Task subgroup grouping method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, and provides a task subgroup grouping method and device, electronic equipment and a storage medium, and the method comprises the steps: carrying out the semantic analysis of a task instruction, obtaining a task target, and determining an intention atlas composed of task nodes and a dependency relationship based on a preset task ontology atlas and the task target; constructing an agent capability map by structurally organizing capability information of a plurality of available agents; and calculating the capability matching degree and intention consistency between the task node and the intelligent agent in the intention map, and forming a task sub-group adaptive to the task node based on the capability matching degree and the intention consistency. According to the method, a unified cognitive structure for a task target is provided for all agents by constructing an intention map, and the problem of collaboration obstacle caused by lack of a shared semantic framework is solved; based on the capability matching degree and the intention consistency, the task subgroup used for executing the task nodes in the intention graph is obtained, and the overall collaborative intelligence and task execution robustness of the multi-agent system are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, apparatus, electronic device, and storage medium for task subgrouping. Background Technology

[0002] With the continuous improvement of large language model capabilities, intelligent agents built based on such models are gradually becoming a key component supporting complex systems such as intelligent office, R&D assistants, and knowledge platforms.

[0003] However, several prominent issues remain unresolved in multi-agent platforms. On one hand, agents in current systems generally lack a unified goal-oriented cognitive structure; most models rely solely on simple prompting and invocation strategies, making it difficult for multiple agents to form a shared semantic framework and establish a stable task collaboration mechanism. On the other hand, task allocation often employs static matching, rule orchestration, or manual specification, failing to dynamically construct and optimize based on the semantic features of the task itself, the dependencies between intents, and the execution chain. This results in problems such as collaboration failures, blurred responsibility boundaries, and redundant computation when facing unstructured tasks. Summary of the Invention

[0004] This invention provides a method, apparatus, electronic device, and storage medium for task subgrouping, which addresses the shortcomings of existing technologies where agents in current systems generally lack a unified target cognitive structure. Most models rely solely on simple prompting and invocation strategies for driving, making it difficult for multiple agents to form a shared semantic framework and establish a stable task collaboration mechanism. Furthermore, task allocation often adopts static matching, rule orchestration, or manual specification, failing to dynamically construct and optimize based on the semantic features of the task itself, the dependencies between intentions, and the execution chain. This results in problems such as collaboration failure, blurred responsibility boundaries, and redundant calculations when the system faces unstructured tasks.

[0005] This invention provides a method for grouping task subgroups, comprising the following steps: The system obtains task instructions input by the user, performs semantic parsing on the task instructions to obtain the task objective, and determines an intent graph based on the task objective and a preset task ontology graph. The intent graph includes multiple task nodes and task edges. The task nodes are decomposed from the task objective, and the task edges are used to represent the execution dependencies between the task nodes. Acquire multiple available agents and construct an agent capability map that represents the capabilities of the agents; Matching calculations are performed on the intent graph and the capability graphs of each agent to obtain the capability matching degree and intent level between the task nodes in the intent graph and the agents. Figure 1To the point of being compatible; Based on the aforementioned capability matching degree and the aforementioned intention Figure 1 Consistency is achieved to obtain a task subgroup for executing the task nodes in the intent graph.

[0006] According to a task subgrouping method provided by the present invention, the step of determining an intent graph based on the task objective and a preset task ontology graph includes: Find and anchor the preset target corresponding to the task objective in the task ontology map; Based on the preset actions and preset dependencies associated with the preset target in the task ontology graph, the multiple task nodes and the task edges are generated, and the intent graph is constructed based on the multiple task nodes and the task edges.

[0007] According to a task subgrouping method provided by the present invention, the intention graph and the capability graph of each agent are matched and calculated to obtain the capability matching degree and intention of the task nodes in the intention graph and the agents. Figure 1 Consistency, including: Determine the capability requirements needed to execute the task node; The capability requirements are matched with the capability information in the agent's capability map to determine the capability matching degree; Based on the similarity between the contextual semantic features of task nodes in the intent graph and the semantic representation of agents in the agent capability graph, the intent is determined. Figure 1 To the point of being responsive.

[0008] According to a task subgrouping method provided by the present invention, the task ontology graph includes domain knowledge in the form of a quadruple structure consisting of target, action, action constraint and capability information. In the quadruple structure, the target is the preset target, the action is the preset action, the action constraint is the preset dependency relationship, and the capability information is used to reflect the capability requirements associated with the preset action.

[0009] According to a task subgrouping method provided by the present invention, the method further includes: Obtain an intent subgraph; the intent subgraph is determined based on the target task nodes and target task edges in the intent graph; the target task edge is a task edge that connects the target task node. The intent subgraph is assigned to the corresponding task subgroup so that the task subgroup can perform task collaborative execution. The steps for determining the target task node include: The task nodes in the intent map that meet the preset conditions are identified as the target task nodes; The preset condition is that within the task subgroup, there exists at least one agent with a capability matching degree and intention with the task node. Figure 1 The consistency of all values ​​is greater than their respective preset thresholds.

[0010] According to a task subgrouping method provided by the present invention, the step of assigning the intent subgraph to the corresponding task subgroup for collaborative task execution by the task subgroup includes: Select target task nodes that have not yet been assigned from the intent subgraph as the current nodes to be assigned; Based on the capability information in the agent capability graph, calculate the execution competence of each agent in the task subgroup for the current node to be assigned; Obtain the current available status of each agent within the task subgroup, and select the selected agent from the agents whose execution competence meets the preset claiming conditions and whose current available status is available. The currently unassigned node is assigned to the selected agent, and the selected agent is controlled to execute the assigned task node according to the dependency relationship represented by the target task edge in the intent subgraph.

[0011] According to a task subgrouping method provided by the present invention, the method further includes: During the collaborative execution of the tasks, the communication frequency and task data flow between different task subgroups are continuously monitored; When the communication frequency or the task data flow meets the preset dynamic adjustment conditions, the members of the task subgroup are adjusted.

