Meeting activity execution task decomposition and arrangement management method based on AI technology

Abstract

This invention provides an AI-based method for decomposing and orchestrating conference activity execution tasks, relating to the technical field of conference management. The method involves acquiring conference demand information and converting it into digital representation data, employing a bidirectional recursive decomposition strategy to decompose tasks and generate subtask data; calculating a task similarity matrix and a conflict matrix to construct a directed graph of task execution and generate initial orchestration data; and collecting execution status to generate feedback data, calculating a risk assessment score, and adjusting the task execution plan when a threshold is exceeded. This method can improve the execution efficiency of conference activities, reduce the risk of resource conflicts, and achieve dynamic optimization and adjustment.

Description

Technical Field

[0001] The present invention relates to conference management technology, and in particular to an AI-based conference activity execution task decomposition and arrangement management method. Background Art

[0002] Currently, conference and event organization is gradually moving towards digitalization and intelligence, facing increasing complexity in multi-task collaboration, resource matching, and process scheduling. Traditional conference and event management methods, which rely primarily on manual experience and static planning, struggle to cope with the dynamic changes in task granularity, real-time fluctuations in resource status, and the high-frequency interactions between multiple tasks. This results in low overall orchestration efficiency and unstable execution.

[0003] With the development of technologies such as artificial intelligence, natural language processing, and graph computing, the introduction of AI into conference event management processes can provide intelligent assistance throughout the entire process, from understanding conference requirements to task execution. However, existing solutions often focus on optimizing a single step and lack a full-process task management mechanism that integrates task modeling, dynamic decomposition, conflict detection, risk feedback, and intelligent adjustment. This makes it difficult to effectively meet the real-time scheduling and optimization needs of complex conference scenarios.

[0004] Especially in large-scale, cross-departmental conferences, where tasks are highly interdependent and execution dependencies are complex, static rule-driven systems struggle to adapt to the reality of frequently changing tasks. Therefore, an AI-based method for decomposing and orchestrating conference execution tasks is urgently needed to accurately understand multi-source demands, intelligently decompose multiple tasks, and optimize conflicts, thereby improving the intelligent management and execution efficiency of conferences. Summary of the Invention

[0005] The embodiment of the present invention provides a method for decomposing and arranging the execution tasks of conference activities based on AI technology, which can solve the problems in the existing technology.

[0006] According to a first aspect of the embodiments of the present invention,

[0007] Provides an AI-based method for decomposing and arranging conference activity execution tasks, including:

[0008] Obtaining conference activity demand information and converting the conference activity demand information into digital representation data of the conference activity;

[0009] Based on the digital representation data of conference activities, a bidirectional recursive decomposition strategy is used to decompose tasks. The bidirectional recursive decomposition strategy determines the optimal task decomposition granularity through top-down goal decomposition and bottom-up resource reorganization iterative operations, generating multiple subtask data.

[0010] Calculate the similarity between multiple subtask data to obtain a task similarity matrix, calculate the correlation entropy between multiple subtask data to obtain a task conflict matrix, combine the task similarity matrix and the task conflict matrix to construct a task execution directed graph, and generate initial task scheduling data based on the task execution directed graph;

[0011] The execution status of the initial task scheduling data is collected to generate task execution feedback data, and the task risk assessment score is calculated based on the task execution feedback data. When the task risk assessment score exceeds the preset threshold, a task adjustment instruction is generated based on the task status data and resource status data in the task execution feedback data. The task execution directed graph is updated according to the task adjustment instruction to generate the final task scheduling data.

[0012] In an optional embodiment,

[0013] Obtaining conference activity demand information and converting the conference activity demand information into digital representation data of conference activities includes:

[0014] Constructing a knowledge graph for the field of conference activities, wherein the knowledge graph includes multi-dimensional information nodes and the relationships between the nodes, generating a structured demand collection template based on the knowledge graph for the field of conference activities, and obtaining conference activity demand information through the structured demand collection template;

[0015] Obtain the text content in the conference activity demand information, extract the semantic features of the text content, perform sequence annotation on the semantic features, identify the entity and relationship information in the text content, and generate text structured information based on the entity and relationship information;

[0016] Input the conference activity demand information and text structured information into a multi-head attention network, which calculates attention weights based on information reliability, and performs weighted fusion of the conference activity demand information and text structured information according to the attention weights to obtain fusion features;

[0017] The fusion features are matched and mapped with the information nodes in the knowledge graph of the conference activity field to generate digital representation data of the conference activity.

[0018] In an optional embodiment,

[0019] Based on the digital representation data of conference activities, a bidirectional recursive decomposition strategy is used to decompose tasks. The bidirectional recursive decomposition strategy determines the optimal task decomposition granularity through top-down goal decomposition and bottom-up resource reorganization iterative operations, generating multiple subtask data including:

[0020] Based on the digital representation data of conference activities, a task goal tree is constructed. The task goal tree contains multiple levels of nodes, each node represents a task goal, and there is a hierarchical dependency relationship between task goals;

[0021] Based on the hierarchical dependency, the top-down goal decomposition in the bidirectional recursive decomposition strategy is used to decompose the upper-level task goal into multiple lower-level sub-goals step by step, and the complexity score of each sub-goal is calculated to obtain the initial sub-goal set;

[0022] Calculate the target correlation of adjacent sub-targets in the initial sub-target set, and use the bottom-up resource reorganization strategy in the bidirectional recursive decomposition strategy to merge sub-targets. Merge adjacent sub-targets with correlations higher than the preset correlation threshold into new sub-targets to obtain a reorganized sub-target set.

[0023] Calculating a resource consumption index for each sub-goal in the reorganized sub-goal set, and performing weighted calculation on the resource consumption index and the complexity score to obtain a task decomposition evaluation index;

[0024] Iterate the initial sub-goal set and the reorganized sub-goal set, and optimize based on the task decomposition evaluation index. When the task decomposition evaluation index reaches the optimal value, the optimal task decomposition granularity is obtained.

[0025] The task target tree is decomposed based on the optimal task decomposition granularity, and the task target information and resource requirement information of each sub-target node obtained by the decomposition are constructed into corresponding sub-task data to generate multiple sub-task data.

[0026] In an optional embodiment,

[0027] Calculate the similarity between multiple subtask data to obtain a task similarity matrix, calculate the correlation entropy between multiple subtask data to obtain a task conflict matrix, combine the task similarity matrix and the task conflict matrix to construct a task execution directed graph, and generate initial task scheduling data based on the task execution directed graph, including:

[0028] Extract task type, goal, and resource requirement information from multiple subtask data, generate attribute vectors for each subtask, calculate cosine similarity between any two subtask attribute vectors, and construct a task similarity matrix;

[0029] By comparing the execution time intervals of any two subtasks, the time overlap is calculated, the resource contention is calculated through resource allocation conflict detection, and the task dependency strength is calculated through task predecessor relationship analysis. The task correlation entropy is calculated after normalizing the time overlap, resource contention, and task dependency strength to construct a task conflict matrix.

[0030] Adaptively weighting the task similarity matrix and the task conflict matrix, dynamically adjusting the similarity weights and the conflict weights according to the urgency of the tasks to generate a combined weight matrix, constructing a task execution directed graph based on the combined weight matrix, calculating the sum of the out-degree weights of each vertex in the task execution directed graph, and calculating the transfer influence of the vertex based on the combined weights with the adjacent vertices, and combining the sum of the out-degree weights with the transfer influence to obtain a dynamic priority score for the vertex;

[0031] The vertex with the highest dynamic priority score is selected as the starting node, and the task execution directed graph is traversed. After each vertex is traversed, the dynamic priority scores of the remaining vertices are recalculated, and the next traversal vertex is determined according to the updated priority scores to generate initial task scheduling data.

[0032] In an optional embodiment,

[0033] By comparing the execution time intervals of any two subtasks, the time overlap is calculated. By detecting resource allocation conflicts, the resource contention is calculated. By analyzing the task precedence relationship, the task dependency strength is calculated.

[0034] Obtain execution time intervals for multiple subtasks, calculate the time fluctuation range of each subtask based on historical execution data, determine a time elasticity coefficient based on the time fluctuation range, and combine the time elasticity coefficient with the execution time interval to construct a fuzzy time window;

[0035] Probability density modeling is performed on the fuzzy time windows of any two subtasks. A time point sequence is generated using the Monte Carlo sampling method. The task overlap probability at each time point is calculated. The task overlap probability is multiplied by the time interval between the sampling time points and the sum is used to obtain the temporal overlap degree.

[0036] Scan and count the resource application lists of each subtask to obtain resource application frequency data, calculate the resource usage frequency based on the resource application frequency data; calculate the resource scarcity based on the ratio of the total resource amount to the total application amount; multiply the usage frequency by the resource scarcity to obtain the occupancy intensity weight of the resource type;

[0037] Based on the occupancy intensity weights, calculate the product of the resource requirements of any two subtasks on the same resource type, multiply the resource requirements by the occupancy intensity weight to obtain the weighted resource requirement; divide the weighted resource requirement by the total resource requirements of the two subtasks to obtain the resource contention degree;

[0038] A task dependency directed graph is constructed. The direct dependency relationship between task nodes is determined through data flow analysis and set as edge weight. The influence propagation path is calculated recursively based on the edge weight. The edge weight of the direct dependency relationship is combined with the attenuation weight on the influence propagation path to obtain the task dependency strength.

