated into a standard message format and assigned scheduling priorities; The set of messages to be scheduled is updated based on the encapsulated intervention scheduling instructions, so that the intervention scheduling instructions participate in subsequent scheduling execution.

2. The group interaction message scheduling and context management method according to claim 1, characterized in that, The process of receiving raw group chat messages, classifying and tagging the raw group chat messages, storing the tagged messages in a buffer window for dynamic sorting, and generating a message set to be scheduled includes: Receive raw group chat messages from the group chat client, parse each raw group chat message and extract message attributes; Based on the message attributes, the original group chat messages are classified and labeled as at least one of the following types: training instructions, interactive feedback, system notifications, or other types. The tagged messages are stored in a buffer window and sorted according to priority weight and timestamp. The priority weight is pre-configured according to the message type, and the priority weight of training instructions is higher than that of interactive feedback. The queue length parameter of the buffer window is dynamically adjusted according to a preset adjustment coefficient based on the current queue length of the buffer window and a preset queue length threshold. Messages are extracted from the dynamically sorted buffer window to generate a message set to be scheduled.

3. The group interaction message scheduling and context management method according to claim 1, characterized in that, The step of associating and storing the scheduling execution result with the corresponding hierarchical node in the hierarchical context tree, and performing reverse pruning on the hierarchical context tree when the preset pruning conditions are met, includes: The training level information is extracted from the scheduling execution result, and the training level information includes task identifier, stage identifier, group identifier and user identifier; Construct or update a hierarchical context tree based on four dimensions: task layer, stage layer, team layer, and user layer; The message content in the scheduling execution result is associated and stored in the corresponding hierarchical node of the hierarchical context tree; Real-time monitoring of the validity period of nodes at each level in the hierarchical context tree; When any node at any level exceeds the preset validity period threshold, the node is marked as a node to be pruned. When the pruning trigger condition is met, reverse pruning is performed on the marked nodes to be pruned, prioritizing the deletion of context information from the user layer and group layer, while retaining the core context information from the task layer and stage layer, and outputting the updated hierarchical context tree.

4. The group interaction message scheduling and context management method according to claim 3, characterized in that, The clipping trigger condition includes at least one of the following: The total amount of context information stored in the hierarchical context tree exceeds the preset token threshold; The validity period of any node at any level exceeds the corresponding validity period threshold; A context compression request was received from the large language model.

5. The group interaction message scheduling and context management method according to claim 1, characterized in that, The step of quantifying the interactive behavior data and calculating a comprehensive participation score based on the quantified features includes: The collected interactive behavior data are mapped to a unified quantitative scoring range to obtain various quantitative feature values; Based on the preset importance of each interactive behavior data, a corresponding weight coefficient is assigned to each quantitative feature value, and the sum of the weight coefficients is 1; The overall participation score is calculated by summing the products of each quantitative feature value and its corresponding weight coefficient.

6. The group interaction message scheduling and context management method according to claim 1, characterized in that, The step of creating different data tables for the same scheduling event and establishing the same association identifier for the data in the different data tables includes: A first data table is created for the scheduling logs generated during the scheduling execution process. The scheduling logs include at least message identifier, scheduling timestamp, source sender, target receiver, scheduling level, and scheduling status. A second data table is created for the hierarchical context tree generated by the hierarchical context management step, the intervention records generated by the participation measurement and intervention steps, and the training content; Generate the same association identifier for the scheduling log, hierarchical context tree, intervention record, and training content corresponding to the same scheduling event; Write the association identifier into the first data table and the second data table respectively to establish a mapping relationship between the first data table and the second data table; The data in the first data table and the second data table are synchronously written to the persistent storage medium.

7. A group interaction message scheduling and context management apparatus, applicable to the group interaction message scheduling and context management method according to any one of claims 1 to 6, characterized in that, include: The receiving module is used to receive raw group chat messages, classify and label the raw group chat messages, store the labeled messages in a buffer window for dynamic sorting, and generate a message set to be scheduled. The output module is used to output the scheduling execution result based on the set of messages to be scheduled, including: matching the message tag of each message in the set of messages to be scheduled with a preset scheduling rule base; if the match is successful, the message is determined to be a deterministic instruction and the scheduling is performed according to a preset scheduling level; if the match fails, the message is determined to be a non-deterministic instruction, the context information associated with the current message is retrieved from the hierarchical context tree, the context information and the non-deterministic instruction are submitted to the large language model for processing, and the scheduling is performed based on the processing result. The pruning module is used to associate and store the scheduling execution result with the corresponding hierarchical node in the hierarchical context tree, and to perform reverse pruning on the hierarchical context tree when the preset pruning conditions are met. The intervention module is used to collect user interaction behavior data during the training process, quantify the features of the interaction behavior data, and calculate a comprehensive participation score based on the quantified features; when the comprehensive participation score is lower than a preset threshold, an intervention scheduling instruction is generated, and the message set to be scheduled is updated based on the intervention scheduling instruction. A creation module is used to create different data tables for the same scheduling event and establish the same association identifier for the data in the different data tables.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the group interaction message scheduling and context management method as described in any one of claims 1 to 6.

9. A non-transitory readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the group interaction message scheduling and context management method as described in any one of claims 1 to 6.

Images