[0012] The present invention also provides a task subgrouping device, comprising the following units: The acquisition unit is used to acquire the task instructions input by the user, perform semantic parsing on the task instructions to obtain the task target, and determine the intent graph based on the task target and a preset task ontology graph. The intent graph includes multiple task nodes and task edges. The task nodes are obtained by decomposing the task target, and the task edges are used to represent the execution dependencies between the task nodes. A construction unit is used to acquire multiple available agents and construct an agent capability map that represents the capability information of the agents. The matching unit is used to perform matching calculations on the intent graph and the capability graphs of each agent to obtain the capability matching degree and intention-to-control (ITC) score between the task nodes in the intent graph and the agents. Figure 1 To the point of being compatible; Semantic clustering unit, used to perform clustering based on the capability matching degree and the semantic clustering unit. Figure 1 Consistency is achieved to obtain a task subgroup for executing the task nodes in the intent graph.

[0013] 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 executes the program to implement the task subgrouping method as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the task subgrouping method as described above.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the task subgrouping method as described above.

[0016] The present invention provides a task subgrouping method, apparatus, electronic device, and storage medium that performs semantic parsing on task instructions to obtain task objectives. Based on a preset task ontology graph and task objectives, it determines an intent graph composed of task nodes and dependencies. Simultaneously, it constructs an agent capability graph by structurally organizing the capability information of multiple available agents. Furthermore, it calculates the capability matching degree and intent expression between task nodes and agents in the intent graph. Figure 1 This method establishes a unified cognitive structure for all agents regarding task objectives by constructing an intent graph, thus resolving the collaboration barriers caused by the lack of a shared semantic framework. Based on capability matching and intent... Figure 1 Consistency is achieved by obtaining task subgroups for task nodes in the execution intent graph, replacing static or manual task allocation methods. This allows task allocation to be optimized based on the semantic logic and execution chain of the task itself, thereby effectively eliminating the problems of collaborative failure, ambiguous responsibilities and redundant calculations in unstructured task scenarios, and significantly improving the overall collaborative intelligence and task execution robustness of multi-agent systems. Attached Figure Description

[0017] 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.

[0018] Figure 1 This is a flowchart illustrating the task subgrouping method provided by the present invention.

[0019] Figure 2 This is a schematic diagram of the task subgrouping device provided by the present invention.

[0020] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0021] 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.

[0022] Figure 1 This is a flowchart illustrating the task subgrouping method provided by the present invention, as shown below. Figure 1 As shown, the method includes steps 110, 120, 130 and 140.

[0023] Step 110: Obtain the task instruction input by the user, perform semantic parsing on the task instruction to obtain the task objective, and determine the intent graph based on the task objective and the preset task ontology graph; the intent graph includes multiple task nodes and task edges, the task nodes are decomposed from the task objective, and the task edges are used to represent the execution dependencies between the task nodes.

[0024] Specifically, task instructions input by the user can be obtained. These task instructions are those input by the user to initiate a task and reflect information about the target task. Task instructions may include natural language text, voice commands, document uploads, and instructions generated through graphical interface interaction, etc., and this embodiment of the invention does not specifically limit them.

[0025] After obtaining the task instructions, semantic parsing can be performed on the task instructions to obtain the task objectives. Based on the task objectives and the preset task ontology graph, the intent graph is determined. The intent graph includes multiple task nodes and task edges. The task nodes are obtained by decomposing the task objectives, and the task edges are used to represent the execution dependencies between task nodes.

[0026] Here, when the system receives a task instruction from the user, such as "Help me analyze last quarter's sales data and generate a PPT report," it invokes a large language model. This large language model, leveraging its powerful natural language understanding and contextual reasoning capabilities, performs deep semantic analysis of the task instruction, going beyond mere keyword recognition. It identifies the core intent within the instruction, namely the task objective.

[0027] Here, the task ontology graph is a pre-built, structured domain knowledge base. As a unified cognitive framework for the entire system, the task ontology graph is used to uniformly express and understand task-related concepts. Specifically, the task ontology graph can include knowledge about objectives, actions, execution constraints, and required capabilities. For example, for the objective of generating a report, the task ontology graph might define a series of actions required, such as data collection, data analysis, chart generation, and text writing, and specify the order of these actions, i.e., execution constraints. The construction of the task ontology graph can be based on industry standards, expert knowledge, or obtained through knowledge extraction from a large number of domain documents.

[0028] The semantic parsing process aims to transform unstructured user task instructions into a machine-understandable structured intent graph. In embodiments of this invention, this step is to accurately understand the user's true intent by leveraging the unified semantic standard provided by the task ontology graph.

[0029] In an intent graph, task nodes represent specific sub-tasks or actions required to complete the overall task objective. For example, for the task "Analyze Sales Data and Generate a PowerPoint Presentation," task nodes could include reading a sales data file, summarizing sales by month, generating a sales trend line chart, writing an analysis summary, creating a PowerPoint file, and inserting charts and summaries. Task edges represent the logical relationships between task nodes, the most common being dependencies. For example, in the example above, there is a task edge pointing from reading a sales data file to summarizing sales by month, indicating that the latter depends on the completion of the former. Task edges can also represent data flow relationships, such as the output of one node being the input of another node.

[0030] Step 120: Obtain multiple available agents and construct an agent capability map that represents the capability information of the agents.

[0031] Specifically, multiple available agents can be acquired to construct an agent capability graph representing their capabilities. An agent is a software entity possessing a certain degree of autonomy and capability, typically built upon a large language model and encapsulating specific tools or skills. In a multi-agent system, multiple agents with different capabilities can exist and be readily available. For example, there might be an agent skilled in data analysis, an agent skilled in document writing, and an agent skilled in calling APIs (Application Programming Interfaces).

[0032] Here, the agent's capability information can include explicit capability descriptions, historical behavior data, and implicit intent preferences. Explicit capability descriptions refer to capability tags registered by the developer or the agent itself, such as data visualization, natural language generation, and code execution. Historical behavior data refers to the types of tasks the agent has successfully executed in the past, along with efficiency and quality evaluation data. Implicit intent preferences refer to the agent's superior performance in handling specific intent tasks, as discovered through analysis of its historical behavior.