[0039] In an optional embodiment,

[0040] Collecting the execution status of the initial task scheduling data to generate task execution feedback data, and calculating the task risk assessment score based on the task execution feedback data includes:

[0041] Collect the execution status of the initial task scheduling data, dynamically adjust the sampling interval based on the state change frequency of the task execution node, and group the collected execution status according to adaptive time windows; calculate the fluctuation variance of the execution status for each time window, automatically adjust the time decay weight based on the fluctuation variance, smooth the execution status based on the time decay weight, and generate task execution feedback data;

[0042] A multi-layer risk status propagation network is constructed, in which the bottom-layer nodes correspond to risk factors, the middle-layer nodes represent risk cluster categories, and the top-layer nodes represent the comprehensive risk level. The risk propagation coefficient between nodes is calculated based on the task execution feedback data. Based on the risk propagation coefficient, the node status of each layer is updated through bottom-up risk accumulation to obtain an initial risk state. The node status of each layer is then corrected through top-down constraint propagation to generate a risk state vector that includes task progress risk, resource utilization risk, and quality compliance risk.

[0043] The risk state vector and the task execution feedback data are subjected to feature combination, the correlation coefficient matrix between the features is calculated, a combination weight is assigned to each feature based on the correlation coefficient matrix, and the risk state vector and the task execution feedback data are weighted and fused according to the combination weight to generate a task risk assessment score.

[0044] In an optional embodiment,

[0045] Generate task adjustment instructions based on the task status data and resource status data in the task execution feedback data, update the task execution directed graph based on the task adjustment instructions, and generate the final task scheduling data including:

[0046] The task status data and resource status data in the task execution feedback data are combined with the task risk assessment score to form a state input, and a task priority adjustment strategy network and a resource allocation strategy network are constructed based on the state input; the task priority adjustment strategy network generates a task priority adjustment coefficient according to the task status data, and the resource allocation strategy network generates a resource allocation ratio according to the resource status data;

[0047] The state input, task priority adjustment coefficient, resource allocation ratio and its corresponding task risk assessment score are stored in the experience replay pool; the policy network is trained based on the experience replay pool, the network parameters are optimized according to the changes in the task risk assessment score, and the optimized task priority adjustment coefficient and resource allocation ratio are combined into task adjustment instructions;

[0048] In the task execution directed graph, task nodes that need to be adjusted are identified based on the task priority adjustment coefficient in the task adjustment instruction; the degree of influence between nodes is calculated based on the task status data and resource status data; and the set of nodes whose influence exceeds the influence threshold is used to construct a subgraph to be adjusted.

[0049] Establishing a constraint network in the subgraph to be adjusted, updating dependencies between nodes based on task status data and resource status data, adjusting task priorities and resource allocation according to task adjustment instructions, and updating the task execution directed graph;

[0050] Task adjustment instructions and task execution feedback data are combined into training samples for storage; similar samples are retrieved based on current task status data and resource status data, the strategy network is continuously optimized, and the final task scheduling data is generated.

[0051] According to a second aspect of the embodiments of the present invention,

[0052] An electronic device is provided, comprising:

[0053] processor;

[0054] a memory for storing processor-executable instructions;

[0055] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0056] According to a third aspect of the embodiments of the present invention,

[0057] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0058] In this embodiment, a bidirectional recursive decomposition strategy is adopted for task decomposition, which realizes the organic combination of top-down goal decomposition and bottom-up resource reorganization, ensures the scientificity and rationality of task decomposition, and avoids the resource mismatch and low execution efficiency caused by traditional one-way task decomposition. By constructing a task similarity matrix and a task conflict matrix to generate a task execution directed graph, the dependencies and execution order between tasks are systematically presented, providing a clear path guidance for the execution of conference activities, and effectively improving the organizational efficiency and quality of conference activities. Based on the task execution feedback data, real-time assessment and dynamic adjustment of task risks are realized. When the risk assessment score exceeds the preset threshold, the task execution plan can be updated in time, which enhances the risk response capability and resource allocation flexibility during the execution of conference activities, and ensures the smooth progress of conference activities. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1This is a flowchart of a method for decomposing and arranging conference activity execution tasks based on AI technology according to an embodiment of the present invention;

[0060] Figure 2 A diagram showing the relationship between the task combination weight matrix and the dynamic priority score according to an embodiment of the present invention;

[0061] Figure 3 This is a time overlap probability density distribution diagram of an embodiment of the present invention;

[0062] Figure 4 Schematic diagram of the directed graph structure of task execution according to an embodiment of the present invention. DETAILED DESCRIPTION

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0064] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0065] Figure 1 FIG. 1 is a flow chart of a method for decomposing and arranging conference activities based on AI technology according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0066] Obtaining conference activity demand information and converting the conference activity demand information into digital representation data of the conference activity;

[0067] Based on the digital representation data of conference activities, a bidirectional recursive decomposition strategy is used to decompose tasks. The bidirectional recursive decomposition strategy determines the optimal task decomposition granularity through top-down goal decomposition and bottom-up resource reorganization iterative operations, generating multiple subtask data.

[0068] Calculate the similarity between multiple subtask data to obtain a task similarity matrix, calculate the correlation entropy between multiple subtask data to obtain a task conflict matrix, combine the task similarity matrix and the task conflict matrix to construct a task execution directed graph, and generate initial task scheduling data based on the task execution directed graph;

[0069] The execution status of the initial task scheduling data is collected to generate task execution feedback data, and the task risk assessment score is calculated based on the task execution feedback data. When the task risk assessment score exceeds the preset threshold, a task adjustment instruction is generated based on the task status data and resource status data in the task execution feedback data. The task execution directed graph is updated according to the task adjustment instruction to generate the final task scheduling data.

[0070] In an optional embodiment,

[0071] Obtaining conference activity demand information and converting the conference activity demand information into digital representation data of conference activities includes:

[0072] Constructing a knowledge graph for the field of conference activities, wherein the knowledge graph includes multi-dimensional information nodes and the relationships between the nodes, generating a structured demand collection template based on the knowledge graph for the field of conference activities, and obtaining conference activity demand information through the structured demand collection template;

[0073] Obtain the text content in the conference activity demand information, extract the semantic features of the text content, perform sequence annotation on the semantic features, identify the entity and relationship information in the text content, and generate text structured information based on the entity and relationship information;

[0074] Input the conference activity demand information and text structured information into a multi-head attention network, which calculates attention weights based on information reliability, and performs weighted fusion of the conference activity demand information and text structured information according to the attention weights to obtain fusion features;

[0075] The fusion features are matched and mapped with the information nodes in the knowledge graph of the conference activity field to generate digital representation data of the conference activity.

[0076] For example, a knowledge graph in the field of conference activities is first constructed. This knowledge graph contains multi-dimensional information nodes related to conference activities and the association relationships between nodes. Multi-dimensional information nodes include basic information nodes such as conference type nodes, participant nodes, conference venue nodes, conference equipment nodes, and conference time nodes, as well as derived information nodes such as conference process nodes, conference agenda nodes, and conference materials nodes. The association relationships between nodes include inheritance relationships, composition relationships, and dependency relationships. For example, there is an inheritance relationship between the international academic conference node and the conference type node, and a dependency relationship between the conference agenda node and the conference time node. The nodes and relationships in the knowledge graph can be used to describe the various dimensional characteristics of conference activities and their interactions.

[0077] A structured demand collection template is generated based on a knowledge graph for conference events. This template sets collection items according to the node hierarchy of the knowledge graph, and the collection items are organized using relationships within the knowledge graph. Each collection item in the collection template includes attribute information such as the collection item name, data type, value range, and whether it is required. For example, the data type of the meeting type collection item is an enumeration, and the value range includes preset meeting type options such as academic conference, business conference, and training conference. This template allows for standardized collection of meeting event demand information. Specifically, the collection template includes multiple collection areas: basic information collection area, venue requirements collection area, personnel requirements collection area, equipment requirements collection area, and process requirements collection area. The basic information collection area is used to collect basic information such as the meeting name, meeting time, and meeting location; the venue requirements collection area is used to collect venue requirements such as venue capacity and layout; the personnel requirements collection area is used to collect personnel requirements such as the number of attendees and special guests; the equipment requirements collection area is used to collect equipment requirements such as projection equipment and audio equipment; and the process requirements collection area is used to collect process requirements such as the meeting agenda and meeting materials.