[0033] By structuring the capability information of all available agents, an agent capability graph can be constructed. In the agent capability graph, each agent can be regarded as a node, and its capability information is the attribute of the node. The connection relationship between nodes in the agent capability graph is dynamically constructed based on factors such as the functional complementarity between agent capabilities, the sequential logic of task flow, and semantic similarity.

[0034] Step 130: Perform matching calculations on the intent graph and the capability graphs of each agent to obtain the capability matching degree and intent level between the task nodes in the intent graph and the agents. Figure 1 To the point of being responsive.

[0035] Specifically, after obtaining the intent map and the capability maps of each agent, matching calculations can be performed on the intent map and the capability maps of each agent to obtain the capability matching degree and the intention-to-control (ID) ratio between the task nodes in the intent map and the agents. Figure 1 Consistency. For example, each task node in the intent graph can be matched with each agent in the agent's capability graph. The calculation process generates two core metrics: capability matching degree and intent. Figure 1 To the point of being responsive.

[0036] Here, capability matching degree is a quantitative indicator used to evaluate the degree to which an agent's capability information matches the requirements of a task node. Its core purpose is to measure whether the agent's capabilities meet the execution requirements of a specific task. Capability matching degree is usually expressed numerically, such as a floating-point number between 0 and 1, with higher values ​​indicating better matching. In the specific calculation process, matching algorithms of varying complexity can be used depending on the actual scenario requirements. This can range from simple matching methods based on keyword recognition and overlap to methods such as similarity calculation based on semantic vector representation, to achieve a deeper understanding of the semantic connotations of capabilities and tasks.

[0037] Here, the description of the capability requirements for executing the task node and the agent's capability information text can be encoded into semantic vectors using a pre-trained language model. Then, the cosine similarity between these two semantic vectors is calculated to evaluate their fit at a deep semantic level, thus obtaining the capability matching degree. For example, for the task node of generating a sales trend line chart, an agent with data visualization or chart generation capabilities can obtain a higher capability matching degree score.

[0038] Here, meaning Figure 1 Consistency is a deep-level indicator that assesses the degree to which an agent's behavioral patterns fit the context of the task. Its focus is on whether the agent's intrinsic goals and behavioral style are consistent with the overall semantic environment of the task. Figure 1 Consistent computation requires comprehensive analysis of the contextual semantic information of task nodes in the intent graph, and comprehensive judgment based on the agent's historical behavior data.

[0039] In one optional embodiment, graph neural network techniques are first used to aggregate the semantics of the task node to be assigned and its context nodes in the intent graph, such as predecessor and successor nodes, to generate an enhanced task vector that incorporates overall task context information. Simultaneously, the agent's historical behavioral data, such as the types of successfully completed tasks and the style characteristics of the output content, are analyzed and encoded to form an agent behavior preference vector. Finally, by calculating the cosine similarity between this enhanced task vector and the agent behavior preference vector, the intent graph can be quantified. Figure 1 Consistency indicators.

[0040] For example, when faced with the task of writing formal quarterly financial statement summaries, an agent accustomed to generating serious, structured text demonstrates a high level of competence. Figure 1 The other agent, while possessing text generation capabilities, excels at creating humorous content; its intention... Figure 1 Consistency was significantly lower because their behavioral style did not match the formal context required by the task.

[0041] Step 140, based on the capability matching degree and the intention Figure 1 Consistency is achieved to obtain a task subgroup for executing the task nodes in the intent graph.

[0042] Specifically, it can be based on capability matching degree and willingness Figure 1 Consistency is obtained to obtain a task subgroup for executing task nodes in the intent graph.

[0043] It should be understood that all available agents can be considered as objects to be classified, based on the degree of matching of each agent's capabilities with specific task clusters in the intent graph. Figure 1Consistency is used as the basis for clustering. The system can automatically identify task nodes that are semantically closely related to a certain class and possess both high capability matching degree and high semantic consistency. Figure 1 Agents with similar skills are grouped into the same subgroup. For example, for data processing and analysis tasks in intent graphs, agents skilled in data reading, data cleaning, and statistical calculation will be grouped into one task subgroup; while for report generation and beautification tasks, agents specializing in text writing, PPT creation, and chart beautification will be grouped into another task subgroup.

[0044] Here, in the process of implementing semantic clustering, various technical approaches can be adopted, including classic algorithms such as K-Means, DBSCAN (Density-Based Spatial Clustering of Applications with Noise), or community detection based on graph neural networks. However, the design of its distance or similarity metrics needs to comprehensively consider the ability matching degree and the intention. Figure 1 To the point of being responsive.

[0045] Among them, task subgroups are a type of agent grouping structure generated by semantic clustering. Task subgroups are formed by agents that demonstrate high ability matching and high willingness in specific task scenarios. Figure 1 A cohesive intelligent agent composition, in which the members of the collective intelligent agent form a high degree of synergy in goal cognition and functional skills.

[0046] The method provided in this invention performs semantic parsing on task instructions to obtain task objectives. Based on a preset task ontology graph and task objectives, it determines an intent graph composed of task nodes and dependencies. Simultaneously, it constructs an agent capability graph by structurally organizing the capability information of multiple available agents. Furthermore, it calculates the capability matching degree and intent expression between task nodes and agents in the intent graph. Figure 1 This method establishes a unified cognitive structure for all agents regarding task objectives by constructing an intent graph, thus resolving the collaboration barriers caused by the lack of a shared semantic framework. Based on capability matching and intent... Figure 1 Consistency is achieved by obtaining task subgroups for task nodes in the execution intent graph, replacing static or manual task allocation methods. This allows task allocation to be optimized based on the semantic logic and execution chain of the task itself, thereby effectively eliminating the problems of collaborative failure, ambiguous responsibilities and redundant calculations in unstructured task scenarios, and significantly improving the overall collaborative intelligence and task execution robustness of multi-agent systems.