[0078] After obtaining the text content in the conference event demand information, it is necessary to extract semantic features and perform sequence annotation on the text content. First, use the word vector model to convert the text content into a word vector sequence, and then use the bidirectional long short-term memory network to extract the contextual semantic features of the text. In specific implementation, the word vector model uses a pre-trained Chinese word vector model, which is trained with massive Chinese corpus and can map Chinese characters to a low-dimensional dense vector space. The bidirectional long short-term memory network contains recurrent neural network layers in both forward and backward directions, which can simultaneously capture the forward and backward contextual information of the text. For example, for the text content "A conference room that can accommodate 200 people is required", it is first converted into a word vector sequence, and then the bidirectional long short-term memory network is used to extract the semantic features of the text containing the venue capacity requirement.

[0079] When performing sequence labeling on the extracted semantic features, a conditional random field model is used to identify entity and relationship information in the text. Specifically, the semantic features are input into the conditional random field model, which calculates the labeling probability based on the feature template and outputs the optimal labeling sequence. Labeling categories include entity labels such as meeting type labels, venue labels, equipment labels, and personnel labels, as well as relationship labels such as venue capacity relations and equipment quantity relations. For example, for the above text, the conditional random field model labels "conference room" as the venue entity and the relationship between "200 people" and "conference room" as the venue capacity relation. Based on the labeling results, unstructured text content can be converted into structured information containing entities and relationships.

[0080] The conference and event demand information and text structured information are fed into a multi-head attention network for information fusion. The multi-head attention network consists of multiple attention sub-networks, each of which independently calculates the attention score between the query vector, key vector, and value vector. The query vector is generated from the conference and event demand information, while the key vector and value vector are generated from the text structured information. The attention score indicates the degree of association between different pieces of information. In specific implementation, the information reliability is first calculated. Information with higher reliability receives a higher attention weight during fusion. Information reliability is comprehensively evaluated based on multiple dimensions, including information source, information completeness, and information consistency. For example, information collected using structured templates is more reliable than information extracted from unstructured text. Each attention sub-network then calculates attention weights based on the information reliability, and a weighted sum is taken of the conference and event demand information and text structured information to produce the fused features.

[0081] Finally, the fused features are matched and mapped with information nodes in the knowledge graph for the conference activity domain. Specifically, the similarity between the fused features and each information node in the knowledge graph is calculated, and the information node with the highest similarity is selected as the mapping target. Similarity calculations are based on the attribute features and relationship features of the nodes. Attribute features include the node's attribute value and attribute type, while relationship features include the node's adjacent nodes and relationship type. For example, if the fused features include the attribute value of "academic conference," they will have a high similarity with the academic conference node in the knowledge graph. Feature mapping can be used to convert the fused features into a standardized representation based on the knowledge graph, generating digital representation data for conference activities. This digital representation data contains the characteristics of each dimension of the conference activity, and the organizational structure of the knowledge graph enables the expression of associations between features.

[0082] In this embodiment, a structured demand collection template is constructed through the knowledge graph of the conference and activity field, and the conference and activity demand information is obtained in a standardized manner, thereby improving the completeness and accuracy of the demand information; a bidirectional long short-term memory network and a conditional random field model are used to extract semantic features and sequence label the text content, accurately identifying entity and relationship information in the text; a multi-head attention network is used to perform feature fusion based on information reliability, effectively integrating conference and activity information from different sources and improving the accuracy of information fusion; and a standardized representation of conference and activity information is achieved through node matching mapping of the knowledge graph, making the digital representation data well structured and interpretable. This method can convert unstructured conference and activity demand information into standardized digital representation data, providing a reliable data foundation for subsequent intelligent analysis and management of conference and activity.

[0083] In an optional embodiment,

[0084] Based on the digital representation data of conference activities, a bidirectional recursive decomposition strategy is used to decompose tasks. The bidirectional recursive decomposition strategy determines the optimal task decomposition granularity through top-down goal decomposition and bottom-up resource reorganization iterative operations, generating multiple subtask data including:

[0085] Based on the digital representation data of conference activities, a task goal tree is constructed. The task goal tree contains multiple levels of nodes, each node represents a task goal, and there is a hierarchical dependency relationship between task goals;

[0086] Based on the hierarchical dependency, the top-down goal decomposition in the bidirectional recursive decomposition strategy is used to decompose the upper-level task goal into multiple lower-level sub-goals step by step, and the complexity score of each sub-goal is calculated to obtain the initial sub-goal set;

[0087] Calculate the target correlation of adjacent sub-targets in the initial sub-target set, and use the bottom-up resource reorganization strategy in the bidirectional recursive decomposition strategy to merge sub-targets. Merge adjacent sub-targets with correlations higher than the preset correlation threshold into new sub-targets to obtain a reorganized sub-target set.

[0088] Calculating a resource consumption index for each sub-goal in the reorganized sub-goal set, and performing weighted calculation on the resource consumption index and the complexity score to obtain a task decomposition evaluation index;

[0089] Iterate the initial sub-goal set and the reorganized sub-goal set, and optimize based on the task decomposition evaluation index. When the task decomposition evaluation index reaches the optimal value, the optimal task decomposition granularity is obtained.

[0090] The task target tree is decomposed based on the optimal task decomposition granularity, and the task target information and resource requirement information of each sub-target node obtained by the decomposition are constructed into corresponding sub-task data to generate multiple sub-task data.

[0091] For example, a task goal tree is constructed based on the digital representation data of conference activities. The task goal tree adopts a tree-like hierarchical structure, which contains task goal nodes at multiple levels. The task goal nodes form a hierarchical dependency relationship through the parent-child relationship. The upper-level nodes represent relatively macro task goals, and the lower-level nodes represent specific execution goals. Each task goal node contains task goal information and resource requirement information. The task goal information describes the specific target content that needs to be completed by the node, and the resource requirement information describes the resource requirements such as personnel, equipment, and venues required to complete the goal. For example, for an academic conference, the root node may be "holding an academic seminar", and its lower-level nodes include task goal nodes such as "conference registration", "conference agenda arrangement", and "on-site service". These nodes can be further subdivided into more specific sub-task goal nodes.

[0092] Using a top-down goal decomposition approach within a bidirectional recursive decomposition strategy, the upper-level task goal node is decomposed into multiple lower-level sub-goal nodes based on the hierarchical dependencies within the task goal tree. The decomposition process first analyzes the components of the upper-level task goal, identifying the key sub-goals required to achieve it. These sub-goals are then mapped to lower-level task goal nodes. A complexity score is calculated for each sub-goal node, based on a comprehensive assessment of multiple dimensions, including the time cost, labor cost, and technical difficulty required to achieve the goal. In implementation, weight coefficients are assigned to each dimension, and the weighted sum of the scores is used to obtain the final complexity score. For example, "arranging the meeting agenda" can be decomposed into sub-goals such as "setting the meeting schedule," "arranging the speaking order," and "coordinating the meeting time." "Coordinating the meeting time" has a higher complexity score because it involves communication and coordination among multiple parties. This recursive decomposition yields an initial set of sub-goals.

[0093] Calculate the goal correlation for adjacent subgoals in the initial subgoal set. The goal correlation indicates the degree of association between two subgoals. A higher correlation indicates a greater suitability for merging the two subgoals. The goal correlation is calculated based on the resource similarity, execution dependency, and timing overlap of the subgoals. Resource similarity indicates the degree of overlap in the resources required by two subgoals, execution dependency indicates the strength of the execution dependency between the two subgoals, and timing overlap indicates the degree of overlap in the execution times of the two subgoals. For example, the subgoals "arrange the order of speeches" and "coordinate meeting times" have high resource similarity and execution dependency, resulting in a high goal correlation.

[0094] Using the bottom-up resource reorganization strategy within the bidirectional recursive decomposition strategy, adjacent subgoals with a correlation above a preset correlation threshold are merged. The merging process first selects a pair of adjacent subgoals with the highest correlation and merges them into a new subgoal. The correlation between the new subgoal and its adjacent subgoals is then recalculated. This merging process repeats until no more subgoal pairs meet the merge criteria. The merged new subgoal inherits the resource requirements of the original goal and integrates any overlapping resources. For example, "arranging the order of speeches" and "coordinating venue times" can be merged into a new subgoal, "arranging and coordinating speech times." This resource reorganization results in a set of reorganized subgoals.

[0095] A resource consumption index is calculated for each subgoal in the reorganized subgoal set. This index reflects the total amount of resources required to complete the subgoal, encompassing multiple dimensions such as human resources, equipment resources, and site resources. Resource consumption for each dimension is quantified based on the resource requirements of the subgoal. The resource consumption index is weighted with the previously calculated complexity score to produce the task decomposition evaluation index. This index comprehensively reflects the rationality of the task decomposition plan; a smaller index indicates a more optimal decomposition plan.

[0096] The initial set of subgoals and the reorganized set of subgoals are iteratively optimized. In each iteration, top-down goal decomposition and bottom-up resource reorganization are performed alternately, and the corresponding task decomposition evaluation indicators are calculated. When the relative rate of change of the task decomposition evaluation indicators is less than a preset threshold, the optimal state is reached, and the subgoal division scheme for the current iteration is determined to be the optimal task decomposition granularity. Based on the optimal task decomposition granularity, the task goal tree is decomposed, and task goal information and resource requirement information are extracted from the decomposed subgoal nodes. The extracted information is organized according to a predefined data structure and constructed into subtask data containing goal descriptions, resource allocation, and other content. For example, for the subgoal "speech time arrangement and coordination," the generated subtask data includes specific time scheduling requirements, required coordination personnel, and other information. Finally, multiple structured subtask data are output for subsequent task allocation and execution management.

[0097] In this embodiment, the hierarchical dependency relationship between task objectives is clearly described through the hierarchical representation of the task objective tree; a bidirectional recursive decomposition strategy combining top-down objective decomposition with bottom-up resource reorganization is adopted to ensure the integrity of task decomposition and optimize resource utilization; task decomposition evaluation indicators based on objective complexity scores and resource consumption indicators can effectively evaluate the rationality of task decomposition schemes; the optimal task decomposition granularity is determined through iterative optimization to achieve a balance between the degree of task objective subdivision and resource allocation efficiency; the subtask data finally generated contains both clear objective descriptions and detailed resource allocation information, providing a reliable basis for subsequent task execution. This method can adaptively decompose complex conference activity tasks into a reasonable set of subtasks, improving task execution efficiency and resource utilization.

[0098] In an optional embodiment,

[0099] Calculate the similarity between multiple subtask data to obtain a task similarity matrix, calculate the correlation entropy between multiple subtask data to obtain a task conflict matrix, combine the task similarity matrix and the task conflict matrix to construct a task execution directed graph, and generate initial task scheduling data based on the task execution directed graph, including:

[0100] Extract task type, goal, and resource requirement information from multiple subtask data, generate attribute vectors for each subtask, calculate cosine similarity between any two subtask attribute vectors, and construct a task similarity matrix;

[0101] By comparing the execution time intervals of any two subtasks, the time overlap is calculated, the resource contention is calculated through resource allocation conflict detection, and the task dependency strength is calculated through task predecessor relationship analysis. The task correlation entropy is calculated after normalizing the time overlap, resource contention, and task dependency strength to construct a task conflict matrix.

[0102] Adaptively weighting the task similarity matrix and the task conflict matrix, dynamically adjusting the similarity weights and the conflict weights according to the urgency of the tasks to generate a combined weight matrix, constructing a task execution directed graph based on the combined weight matrix, calculating the sum of the out-degree weights of each vertex in the task execution directed graph, and calculating the transfer influence of the vertex based on the combined weights with the adjacent vertices, and combining the sum of the out-degree weights with the transfer influence to obtain a dynamic priority score for the vertex;

[0103] The vertex with the highest dynamic priority score is selected as the starting node, and the task execution directed graph is traversed. After each vertex is traversed, the dynamic priority scores of the remaining vertices are recalculated, and the next traversal vertex is determined according to the updated priority scores to generate initial task scheduling data.

[0104] For example, task type, task objectives, and resource requirement information are extracted from multiple subtask data to generate an attribute vector for each subtask. Task type information includes the nature of the task, such as meeting preparation, meeting site, or meeting service. Task objective information includes a detailed description of the task's objectives and completion criteria. Resource requirement information includes the required personnel, equipment, and venue types. This information is converted into a unified attribute vector, where each dimension in the attribute vector corresponds to a task characteristic. For example, for the "Conference Registration Service" task, its attribute vector includes information on dimensions such as conference service type, registration personnel requirements, and registration equipment requirements.

[0105] Calculate the cosine similarity between the attribute vectors of any two subtasks to determine the degree of similarity between the two tasks. Cosine similarity measures the similarity of task features by calculating the cosine of the angle between two attribute vectors. A higher similarity value indicates greater similarity between the two tasks. The similarity values ​​between all pairs of tasks are organized into a task similarity matrix, which reflects the similarity relationships between tasks in the task set. For example, the "Conference Registration Service" task has a high similarity to the "Conference Participant Information Collection" task, but a lower similarity to the "Venue Setup" task.

[0106] The temporal overlap is calculated by comparing the execution time intervals of any two subtasks. Temporal overlap indicates the degree of temporal overlap between two tasks and is measured by calculating the ratio of the intersection of the two task execution time intervals to the union of the two task execution time intervals. The greater the temporal overlap, the more likely the two tasks will conflict during execution. For example, the execution times of the "Conference Registration Service" and "Venue Setup" tasks partially overlap, requiring a proper execution order.

[0107] Resource contention is calculated through resource allocation conflict detection. Resource contention indicates the degree of conflict between two tasks regarding resource usage. It is calculated by detecting whether the resource requirements of the two tasks overlap. For each resource type, if both tasks require that type of resource and the total amount of resources is limited, resource contention exists. A higher resource contention indicates a higher likelihood of resource conflict between the two tasks. For example, two tasks requiring the use of the same conference room may have a high degree of resource contention.

[0108] Task dependency strength is calculated by analyzing task precedence relationships. Task dependency strength represents the execution dependency between two tasks and is calculated by analyzing the input-output relationship between the tasks. If the output of one task is a required input for another, then the two tasks have a dependency relationship. The greater the dependency strength, the less likely the execution order of the two tasks can be adjusted. For example, "Collecting meeting information" must be completed before "Preparing meeting materials," indicating a strong dependency between the two tasks.

[0109] The temporal overlap, resource contention, and task dependency strength are normalized to a range between zero and one. Task correlation entropy is then calculated. This comprehensively reflects the degree of conflict between two tasks during execution; a higher correlation entropy indicates more severe conflict. The correlation entropy values ​​for all task pairs are organized into a task conflict matrix.

[0110] An adaptive weighted combination of the task similarity matrix and the task conflict matrix is ​​performed. Similarity and conflict weights are dynamically adjusted based on the urgency of the task. Urgent tasks prioritize conflict factors, while non-urgent tasks prioritize similarity factors. A combined weight matrix is ​​generated through weighted combination, and a directed task execution graph is constructed based on this matrix. Each vertex in the directed graph represents a task, and directed edges between vertices represent the execution order of tasks. The edge weights are determined by the combined weights.

[0111] The sum of the out-degree weights of each vertex in the task execution directed graph is calculated. This sum represents the strength of the task's direct connection with other tasks. The transitive influence of a vertex is also calculated. This transitive influence assesses the scope of a task's indirect influence by considering the combined weights of its neighboring vertices. Combining the sum of the out-degree weights with the transitive influence yields a dynamic priority score for the vertex, which comprehensively reflects the task's importance within the entire task network.

[0112] The vertex with the highest dynamic priority score is selected as the starting node for the traversal. This is typically the most influential or urgent task. During the traversal, after each vertex is visited, the dynamic priority scores of the remaining vertices are recalculated. This priority score update takes into account the impact of the visited vertex on the remaining tasks, ensuring dynamic adaptability of task scheduling. The next vertex to traverse is determined based on the updated priority score. This process is repeated until all vertices have been traversed, ultimately generating initial task scheduling data. This scheduling data reflects the execution order and priority of the tasks.

[0113] Figure 2 This is a relationship diagram between the task combination weight matrix and the dynamic priority score according to an embodiment of the present invention, as shown in FIG. Figure 2 The figure shows a directed task execution graph consisting of 8 task nodes and 12 connecting edges, along with a table comparing the corresponding task priority scores. Each node in the figure represents a subtask, and the thickness of the connecting lines between nodes intuitively indicates the size of the combined weights. The combined weight between Task 1 and Task 2 is the highest, reaching 0.82, while the combined weight between Task 1 and Task 5 is the lowest, at only 0.23. These combined weights are derived through an adaptive weighted combination of the task similarity matrix and the task conflict matrix, fully accounting for inter-task similarities and potential conflicts. In terms of dynamic priority scores, Task 2 has the highest score of 0.95, due to its large sum of outdegree weights (the sum of the combined weights with Tasks 3 and 6) and its strong transfer influence. Task 8 has the lowest score of 0.52, primarily due to its position at the end of the task chain and its limited transfer influence. Comparing the dynamic priority of this technical solution with the traditional static priority sorting method (such as the deadline-based EDD algorithm, full name Earliest Due Date), it can be found that the priority of this technical solution is generally higher than the static method, and it can better capture the impact of the dependency relationship between tasks on the overall execution efficiency. For example, for the key node task 2, the priority given by this solution is 0.95, which is significantly higher than the static method of about 0.80, which shows that this solution can identify and strengthen the importance of key tasks. It can also be seen in the figure that the task execution directed graph presents a clear network structure, reflecting the complex interdependence between tasks in the actual work environment. Compared with the traditional linear or tree-like task structure, the directed graph constructed by this technical solution can more comprehensively express the multi-dimensional relationship between tasks.