[0047] Based on the above embodiments, step 110, which involves determining the intent graph based on the task objective and a preset task ontology graph, includes: Step 111: Locate and anchor the preset target corresponding to the task objective in the task ontology map; Step 112: Generate the multiple task nodes and task edges based on the preset actions and preset dependencies associated with the preset target in the task ontology graph, and construct the intent graph based on the multiple task nodes and task edges.

[0048] Specifically, when the task instruction is "help me analyze the sales data of the previous quarter and generate a PPT report", the task objective can be identified as "generate a quarterly sales performance report".

[0049] Next, the system searches for and anchors the pre-defined target corresponding to the task objective within the task ontology graph. The system matches and aligns the task objective identified in the previous step with existing pre-defined targets in the task ontology graph. The task ontology graph may contain a pre-defined target template named "Generate Periodic Business Reports." By calculating the semantic similarity between the text vector of the user's task objective and the text vector of the pre-defined target name in the graph, the system can successfully anchor the user's specific task objective to the pre-defined target. This anchoring process provides a clear starting point and structured guidance for subsequent task decomposition.

[0050] Finally, based on the preset actions and preset dependencies associated with the preset target in the task ontology graph, multiple task nodes and task edges are generated, and an intent graph is constructed based on these multiple task nodes and task edges.

[0051] Once anchoring is complete, the system can perform logical deduction and hierarchical expansion based on the predefined structural paths in the task ontology graph. For example, associated with the preset goal of "generating periodic business reports" are several preset action units, such as acquiring data, processing data, generating visualization charts, and writing report text. The task ontology graph also explicitly defines the preset dependencies between these action units to represent their execution order. Based on the aforementioned preset template and combined with specific parameters parsed from user instructions, such as data source type and report format requirements, the system instantiates the actions and dependencies in the template, thereby generating task nodes and task edges corresponding to this specific task, ultimately constructing a complete intent graph. The execution order of the task nodes output in this process, such as the sequential execution of steps like opening data tables, analyzing key fields, and generating report content as reflected in chained reasoning, is systematically generated in this way.

[0052] The method provided in this invention anchors user task objectives to preset objectives in a task ontology graph, and generates task nodes and task edges based on predefined preset actions and dependencies in the task ontology graph, thereby constructing a structured intent graph. This method automatically transforms abstract and ambiguous user task instructions into intent graphs with clear execution logic and dependencies, effectively solving the problem of unstructured tasks being difficult to directly decompose and schedule, and improving the automation and executability of complex task planning.

[0053] Based on the above embodiments, step 130 includes: Step 131: Determine the capability requirements for executing the task node; Step 132: Match the capability requirements with the capability information in the agent capability map to determine the capability matching degree; Step 133: Based on the similarity between the contextual semantic features of the task nodes in the intent graph and the semantic representation of the agents in the agent capability graph, determine the intent... Figure 1 To the point of being responsive.

[0054] Specifically, first, the capability requirements for executing task nodes are determined. For each task node in the intent graph, the system needs to define its corresponding capability requirements. Here, the information on capability requirements can be obtained directly from the task ontology graph.

[0055] Secondly, the capability requirements are matched with the capability information in the agent's capability graph to determine the capability matching degree. The system will traverse the agent's capability graph and check whether the capability information of each agent includes the capability requirements required by the task node.

[0056] Here, the capability matching degree can be calculated using a Boolean matching method, where a successful match is determined when the agent's capability information includes the capability requirements of the task node; alternatively, a semantic vector-based matching method can be used, where the capability requirement text of the task node and the capability information text of the agent are converted into semantic vectors respectively, and a continuous capability matching degree value is obtained by calculating the cosine similarity between the two. Furthermore, the agent's historical task performance can be introduced as a weighting factor, referencing its success rate or score in performing similar tasks to adjust the matching degree; however, this embodiment of the invention does not specifically limit this approach.

[0057] Here, meaning Figure 1 Consistency is determined based on the similarity calculation between the contextual semantic features of task nodes in the intent graph and the semantic representation of agents in the agent capability graph.

[0058] Here, the contextual semantic features of a task node are composed of its own description and its structured position information in the intent graph, including contextual features such as the output of upstream task nodes and the input of downstream task nodes. For example, a text summary node that undertakes the task of reading academic papers and outputs to the literature review generation task will present academic and rigorous contextual semantics. The semantic representation of the agent is a comprehensive vector, generated by fusing and encoding multi-source information such as the agent's ability information text, historical task execution logs, and preset behavioral preferences, to characterize its overall behavioral profile. Based on the similarity calculation of the above two types of semantic vectors, a quantified intent can be obtained. Figure 1 Consistency score.

[0059] The method provided in this invention calculates the capability matching degree by matching the capability requirements of task nodes with the capability information of intelligent agents, and calculates the meaning by combining the contextual semantics of task nodes with the semantic representation of intelligent agents. Figure 1 This approach achieves a dual match between the agent's functional attributes and task requirements, and between the agent's behavioral patterns and the task context. Through this method, an agent with both the corresponding capabilities and the appropriate task context style can be accurately matched to each task node, effectively avoiding execution deviations caused by insufficient agent capabilities or incompatible behavioral styles, and improving the accuracy and quality of complex task execution.

[0060] Based on the above embodiments, the task ontology graph includes domain knowledge in a four-tuple structure consisting of target, action, action constraint, and capability information; In the quadruple structure, the target is the preset target, the action is the preset action, the action constraint is the preset dependency relationship, and the capability information is used to reflect the capability requirements associated with the preset action.

[0061] Specifically, the task ontology graph includes domain knowledge in the form of a four-tuple structure consisting of objectives, actions, action constraints, and capability information. This four-tuple structure (objective, action, action constraint, capability information) provides a standardized representation paradigm for domain knowledge.

[0062] In this four-tuple structure, the target is a preset target, the action is a preset action, the action constraint is a preset dependency relationship, and the capability information is used to reflect the capability requirements associated with the preset action.