[0114] This application constructs a task similarity matrix and a task conflict matrix and performs adaptive weighted combination to establish a task execution directed graph and generate initial task scheduling data based on a dynamic priority mechanism, which can achieve accurate modeling and dynamic sorting of the collaborative and conflict relationships between multiple subtasks. Existing technologies often use static priorities or fixed rule allocation in task scheduling, which makes it difficult to simultaneously take into account the relevance and conflict between tasks, easily leading to resource conflicts, execution blockages, or low scheduling efficiency. The starting point of the improvement of this application is to consider the dynamic balance between the collaborative feasibility reflected by task similarity and the execution risk represented by task conflict. It comprehensively constructs attribute vectors based on task type, goal, and resource requirement information to quantify the similarity between tasks. At the same time, it introduces time overlap, resource competition, and predecessor relationships to evaluate the degree of task conflict. It integrates different factors through adaptive weighting and dynamically adjusts the combination weights based on the urgency of the task, making task scheduling more flexible and real-time. Compared with traditional methods, this solution can not only effectively avoid the allocation of conflicting resources, but also improve the overall task execution efficiency and rationality, with higher scheduling flexibility and execution stability.

[0115] In an optional embodiment, calculating the time overlap by comparing the execution time intervals of any two subtasks, calculating the resource contention by detecting resource allocation conflicts, and calculating the task dependency strength by analyzing the task predecessor relationships include:

[0116] Obtain execution time intervals for multiple subtasks, calculate the time fluctuation range of each subtask based on historical execution data, determine a time elasticity coefficient based on the time fluctuation range, and combine the time elasticity coefficient with the execution time interval to construct a fuzzy time window;

[0117] Probability density modeling is performed on the fuzzy time windows of any two subtasks. A time point sequence is generated using the Monte Carlo sampling method. The task overlap probability at each time point is calculated. The task overlap probability is multiplied by the time interval between the sampling time points and the sum is used to obtain the temporal overlap degree.

[0118] Scan and count the resource application lists of each subtask to obtain resource application frequency data, calculate the resource usage frequency based on the resource application frequency data; calculate the resource scarcity based on the ratio of the total resource amount to the total application amount; multiply the usage frequency by the resource scarcity to obtain the occupancy intensity weight of the resource type;

[0119] Based on the occupancy intensity weights, calculate the product of the resource requirements of any two subtasks on the same resource type, multiply the resource requirements by the occupancy intensity weight to obtain the weighted resource requirement; divide the weighted resource requirement by the total resource requirements of the two subtasks to obtain the resource contention degree;

[0120] A task dependency directed graph is constructed. The direct dependency relationship between task nodes is determined through data flow analysis and set as edge weight. The influence propagation path is calculated recursively based on the edge weight. The edge weight of the direct dependency relationship is combined with the attenuation weight on the influence propagation path to obtain the task dependency strength.

[0121] This embodiment provides a task coordination analysis method based on subtask time intervals, resource occupancy and dependency relationships, which achieves efficient scheduling and optimized execution of tasks by calculating the time overlap, resource competition and task dependency intensity between tasks.

[0122] First, we obtain the execution time interval data for multiple subtasks. For example, we extract the execution time interval data from the historical execution records for four subtasks, A, B, C, and D. The execution time interval for subtask A is [10, 20], indicating that it starts at 10:00 and ends at 20:00. The execution time interval for subtask B is [15, 25]; the execution time interval for subtask C is [5, 15]; and the execution time interval for subtask D is [22, 30].

[0123] The time fluctuation range of each subtask is calculated based on historical execution data. For subtask A, by collecting its historical execution records, it is found that its actual execution time fluctuates within ±10% of the nominal value, resulting in a time fluctuation range of ±2. The time fluctuation range of subtask B is ±3; the time fluctuation range of subtask C is ±1.5; and the time fluctuation range of subtask D is ±2.4.

[0124] Determine the time elasticity coefficient based on the time fluctuation range. The time elasticity coefficient is calculated by dividing the time fluctuation range by the nominal execution time. For subtask A, the time elasticity coefficient is 2 / 10 = 0.2; for subtask B, it is 0.3; for subtask C, it is 0.15; and for subtask D, it is 0.3.

[0125] The time elasticity coefficient is combined with the execution time interval to construct a fuzzy time window. The fuzzy time window for subtask A is [8, 22], obtained by shifting the original start time forward by 2 units and the end time backward by 2 units. The fuzzy time window for subtask B is [12, 28]; the fuzzy time window for subtask C is [3.5, 16.5]; and the fuzzy time window for subtask D is [19.6, 32.4].

[0126] Probability density modeling is performed for any two subtasks' fuzzy time windows. A fuzzy time window represents the distribution of the actual execution probability of a task within a certain interval. The boundaries are not completely determined, but rather have a certain degree of fuzziness. Probability density modeling is used to quantify the distribution of the probability of task execution at different time points. Taking subtasks A and B as an example, a probability density value is assigned to each time point within the fuzzy time window, with the central time point having the highest probability and decreasing toward the ends. Using a triangular probability density function, the probability density of subtask A at time point 10 is 0.1, the probability density at time point 15 is 0.3, and the probability density at time point 20 is 0.1. The probability density of subtask B at time point 15 is 0.08, the probability density at time point 20 is 0.25, and the probability density at time point 25 is 0.08.

[0127] A time sequence is generated using the Monte Carlo sampling method. Based on the aforementioned probability density function, 1000 sampling points are randomly generated within the fuzzy time window, including the start and end time points. For subtask A, the generated partial time sampling points are [8.2, 9.5, 10.8, 12.4, 14.7, 15.9, 17.3, 18.6, 20.2, 21.8]; for subtask B, the generated partial time sampling points are [12.3, 13.8, 15.2, 16.7, 18.4, 20.1, 21.9, 23.5, 25.3, 27.8].

[0128] Calculate the probability of task overlap at each time point. For any time point t, if that time point falls within the fuzzy time windows of both subtasks, calculate the probability of simultaneous execution of the two subtasks at that time point. This is the product of the probability density values ​​of the two tasks at that time point. At time point 16, the probability density of subtask A is 0.25, and the probability density of subtask B is 0.15, resulting in an overlap probability of 0.0375.

[0129] The temporal overlap is calculated by multiplying the task overlap probability by the time interval between the sampling points and summing the results. Temporal overlap measures the risk of conflict between two tasks. A larger value indicates a higher probability of conflict. If the time interval between the sampling points is set to 0.1, the temporal overlap between subtasks A and B is 0.0375 × 0.1 + 0.0412 × 0.1 + ... + 0.0298 × 0.1 = 0.421, indicating that the two subtasks overlap approximately 42.1% of the time.

[0130] Scan and count the resource request lists for each subtask to obtain resource request frequency data. Assume that the system contains three types of resources: CPU, memory, and disk. Subtask A requests 5 units of CPU resources, 10 units of memory resources, and 20 units of disk resources; subtask B requests 8 units of CPU resources, 5 units of memory resources, and 15 units of disk resources. By counting the resource request lists of all subtasks, we find that the total number of CPU resource requests is 25, the total number of memory resource requests is 35, and the total number of disk resource requests is 80.

[0131] Calculate resource usage frequency based on resource request frequency data. The CPU resource usage frequency is 25 / 140 = 0.179, the memory resource usage frequency is 35 / 140 = 0.25, and the disk resource usage frequency is 80 / 140 = 0.571.

[0132] Resource scarcity is calculated based on the ratio of total resource usage to total requested resources. For example, if the total CPU resource usage is 20 units and the total requested resources are 25 units, the resource scarcity is 25 / 20 = 1.25. For example, if the total memory resource usage is 40 units and the total requested resources are 35 units, the resource scarcity is 35 / 40 = 0.875. For example, if the total disk resource usage is 100 units and the total requested resources are 80 units, the resource scarcity is 80 / 100 = 0.8.

[0133] Multiply the usage frequency by the resource scarcity to get the resource type's occupancy intensity weight. The CPU resource occupancy intensity weight is 0.179 × 1.25 = 0.224; the memory resource occupancy intensity weight is 0.25 × 0.875 = 0.219; and the disk resource occupancy intensity weight is 0.571 × 0.8 = 0.457.

[0134] Based on the occupancy intensity weights, calculate the product of the resource requirements of any two subtasks for the same resource type and multiply this product by the occupancy intensity weight to obtain the weighted resource requirement. The product of the CPU requirements of subtasks A and B is 5 × 8 = 40, resulting in a weighted resource requirement of 40 × 0.224 = 8.96. The product of the memory requirements is 10 × 5 = 50, resulting in a weighted resource requirement of 50 × 0.219 = 10.95. The product of the disk requirements is 20 × 15 = 300, resulting in a weighted resource requirement of 300 × 0.457 = 137.1.

[0135] Divide the weighted resource requirement by the total resource requirement of the two subtasks to get the resource contention. The total resource requirement of subtasks A and B is (5 + 10 + 20) + (8 + 5 + 15) = 63, and the resource contention between them is (8.96 + 10.95 + 137.1) / 63 = 2.49.

[0136] Construct a task dependency directed graph. Use data flow analysis to determine the direct dependencies between task nodes and set them as edge weights. For example, if 60% of subtask A's output data is used by subtask B, set the edge weight from A to B to 0.6; if 30% of subtask B's output data is used by subtask D, set the edge weight from B to D to 0.3; if 40% of subtask C's output data is used by subtask A, set the edge weight from C to A to 0.4; and if 25% of subtask C's output data is used by subtask B, set the edge weight from C to B to 0.25.

[0137] The influence propagation path is recursively calculated based on edge weights. For the influence propagation path from C to D, there are two paths: C→A→B→D and C→B→D. The weight of the C→A→B→D path is 0.4×0.6×0.3=0.072; the weight of the C→B→D path is 0.25×0.3=0.075.

[0138] The task dependency strength is calculated by combining the edge weight of the direct dependency with the decay weight along the influence propagation path. A decay factor of 0.8 is used, meaning that the influence of the indirect dependency decays as the path length increases. The task dependency strength from subtask C to D is 0.072 × 0.8² + 0.075 × 0.8 = 0.106, indicating that subtask C has a 10.6% dependency influence on subtask D.

[0139] Figure 3 This is a time overlap probability density distribution diagram of an embodiment of the present invention, such as Figure 3The figure shows the probability distribution of the time overlap between tasks T2 and T4. This technical solution utilizes probability density modeling of fuzzy time windows and Monte Carlo sampling to obtain a more realistic probability distribution curve for time overlap. The figure clearly shows that the probability density distribution calculated by this technical solution exhibits an asymmetric bimodal structure, with a probability density of 0.224 at 10 hours and a peak of 0.241 at 15 hours, reflecting the complex temporal relationship between the two tasks during execution. In contrast, linear probability distribution methods (such as the linear decay model) exhibit a flatter unimodal distribution, with a peak of only 0.195 at 16 hours. Uniform probability distribution methods (such as the standard PERT method) simply assume a constant probability density of 0.143 between 0 and 20 hours, failing to capture the dynamic characteristics of time overlap. At 20 hours, the probability density calculated by this technical solution is 0.150, which is higher than the uniform distribution but lower than the linear distribution. This demonstrates that this solution more accurately identifies overlap at the boundaries of the time window. The advantage of this technical solution lies in its consideration of the combined effects of time fluctuations and elasticity. The probability density generated by a large number of sampling points more closely approximates the statistical laws of actual task execution. This precise probability density modeling is crucial for accurately calculating temporal overlap, enabling the system to better predict and handle time conflicts during task execution, thereby improving overall scheduling efficiency.

[0140] In this embodiment, by introducing fuzzy time windows and probabilistic modeling, the risk of temporal overlap between tasks is precisely quantified, effectively improving the accuracy of temporal conflict identification. By integrating resource usage frequency and scarcity to assess resource occupancy intensity, a detailed modeling of resource competition relationships is achieved. Combining data flow dependency analysis with the attenuation mechanism of recursive propagation paths, the direct and indirect dependencies between tasks are accurately characterized. Compared to the existing techniques that often use static configuration or rough rules to handle task conflicts, which cannot balance the timing flexibility and resource coordination between tasks, this solution takes dynamic feature modeling as its core, takes into account execution uncertainty and system resource constraints, constructs a multi-dimensional conflict assessment mechanism, and achieves a comprehensive quantitative analysis of scheduling risks, effectively improving the accuracy and robustness of task scheduling and the overall operating efficiency of the system.

[0141] In an optional embodiment, collecting the execution status of the initial task scheduling data to generate task execution feedback data, and calculating the task risk assessment score based on the task execution feedback data includes:

[0142] Collect the execution status of the initial task scheduling data, dynamically adjust the sampling interval based on the state change frequency of the task execution node, and group the collected execution status according to adaptive time windows; calculate the fluctuation variance of the execution status for each time window, automatically adjust the time decay weight based on the fluctuation variance, smooth the execution status based on the time decay weight, and generate task execution feedback data;

[0143] A multi-layer risk status propagation network is constructed, in which the bottom-layer nodes correspond to risk factors, the middle-layer nodes represent risk cluster categories, and the top-layer nodes represent the comprehensive risk level. The risk propagation coefficient between nodes is calculated based on the task execution feedback data. Based on the risk propagation coefficient, the node status of each layer is updated through bottom-up risk accumulation to obtain an initial risk state. The node status of each layer is then corrected through top-down constraint propagation to generate a risk state vector that includes task progress risk, resource utilization risk, and quality compliance risk.

[0144] The risk state vector and the task execution feedback data are subjected to feature combination, the correlation coefficient matrix between the features is calculated, a combination weight is assigned to each feature based on the correlation coefficient matrix, and the risk state vector and the task execution feedback data are weighted and fused according to the combination weight to generate a task risk assessment score.

[0145] This embodiment provides a task risk assessment method, which includes collecting the execution status of initial task scheduling data to generate task execution feedback data, and calculating a task risk assessment score based on the task execution feedback data.

[0146] Exemplarily, the system collects the execution status of the initial task scheduling data. This process dynamically adjusts the sampling time interval based on the state change frequency of the task execution node to achieve a balance between sampling efficiency and data accuracy. Specifically, the system monitors the frequency value of the state change of the execution node. When it detects that the state change frequency exceeds a preset threshold (for example, the number of state changes per minute is greater than 5 times), the system automatically shortens the sampling interval (such as from the original 60 seconds to 30 seconds); conversely, when the state change frequency is lower than another preset threshold (for example, the number of state changes per minute is less than 1 time), the system extends the sampling interval (such as from the original 30 seconds to 60 seconds). This adaptive sampling strategy can effectively reduce data redundancy while ensuring the capture of key state change points.

[0147] The collected execution status is grouped according to adaptive time windows. The time window size is dynamically adjusted based on task complexity and execution phase. For example, for development tasks, a smaller window (e.g., 2 hours) can be set during the coding phase, and a larger window (e.g., 4 hours) can be set during the testing phase. For each time window, the system calculates the execution status fluctuation variance, which indicates the stability of the execution status within that time window. In practical applications, the fluctuation variance can be obtained by calculating the statistical dispersion of status indicators (such as CPU utilization, memory usage, and task completion percentage) within the window.

[0148] Based on the calculated volatility variance, the system automatically adjusts the time-decay weights. A larger volatility variance indicates a more unstable state, so recent data is given a higher weight. A smaller volatility variance indicates a relatively stable state, so a more balanced weight distribution is adopted. For example, when the volatility variance exceeds 0.3, the weight of data for the most recent 30% of the time period can be set to 0.6, and the weight of data for the remaining 70% of the time period can be set to 0.4. When the volatility variance is less than 0.1, the weights can be set to 0.5 and 0.5, respectively, to achieve balanced weighting. Based on these time-decay weights, the system smooths the execution state and generates task execution feedback data. Smoothing effectively filters out the interference of short-term fluctuations and preserves the main trends of state changes.

[0149] After generating task execution feedback data, the system constructs a multi-layered risk status propagation network to assess task risk. This network employs a hierarchical structure, consisting of bottom-layer, middle-layer, and top-layer nodes. Bottom-layer nodes correspond to specific risk factors, such as insufficient resources, schedule delays, and code defects; middle-layer nodes represent risk clusters, such as resource risk, schedule risk, and quality risk; and top-layer nodes represent the overall risk level, reflecting the overall risk status of the task.

[0150] Based on task execution feedback data, the system calculates the risk propagation coefficient between nodes in the network. The risk propagation coefficient reflects the strength of the impact relationship between different risk factors, ranging from 0 to 1, with larger values ​​indicating a stronger impact. For example, in a software development task, the propagation coefficient of "insufficient testing resources" on "decreased code quality" might be 0.7, indicating a strong impact on code quality; whereas the propagation coefficient of "unstable development environment" on "schedule delays" might be 0.3, indicating a relatively weak impact.

[0151] Once the risk propagation coefficient is determined, the system updates the node states at each level through a bottom-up risk accumulation process to obtain the initial risk state. Specifically, starting from the bottom-level risk factor node, the risk state value is multiplied by the corresponding propagation coefficient and then accumulated and propagated to the upper-level nodes, layer by layer, until the top-level node is reached. The system then corrects the node states at each level through a top-down constraint propagation process, propagating the top-level comprehensive risk information back to the lower-level nodes. This corrects the initial risk assessment, resulting in a more comprehensive and accurate risk assessment result.

[0152] Through this two-way communication process, the system generates a risk state vector that includes task schedule risk, resource utilization risk, and quality compliance risk. For example, the risk state vector for a software project might be represented as [0.6, 0.3, 0.7], corresponding to the degree of schedule risk, resource risk, and quality risk, respectively. Larger values ​​indicate higher risk.

[0153] Finally, the system combines the risk state vector with the task execution feedback data to perform a feature combination and calculate a correlation coefficient matrix between the features. This correlation coefficient matrix reflects the strength of the association between each risk factor and the execution indicator, helping to identify key risk indicators. Based on the correlation coefficient matrix, the system assigns a combination weight to each feature. The stronger the correlation between the features, the higher the weight assigned. For example, for a company's product development task, if the correlation coefficient between "code submission frequency" and "quality compliance risk" is 0.8, then this feature may receive a weight of 0.3 in the quality risk assessment; while if the correlation coefficient between "team communication frequency" and "schedule risk" is 0.5, then it may receive a weight of 0.2 in the schedule risk assessment.