[0063] The method provided in this invention provides a task ontology graph that includes domain knowledge in a four-tuple structure consisting of objectives, actions, action constraints, and capability information. The objective in the four-tuple structure is a preset objective, the action is a preset action, the action constraint is a preset dependency relationship, and the capability information reflects the capability requirements associated with the preset action. This achieves a unified representation of the core elements and their relationships within the domain knowledge, providing a direct, structured basis for task objective identification, action sequence generation, dependency relationship establishment, and capability requirement extraction. This ensures the consistency of knowledge and the standardization of the structure from user task instructions to the executable task graph.

[0064] Based on the above embodiments, the method further includes: Step 210, obtain the intent subgraph; the intent subgraph is determined based on the target task nodes and target task edges in the intent graph; the target task edge is the task edge connecting the target task node; Step 220: Assign the intent subgraph to the corresponding task subgroup so that the task subgroup can perform task collaborative execution. The steps for determining the target task node include: The task nodes in the intent map that meet the preset conditions are identified as the target task nodes; The preset condition is that within the task subgroup, there exists at least one agent with a capability matching degree and intention with the task node. Figure 1 The consistency of all values ​​is greater than their respective preset thresholds.

[0065] Specifically, first, an intent subgraph is obtained, which is determined based on the target task nodes and target task edges in the intent graph. The target task edges are the task edges that connect the target task nodes.

[0066] The steps for determining the target task node include: Each task node in the intent graph is evaluated, and task nodes that meet preset conditions are identified as target task nodes. Here, the preset conditions are that within the task subgroup, there exists at least one agent with a capability matching degree and intent matching degree with the task node. Figure 1 The consistency is greater than their respective preset thresholds. Here, the ability matching degree and the willingness... Figure 1 The preset thresholds for consistency can be the same or different. The preset thresholds can be set to 0.8 or 0.7, etc. The embodiments of the present invention do not make specific limitations on this.

[0067] Here, the intent subgraph is determined by traversing the intent graph. Specifically, for each task node, it is examined whether there are agents in a specific task subgroup that meet the preset capability matching degree and intent. Figure 1A consistency threshold is required, for example, both metrics must be higher than 0.8. If a qualified agent exists, the task node is identified as the target task node for that subgroup. Subsequently, the original task edges connecting all such target task nodes are identified as target task edges. The subgraph consisting of all target task nodes and target task edges is the intent subgraph assigned to that task subgroup.

[0068] Then, the intent subgraph is assigned to the corresponding task subgroups so that the task subgroups can perform task collaborative execution.

[0069] The method provided in this embodiment of the invention obtains an intent subgraph; the intent subgraph is determined based on the target task nodes and target task edges in the intent graph; the target task nodes and target task edges are respectively the capability matching degree and intent degree of the agents within the task subgroup. Figure 1 Task nodes and task edges that all meet a preset threshold in terms of consistency are identified; the intent subgraph is then assigned to the corresponding task subgroups for collaborative task execution. This method combines the satisfaction of capability matching degree with intent... Figure 1 Task nodes and associated edges with consistent dual thresholds are combined into an intent subgraph and assigned to matching task subgroups, thus achieving automated decomposition and precise allocation of complex tasks.

[0070] Based on the above embodiments, step 220 includes: Step 221: Select the target task node that has not yet been assigned from the intent subgraph as the current node to be assigned; Step 222: Based on the capability information in the agent capability graph, calculate the execution competence of each agent in the task subgroup for the current node to be assigned; Step 223: Obtain the current available status of each agent within the task subgroup, and select the selected agent from the agents whose execution competence meets the preset claiming conditions and whose current available status is available; Step 224: Assign the currently unassigned node to the selected agent, and control the selected agent to execute the assigned task node according to the dependency relationship represented by the target task edge in the intent subgraph.

[0071] Specifically, firstly, unassigned target task nodes are selected from the intent subgraph as the current nodes to be assigned. Then, based on the capability information in the agent capability graph, the execution competence of each agent within the task subgroup for the current node to be assigned is calculated. Execution competence reflects the agent's comprehensive ability and applicability in executing a specific target task node within the current task context. Here, key parameter vectors such as historical execution success rate, standardized expected time consumption, and resource consumption are extracted from the agent's capability information. These vectors are then normalized, and the normalized parameters are weighted and fused according to preset weights to generate a comprehensive execution competence. This process transforms abstract agent capabilities into concrete and comparable numerical indicators, providing a quantitative decision-making basis for the accurate allocation of task nodes.

[0072] Furthermore, the current availability status of each agent within the task subgroup can be obtained, and a selected agent can be chosen from those agents whose execution competence meets the preset claiming conditions and whose current availability status is available. The current availability status can be available, busy, under maintenance, etc., and this embodiment of the invention does not specifically limit this.

[0073] Here, the pre-defined claiming conditions can be implemented using a global optimal strategy, a threshold-first strategy, or a multi-agent negotiation strategy. The global optimal strategy directly selects the agent with the highest execution competence. The threshold-first strategy selects the first agent whose execution competence exceeds a certain threshold to improve decision-making speed. The multi-agent negotiation strategy allows multiple agents to negotiate through internal communication to decide who will claim the claim. The selected agent is then called the chosen agent.

[0074] Finally, the currently unassigned node is assigned to the selected agent, and the selected agent is controlled to execute the assigned task node based on the dependency relationships represented by the target task edges in the intent subgraph. The execution competence can be used to reflect information such as the agent's real-time load status and the estimated task execution time.

[0075] The method provided in this invention selects unassigned target task nodes from an intent subgraph as currently unassigned nodes; calculates the execution competence of each agent within a task subgroup for the currently unassigned node based on capability information in the agent capability graph; obtains the current availability status of each agent within the task subgroup, and selects a candidate agent from those whose execution competence meets preset claiming conditions and whose current availability status is available; assigns the currently unassigned node to the candidate agent, and controls the candidate agent to execute the assigned task node according to the dependency relationships represented by the target task edges in the intent subgraph. This method achieves accurate task allocation based on capability matching and dynamic load, thereby enabling task nodes to be assigned to agents with matching capabilities and currently available, thus improving the success rate and reliability of task execution.

[0076] In related technologies, the lack of effective role negotiation mechanisms and dynamic group organization capabilities among intelligent agents makes it impossible to achieve on-demand collaboration, boundary adjustment, and reorganization evolution when facing complex tasks, thus limiting the long-term operational capability of multi-agent systems.

[0077] Based on the above embodiments, the method further includes: Step 310: During the collaborative execution of the tasks, continuously monitor the communication frequency and task data flow between different task subgroups; Step 320: When the communication frequency or the task data flow meets the preset dynamic adjustment conditions, the members of the task subgroup are adjusted.

[0078] Specifically, firstly, during the collaborative execution of tasks, the communication frequency and task data flow between different task subgroups are continuously monitored. Communication frequency refers to the frequency of information exchange between agents within two different subgroups. Task data flow refers to the frequency and volume of data where the output of one subgroup is used as input by another subgroup.

[0079] Then, when the communication frequency or task data flow meets the preset dynamic adjustment conditions, the members of the task subgroup are adjusted. The preset dynamic adjustment conditions are a set of rules used to determine whether the current clustering structure is still optimal.

[0080] The dynamic adjustment conditions may include merging conditions, splitting conditions, and member migration conditions, etc., and the embodiments of the present invention do not specifically limit them.

[0081] In an optional embodiment, if the communication frequency and data interaction volume between task subgroup A and task subgroup B consistently exceed a preset threshold, it indicates that the tasks they undertake are highly correlated, and the current grouping structure may be redundant. In this case, a group merging operation can be triggered to merge subgroups A and B into a unified group.

[0082] In another optional embodiment, if multiple sparsely interacting independent subsets are detected within a task subgroup, and the communication strength between subsets is significantly lower than the communication strength within subsets, it indicates that the group may be undertaking multiple weakly related tasks. In this case, a group splitting operation can be triggered to divide the group into multiple independent subgroups according to the communication mode.

[0083] Furthermore, if the interaction frequency between an agent in subgroup A and members of subgroup B is significantly higher than its interaction frequency with members within its own subgroup A, it indicates that the agent is more collaborative with subgroup B at the functional or task-related level. In this case, a member migration operation can be triggered to move the agent from atomic group A to the target subgroup B.

[0084] Understandably, the hierarchical organizational strategy of "intra-group collaboration + inter-group boundary regulation" breaks the scheduling rigidity of traditional static multi-agent systems, achieving a higher level of flexible orchestration. Within each group, tasks are proactively claimed, roles are negotiated, and context is shared. External support allows for controllable intervention and automatic merging and splitting, thereby achieving a collaborative goal that is both autonomous and manageable.

[0085] Furthermore, the system supports full-process behavior backtracking and structural visualization, allowing users to intuitively view intent graph construction, cluster evolution paths, and task allocation processes, greatly improving the system's interpretability, debuggability, and deployment reliability. Simultaneously, the behavior feedback loop mechanism provides continuous optimization support for the language model and clustering strategy, enabling the system to possess long-term evolution and self-reinforcing capabilities.

[0086] The method provided in this invention continuously monitors the communication frequency and data flow between different task subgroups during collaborative task execution. When the communication frequency or data flow meets preset dynamic adjustment conditions, the members of the task subgroups are adjusted. This ensures that the organizational structure of the agents is not static but can adaptively adjust as the task progresses and its state changes. This solves the problem of decreased collaborative efficiency or unreasonable structure that may occur when static grouping faces long-term, complex, or dynamically changing tasks, significantly improving the robustness and long-term performance of the entire system.

[0087] Based on any of the above embodiments, a task subgrouping method includes the following steps: First, the system receives user input tasks, which can take various forms such as natural language, documents, or structured instructions. Then, a large language model is used for semantic parsing, transforming unstructured user instructions into a preliminary structured task sketch. The system matches the parsed task sketch with a predefined task ontology graph, completing task ontology anchoring. Specifically, through a goal-action-constraint matching process, the user task is mapped to a predefined standardized template in the task ontology graph. Based on this matching result, the system generates an intent graph, which serves as a task structure graph representation, consisting of multiple task nodes and dependency edges representing their logical order, thereby transforming ambiguous user requirements into an executable workflow.

[0088] Then, capability descriptions and historical behavior data are extracted from the agents in the system to construct agent capability profiles. Based on this, graph neural network clustering technology is used to group agents according to semantic similarity, forming task subgroups. This clustering process not only considers the similarity of functional attributes but also introduces semantic similarity. Figure 1 Consistency is used as a deep indicator for group segmentation to ensure that agents within the same group are highly coordinated in terms of behavioral style and goal cognition.

[0089] Subsequently, for each task subgroup, the system allocates corresponding task subgraphs from its intent graph. Within each subgroup, group members are activated and bound to the required capability modules. Agents dynamically accept subtasks through role negotiation and task assignment mechanisms, and maintain execution state synchronization through context sharing. Finally, under the system's scheduling, each subgroup executes its task subgraph in parallel. The entire process is coordinated by the parallel collaborative scheduling module to ensure that dependencies are respected.

[0090] Finally, the system constructs a closed-loop learning and optimization mechanism to achieve continuous evolution. The system collects behavioral logs of the agent throughout the process and calculates performance data such as intent deviation and efficiency metrics. Based on this feedback data, the system can fine-tune and evolve semantic parsing models and clustering algorithms, and optimize clustering and prompting structures, thereby enabling the entire system's task processing capabilities to continuously adapt to environmental changes, becoming increasingly accurate and efficient.

[0091] The method provided in this invention employs a semantic clustering and task capability grouping mechanism guided by graph neural networks to achieve self-organizing combination of agents. This mechanism can not only automatically assign agents according to task intent, but also support dynamic group adjustment, member migration and structural evolution, adapting to complex task changes, capability replacement and boundary ambiguity, etc., significantly improving the system's adaptability in unstructured and variable tasks.