[0154] Based on the determined combined weights, the system combines the risk state vector with task execution feedback data to generate a final task risk assessment score. This score is typically expressed on a scale of 0 to 100, with higher scores indicating greater risk. For example, a risk assessment score of 75 for an R&D project indicates high risk and requires focused monitoring and intervention; a score of 25 for another project indicates low risk and allows for the maintenance of existing management strategies.

[0155] In this embodiment, through dynamic sampling and time window grouping, it is possible to accurately adapt to the frequency of task state changes, avoid information lag or data redundancy problems caused by fixed sampling intervals, and achieve efficient monitoring of task execution status. At the same time, through the time-attenuated weight mechanism driven by fluctuation variance, adaptive smoothing of state data is achieved, effectively suppressing the interference of occasional fluctuations on risk judgment. The use of a multi-layer risk state propagation network to perform structured modeling of task risks can comprehensively analyze potential risks in dimensions such as task progress, resource utilization, and quality control from multiple risk sources, and achieve hierarchical expression and dynamic correction of risk characteristics. Through feature correlation analysis and weight fusion mechanism, the comprehensiveness and accuracy of task risk assessment are effectively improved. Compared with the existing technology of evaluating task risks based on only a single indicator or static model, this solution can more accurately identify potential abnormal trends in task execution, thereby significantly improving the scheduling system's risk perception and response capabilities.

[0156] In an optional embodiment, generating a task adjustment instruction based on task status data and resource status data in the task execution feedback data, updating the task execution directed graph based on the task adjustment instruction, and generating the final task scheduling data includes:

[0157] The task status data and resource status data in the task execution feedback data are combined with the task risk assessment score to form a state input, and a task priority adjustment strategy network and a resource allocation strategy network are constructed based on the state input; the task priority adjustment strategy network generates a task priority adjustment coefficient according to the task status data, and the resource allocation strategy network generates a resource allocation ratio according to the resource status data;

[0158] The state input, task priority adjustment coefficient, resource allocation ratio and its corresponding task risk assessment score are stored in the experience replay pool; the policy network is trained based on the experience replay pool, the network parameters are optimized according to the changes in the task risk assessment score, and the optimized task priority adjustment coefficient and resource allocation ratio are combined into task adjustment instructions;

[0159] In the task execution directed graph, task nodes that need to be adjusted are identified based on the task priority adjustment coefficient in the task adjustment instruction; the degree of influence between nodes is calculated based on the task status data and resource status data; and the set of nodes whose influence exceeds the influence threshold is used to construct a subgraph to be adjusted.

[0160] Establishing a constraint network in the subgraph to be adjusted, updating dependencies between nodes based on task status data and resource status data, adjusting task priorities and resource allocation according to task adjustment instructions, and updating the task execution directed graph;

[0161] Task adjustment instructions and task execution feedback data are combined into training samples for storage; similar samples are retrieved based on current task status data and resource status data, the strategy network is continuously optimized, and the final task scheduling data is generated.

[0162] This embodiment provides an adaptive task scheduling method, which dynamically adjusts task priorities and resource allocation based on task execution feedback data to improve the overall execution efficiency of the system.

[0163] For example, task status data, resource status data, and task risk assessment scores are extracted from task execution feedback data to construct the status input. Task status data includes the completion degree, execution duration, and dependency completion status of each task; resource status data includes computing resource utilization, network bandwidth usage, and storage space consumption; and the task risk assessment score is calculated based on historical task execution data and current system load.

[0164] For example, for a distributed computing task, the system may collect the following status data: Task A is 70% complete, and the execution time has exceeded expectations by 30%; Task B is 35% complete, and the execution time is in line with expectations; CPU usage is 89%, memory usage is 65%, and network bandwidth usage is 42%; Task A has a risk assessment score of 0.78, and Task B has a risk assessment score of 0.45.

[0165] Based on the state input, the system constructs a task priority adjustment policy network and a resource allocation policy network. The task priority adjustment policy network consists of an input layer, two hidden layers, and an output layer. The input layer receives task state data, and the output layer generates priority adjustment coefficients for each task. The hidden layer uses the ReLU activation function to process features and uses an attention mechanism to focus on key task state indicators. In the above example, the network might output a priority adjustment coefficient of +0.25 (increase priority) for Task A and a coefficient of -0.15 (decrease priority) for Task B.

[0166] The resource allocation policy network has a similar structure, but it focuses on processing resource status data and outputting allocation ratios for each resource type. This network uses a softmax function to ensure that the sum of all resource allocation ratios is 1. Continuing with the above example, the network might allocate 70% of the CPU resources and 60% of the memory resources to Task A, and 30% of the CPU resources and 40% of the memory resources to Task B.

[0167] The system stores state inputs, task priority adjustment coefficients, resource allocation ratios, and corresponding task risk assessment scores in an experience replay pool. The experience replay pool uses a key-value storage structure and a hash index to accelerate retrieval of similar states. Each stored record contains the state vector, the adjustment strategy employed, and the resulting risk score change.

[0168] The system uses a temporal difference learning algorithm to train the policy network based on data from the experience replay pool. Specifically, the system randomly draws a batch of experience samples, calculates the difference between the expected risk score under the current policy and the actual risk score change, and then updates the network parameters using a backpropagation algorithm. The optimization goal is to minimize the task risk assessment score, thereby increasing the task execution success rate and reducing resource waste. After multiple rounds of training, the network parameters gradually converge, generating more optimal task priority adjustment coefficients and resource allocation ratios, which form the task adjustment instructions.

[0169] After the task adjustment instruction is generated, the system identifies the task nodes that need to be adjusted in the task execution directed graph. Specifically, the system traverses the task execution directed graph and selects the nodes whose absolute value of the priority adjustment coefficient exceeds the adjustment threshold (for example, 0.2) as adjustment targets. For the above example, the adjustment coefficient of task A is +0.25, which exceeds the threshold and is therefore marked as requiring adjustment. Next, the system calculates the degree of influence between the nodes. The degree of influence is calculated by considering the direct dependency between nodes, the degree of shared resources, and the historical execution correlation. The system uses a sliding window algorithm to analyze recent task execution data and identify strongly correlated task pairs. For example, if the execution status of task C is highly correlated with task A (correlation coefficient 0.85, exceeding the influence threshold of 0.7), task C is also included in the scope to be adjusted. The system constructs these nodes into a subgraph to be adjusted.

[0170] Within the subgraph to be adjusted, the system builds a constraint network and updates the dependencies between nodes. The constraint network is represented using a graph data structure, with nodes representing tasks and edges representing dependencies or resource constraints. The system updates edge weights based on the latest task status data to reflect the current strength of the dependency. For example, if a new data dependency is discovered between Task A and Task D, the system adds an edge from Task D to Task A to the constraint network with a weight of 0.9, indicating a strong dependency.

[0171] The system adjusts the priorities and resource allocations of tasks in the subgraph to be adjusted based on the task adjustment instructions. For tasks with increased priority (such as Task A), the system uses a topological sorting algorithm to advance them to the earliest execution position allowed by the dependencies. For resource allocation adjustments, the system modifies the resource scheduler's allocation parameters to ensure that high-priority tasks receive sufficient resources. Once the update is complete, the system synchronizes the changes to the task execution directed graph.

[0172] During implementation, the system stores each task adjustment instruction and task execution feedback data as training samples. These samples include the pre-adjustment state, the adjustment actions taken, and the post-adjustment system performance. Using the K-nearest neighbor algorithm, the system retrieves similar historical samples based on current task state and resource status data, analyzes historical adjustment results, and further optimizes the current policy network parameters.

[0173] Based on the updated task execution directed graph, the system generates task orchestration data, including the execution order, priority, resource allocation plan, and estimated completion time for each task. This task orchestration data is output in JSON format and distributed to the task scheduler for execution. For example, the final orchestration data might show that Task A's priority has been increased from 3 to 4, its CPU resource allocation has increased from 50% to 70%, and its estimated completion time has been reduced by 25%.

[0174] Figure 4 This is a schematic diagram of the directed graph structure of task execution according to an embodiment of the present invention. Figure 4 The graph, shown in Figure 1, illustrates the complete task dependency structure and critical path. The graph contains 16 task nodes distributed across six levels, with directed edges representing inter-task dependencies. Tasks 3, 7, 11, 13, 14, and 16 (circular nodes) form the critical path, with a total execution time of 87.5 minutes and a total impact of 4.61. Each task node is labeled with a priority value, ranging from 0.4 (Task 10) to 0.95 (Task 16). Nodes on the critical path generally have high priorities. The impact between tasks is clearly marked next to the connecting lines. The impact from Task 1 to Task 3 is the highest, at 0.95; Task 3 to Task 7 is 0.92; Task 11 to Task 14 is 0.91; and Task 14 to Task 16 is 0.94. These high-impact connections are prevalent on the critical path. The subgraph to be adjusted is marked with a dashed box in the figure. This contains nodes with an impact exceeding the threshold of 0.85 and will be prioritized for adjustment. Through this graphical representation, the system can intuitively identify bottlenecks and optimization opportunities during task execution, thereby applying targeted task priority adjustment coefficients and resource allocation ratios to improve overall execution efficiency. Compared to traditional static task scheduling, this dynamic adjustment method can more accurately respond to state changes and resource fluctuations during execution.