[0092] The task subgrouping device provided by the present invention is described below. The task subgrouping device described below and the task subgrouping method described above can be referred to in correspondence.

[0093] Based on any of the above embodiments, the present invention provides a task subgrouping device. Figure 2 This is a schematic diagram of the structure of the task subgrouping device provided by the present invention, as shown below. Figure 2 As shown, the device includes: The acquisition unit 210 is used to acquire the task instruction input by the user, perform semantic parsing on the task instruction to obtain the task target, and determine the intent graph based on the task target and a preset task ontology graph. The intent graph includes multiple task nodes and task edges. The task nodes are obtained by decomposing the task target, and the task edges are used to represent the execution dependency relationship between the task nodes. Construction unit 220 is used to acquire multiple available agents and construct an agent capability map that represents the capability information of the agents; Matching unit 230 is used to perform matching calculations on the intent graph and the capability graphs of each agent to obtain the capability matching degree and intention-to-control (ITC) score between the task nodes in the intent graph and the agents. Figure 1 To the point of being compatible; Semantic clustering unit 240, used for clustering based on the capability matching degree and the semantic clustering unit 240. Figure 1 Consistency is achieved to obtain a task subgroup for executing the task nodes in the intent graph.

[0094] The apparatus provided in this invention performs semantic parsing on task instructions to obtain task objectives. Based on a preset task ontology graph and task objectives, it determines an intent graph composed of task nodes and dependencies. Simultaneously, it constructs an agent capability graph by structurally organizing the capability information of multiple available agents. Furthermore, it calculates the capability matching degree and intent expression between task nodes and agents in the intent graph. Figure 1 This method establishes a unified cognitive structure for all agents regarding task objectives by constructing an intent graph, thus resolving the collaboration barriers caused by the lack of a shared semantic framework. Based on capability matching and intent... Figure 1 Consistency is achieved by obtaining task subgroups for task nodes in the execution intent graph, replacing static or manual task allocation methods. This allows task allocation to be optimized based on the semantic logic and execution chain of the task itself, thereby effectively eliminating the problems of collaborative failure, ambiguous responsibilities and redundant calculations in unstructured task scenarios, and significantly improving the overall collaborative intelligence and task execution robustness of multi-agent systems.

[0095] Based on any of the above embodiments, the acquisition unit 210 is specifically used for: Find and anchor the preset target corresponding to the task objective in the task ontology map; Based on the preset actions and preset dependencies associated with the preset target in the task ontology graph, the multiple task nodes and the task edges are generated, and the intent graph is constructed based on the multiple task nodes and the task edges.

[0096] Based on any of the above embodiments, the matching unit 230 is specifically used for: Determine the capability requirements needed to execute the task node; The capability requirements are matched with the capability information in the agent's capability map to determine the capability matching degree; Based on the similarity between the contextual semantic features of task nodes in the intent graph and the semantic representation of agents in the agent capability graph, the intent is determined. Figure 1 To the point of being responsive.

[0097] Based on any of the above embodiments, the task ontology graph includes domain knowledge in a four-tuple structure consisting of target, action, action constraint, and capability information; In the quadruple structure, the target is the preset target, the action is the preset action, the action constraint is the preset dependency relationship, and the capability information is used to reflect the capability requirements associated with the preset action.

[0098] Based on any of the above embodiments, a cooperative execution unit is further included, wherein the cooperative execution unit specifically includes: An intent subgraph acquisition unit is used to acquire an intent subgraph; the intent subgraph is determined based on the target task nodes and target task edges in the intent graph; the target task edge is a task edge that connects the target task node. The collaborative execution subunit is used to assign the intent subgraph to the corresponding task subgroup so that the task subgroup can perform collaborative task execution. It also includes a determining unit, which is specifically used for: The task nodes in the intent map that meet the preset conditions are identified as the target task nodes; The preset condition is that within the task subgroup, there exists at least one agent with a capability matching degree and intention with the task node. Figure 1 The consistency of all values ​​is greater than their respective preset thresholds.

[0099] Based on any of the above embodiments, the cooperative execution subunit is specifically used for: Select target task nodes that have not yet been assigned from the intent subgraph as the current nodes to be assigned; Based on the capability information in the agent capability graph, calculate the execution competence of each agent in the task subgroup for the current node to be assigned; Obtain the current available status of each agent within the task subgroup, and select the selected agent from the agents whose execution competence meets the preset claiming conditions and whose current available status is available. The currently unassigned node is assigned to the selected agent, and the selected agent is controlled to execute the assigned task node according to the dependency relationship represented by the target task edge in the intent subgraph.

[0100] Based on any of the above embodiments, an adjustment unit is further included, wherein the adjustment unit is specifically used for: During the collaborative execution of the tasks, the communication frequency and task data flow between different task subgroups are continuously monitored; When the communication frequency or the task data flow meets the preset dynamic adjustment conditions, the members of the task subgroup are adjusted.

[0101] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 3 As shown, the electronic device may include: a processor 310, a communication interface 320, a memory 330, and a communication bus 340, wherein the processor 310, the communication interface 320, and the memory 330 communicate with each other through the communication bus 340. The processor 310 can call logical instructions in the memory 330 to execute a task subgrouping method, which includes: acquiring a task instruction input by a user; performing semantic parsing on the task instruction to obtain a task objective; determining an intent graph based on the task objective and a preset task ontology graph; the intent graph includes multiple task nodes and task edges, the task nodes being decomposed from the task objective, and the task edges representing the execution dependencies between the task nodes; acquiring multiple available agents and constructing an agent capability graph representing the agent's capability information; performing matching calculations on the intent graph and each agent capability graph to obtain the capability matching degree and intention between the task nodes in the intent graph and the agent. Figure 1 Consistency; based on the said capability matching degree and the said intention Figure 1 Consistency is achieved to obtain a task subgroup for executing the task nodes in the intent graph.

[0102] Furthermore, the logical instructions in the aforementioned memory 330 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, essentially, or the part that contributes to the prior art, or a part 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.

[0103] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the task subgrouping method provided by the above methods. The method includes: acquiring a task instruction input by a user; performing semantic parsing on the task instruction to obtain a task objective; determining an intent graph based on the task objective and a preset task ontology graph; the intent graph includes multiple task nodes and task edges, the task nodes being decomposed from the task objective, and the task edges being used to represent the execution dependencies between the task nodes; acquiring multiple available agents and constructing an agent capability graph representing the capability information of the agents; performing matching calculations on the intent graph and each agent capability graph to obtain the capability matching degree and intention degree between the task nodes in the intent graph and the agents. Figure 1 Consistency; based on the said capability matching degree and the said intention Figure 1 Consistency is achieved to obtain a task subgroup for executing the task nodes in the intent graph.

[0104] In another aspect, 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 task subgrouping method provided by the above methods. The method includes: acquiring a task instruction input by a user; performing semantic parsing on the task instruction to obtain a task objective; determining an intent graph based on the task objective and a preset task ontology graph; the intent graph includes multiple task nodes and task edges, the task nodes being decomposed from the task objective, and the task edges being used to represent the execution dependencies between the task nodes; acquiring multiple available agents; constructing an agent capability graph representing the capability information of the agents; and performing matching calculations on the intent graph and each agent capability graph to obtain the capability matching degree and intention degree between the task nodes in the intent graph and the agents. Figure 1 Consistency; based on the said capability matching degree and the said intention Figure 1 Consistency is achieved to obtain a task subgroup for executing the task nodes in the intent graph.

[0105] The device 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.

[0106] 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 part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a 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.

[0107] 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 grouping task subgroups, characterized in that, include: The system obtains task instructions input by the user, performs semantic parsing on the task instructions to obtain the task objective, and determines an intent graph based on the task objective and a preset task ontology graph. The intent graph includes multiple task nodes and task edges. The task nodes are decomposed from the task objective, and the task edges are used to represent the execution dependencies between the task nodes. Acquire multiple available agents and construct an agent capability map that represents the capabilities of the agents; The intent graph and the capability graphs of each agent are matched and calculated to obtain the capability matching degree and intent consistency between the task nodes in the intent graph and the agents. Based on the capability matching degree and the intent consistency, a task subgroup is obtained for executing the task nodes in the intent graph.

2. The task subgrouping method according to claim 1, characterized in that, The process of determining the intent graph based on the task objective and a preset task ontology graph includes: Find and anchor the preset target corresponding to the task objective in the task ontology map; Based on the preset actions and preset dependencies associated with the preset target in the task ontology graph, the multiple task nodes and the task edges are generated, and the intent graph is constructed based on the multiple task nodes and the task edges.

3. The task subgrouping method according to claim 2, characterized in that, The step of matching and calculating the intent graph and the capability graphs of each agent to obtain the capability matching degree and intent consistency between the task nodes in the intent graph and the agents includes: Determine the capability requirements needed to execute the task node; The capability requirements are matched with the capability information in the agent's capability map to determine the capability matching degree; The consistency of intent is determined based on the similarity between the contextual semantic features of task nodes in the intent graph and the semantic representation of agents in the agent capability graph.

4. The task subgrouping method according to claim 3, characterized in that, The task ontology graph includes domain knowledge in a four-tuple structure consisting of target, action, action constraint, and capability information. In the quadruple structure, the target is the preset target, the action is the preset action, the action constraint is the preset dependency relationship, and the capability information is used to reflect the capability requirements associated with the preset action.

5. The task subgrouping method according to any one of claims 1 to 4, characterized in that, The method further includes: Obtain an intent subgraph; the intent subgraph is determined based on the target task nodes and target task edges in the intent graph; the target task edge is a task edge that connects the target task node. The intent subgraph is assigned to the corresponding task subgroup so that the task subgroup can perform task collaborative execution. The steps for determining the target task node include: The task nodes in the intent map that meet the preset conditions are identified as the target task nodes; The preset condition is that within the task subgroup, there exists at least one agent whose capability matching degree and intent consistency with the task node are both greater than their respective preset thresholds.

6. The task subgrouping method according to claim 5, characterized in that, Assigning the intent subgraph to the corresponding task subgroup for collaborative task execution by the task subgroup includes: Select target task nodes that have not yet been assigned from the intent subgraph as the current nodes to be assigned; Based on the capability information in the agent capability graph, calculate the execution competence of each agent in the task subgroup for the current node to be assigned; Obtain the current available status of each agent within the task subgroup, and select the selected agent from the agents whose execution competence meets the preset claiming conditions and whose current available status is available. The currently unassigned node is assigned to the selected agent, and the selected agent is controlled to execute the assigned task node according to the dependency relationship represented by the target task edge in the intent subgraph.

7. The task subgrouping method according to claim 5, characterized in that, The method further includes: During the collaborative execution of the tasks, the communication frequency and task data flow between different task subgroups are continuously monitored; When the communication frequency or the task data flow meets the preset dynamic adjustment conditions, the members of the task subgroup are adjusted.

8. A task subgrouping device, characterized in that, include: The acquisition unit is used to acquire the task instructions input by the user, perform semantic parsing on the task instructions to obtain the task target, and determine the intent graph based on the task target and a preset task ontology graph. The intent graph includes multiple task nodes and task edges. The task nodes are obtained by decomposing the task target, and the task edges are used to represent the execution dependencies between the task nodes. A construction unit is used to acquire multiple available agents and construct an agent capability map that represents the capability information of the agents. A matching unit is used to perform matching calculations on the intent graph and the capability graphs of each agent to obtain the capability matching degree and intent consistency between the task nodes in the intent graph and the agents. A semantic clustering unit is used to obtain a task subgroup for executing the task nodes in the intent graph based on the capability matching degree and the intent consistency.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the task subgrouping method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the task subgrouping method as described in any one of claims 1 to 7.

Images