[0175] This embodiment, by constructing a task priority adjustment strategy network and a resource allocation strategy network, can dynamically generate task adjustment instructions according to the real-time task status and resource status, thereby realizing intelligent adjustment and optimization in the task scheduling process. Compared with the existing technology that generally adopts rule-driven or static models to sort tasks and allocate resources, it is unable to effectively deal with the risk fluctuations and resource bottlenecks that occur during task execution. This application adopts a data-driven strategy network mechanism to continuously optimize network parameters through the experience replay pool to improve the adaptability and robustness of the adjustment instructions. Furthermore, by constructing a subgraph to be adjusted and introducing an impact degree analysis and constraint network update method, it is possible to accurately identify key adjustment areas and effectively control the adjustment range, avoiding the chain effects caused by large changes in the global task structure. Ultimately, continuous training and updating of the strategy network is achieved in the feedback closed loop, which significantly enhances the system's responsiveness and self-evolution capabilities in a dynamic environment, and improves the intelligence level of task scheduling and the efficiency of full-process scheduling.

[0176] According to a second aspect of the embodiments of the present invention,

[0177] An electronic device is provided, comprising:

[0178] processor;

[0179] a memory for storing processor-executable instructions;

[0180] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0181] According to a third aspect of the embodiments of the present invention,

[0182] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0183] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0184] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for decomposing and arranging conference activity execution tasks based on AI technology, characterized in that: include: Obtaining conference activity demand information and converting the conference activity demand information into digital representation data of the conference activity; Based on the digital representation data of conference activities, a bidirectional recursive decomposition strategy is used to decompose tasks. The bidirectional recursive decomposition strategy determines the optimal task decomposition granularity through top-down goal decomposition and bottom-up resource reorganization iterative operations, generating multiple subtask data. Calculate the similarity between multiple subtask data to obtain a task similarity matrix, calculate the correlation entropy between multiple subtask data to obtain a task conflict matrix, combine the task similarity matrix and the task conflict matrix to construct a task execution directed graph, and generate initial task scheduling data based on the task execution directed graph; Collect the execution status of the initial task scheduling data to generate task execution feedback data, calculate the task risk assessment score based on the task execution feedback data, and when the task risk assessment score exceeds a preset threshold, generate a task adjustment instruction based on the task status data and resource status data in the task execution feedback data, update the task execution directed graph based on the task adjustment instruction, and generate the final task scheduling data; Calculate the similarity between multiple subtask data to obtain a task similarity matrix, calculate the correlation entropy between multiple subtask data to obtain a task conflict matrix, combine the task similarity matrix and the task conflict matrix to construct a task execution directed graph, and generate initial task scheduling data based on the task execution directed graph, including: Extract task type, goal, and resource requirement information from multiple subtask data, generate attribute vectors for each subtask, calculate cosine similarity between any two subtask attribute vectors, and construct a task similarity matrix; By comparing the execution time intervals of any two subtasks, the time overlap is calculated, the resource contention is calculated through resource allocation conflict detection, and the task dependency strength is calculated through task predecessor relationship analysis. The task correlation entropy is calculated after normalizing the time overlap, resource contention, and task dependency strength to construct a task conflict matrix. Adaptively weighting the task similarity matrix and the task conflict matrix, dynamically adjusting the similarity weights and the conflict weights according to the urgency of the tasks to generate a combined weight matrix, constructing a task execution directed graph based on the combined weight matrix, calculating the sum of the out-degree weights of each vertex in the task execution directed graph, and calculating the transfer influence of the vertex based on the combined weights with the adjacent vertices, and combining the sum of the out-degree weights with the transfer influence to obtain a dynamic priority score for the vertex; The vertex with the highest dynamic priority score is selected as the starting node, and the task execution directed graph is traversed. After each vertex is traversed, the dynamic priority scores of the remaining vertices are recalculated, and the next traversal vertex is determined according to the updated priority scores to generate initial task scheduling data.

2. The method for decomposing and arranging conference activity execution tasks based on AI technology according to claim 1 is characterized in that: Obtaining conference activity demand information and converting the conference activity demand information into digital representation data of conference activities includes: Constructing a knowledge graph for the field of conference activities, wherein the knowledge graph includes multi-dimensional information nodes and the relationships between the nodes, generating a structured demand collection template based on the knowledge graph for the field of conference activities, and obtaining conference activity demand information through the structured demand collection template; Obtain the text content in the conference activity demand information, extract the semantic features of the text content, perform sequence annotation on the semantic features, identify the entity and relationship information in the text content, and generate text structured information based on the entity and relationship information; Input the conference activity demand information and text structured information into a multi-head attention network, which calculates attention weights based on information reliability, and performs weighted fusion of the conference activity demand information and text structured information according to the attention weights to obtain fusion features; The fusion features are matched and mapped with the information nodes in the knowledge graph of the conference activity field to generate digital representation data of the conference activity.

3. The method for decomposing and arranging conference activity execution tasks based on AI technology according to claim 1 is characterized in that: Based on the digital representation data of conference activities, a bidirectional recursive decomposition strategy is used to decompose tasks. The bidirectional recursive decomposition strategy determines the optimal task decomposition granularity through top-down goal decomposition and bottom-up resource reorganization iterative operations, generating multiple subtask data including: Based on the digital representation data of conference activities, a task goal tree is constructed. The task goal tree contains multiple levels of nodes, each node represents a task goal, and there is a hierarchical dependency relationship between task goals; Based on the hierarchical dependency, the top-down goal decomposition in the bidirectional recursive decomposition strategy is used to decompose the upper-level task goal into multiple lower-level sub-goals step by step, and the complexity score of each sub-goal is calculated to obtain the initial sub-goal set; Calculate the target correlation of adjacent sub-targets in the initial sub-target set, and use the bottom-up resource reorganization strategy in the bidirectional recursive decomposition strategy to merge sub-targets. Merge adjacent sub-targets with correlations higher than the preset correlation threshold into new sub-targets to obtain a reorganized sub-target set. Calculating a resource consumption index for each sub-goal in the reorganized sub-goal set, and performing weighted calculation on the resource consumption index and the complexity score to obtain a task decomposition evaluation index; Iterate the initial sub-goal set and the reorganized sub-goal set, and optimize based on the task decomposition evaluation index. When the task decomposition evaluation index reaches the optimal value, the optimal task decomposition granularity is obtained. The task target tree is decomposed based on the optimal task decomposition granularity, and the task target information and resource requirement information of each sub-target node obtained by the decomposition are constructed into corresponding sub-task data to generate multiple sub-task data.

4. The method for decomposing and arranging conference activity execution tasks based on AI technology according to claim 1 is characterized in that: By comparing the execution time intervals of any two subtasks, the time overlap is calculated. By detecting resource allocation conflicts, the resource contention is calculated. By analyzing the task precedence relationship, the task dependency strength is calculated. Obtain execution time intervals for multiple subtasks, calculate the time fluctuation range of each subtask based on historical execution data, determine a time elasticity coefficient based on the time fluctuation range, and combine the time elasticity coefficient with the execution time interval to construct a fuzzy time window; Probability density modeling is performed on the fuzzy time windows of any two subtasks. A time point sequence is generated using the Monte Carlo sampling method. The task overlap probability at each time point is calculated. The task overlap probability is multiplied by the time interval between the sampling time points and the sum is used to obtain the temporal overlap degree. Scan and count the resource application lists of each subtask to obtain resource application frequency data, calculate the resource usage frequency based on the resource application frequency data; calculate the resource scarcity based on the ratio of the total resource amount to the total application amount; multiply the usage frequency by the resource scarcity to obtain the occupancy intensity weight of the resource type; Based on the occupancy intensity weights, calculate the product of the resource requirements of any two subtasks on the same resource type, multiply the resource requirements by the occupancy intensity weight to obtain the weighted resource requirement; divide the weighted resource requirement by the total resource requirements of the two subtasks to obtain the resource contention degree; A task dependency directed graph is constructed. The direct dependency relationship between task nodes is determined through data flow analysis and set as edge weight. The influence propagation path is calculated recursively based on the edge weight. The edge weight of the direct dependency relationship is combined with the attenuation weight on the influence propagation path to obtain the task dependency strength.

5. The method for decomposing and arranging conference activity execution tasks based on AI technology according to claim 1 is characterized in that: Collecting the execution status of the initial task scheduling data to generate task execution feedback data, and calculating the task risk assessment score based on the task execution feedback data includes: Collect the execution status of the initial task scheduling data, dynamically adjust the sampling interval based on the state change frequency of the task execution node, and group the collected execution status according to adaptive time windows; calculate the fluctuation variance of the execution status for each time window, automatically adjust the time decay weight based on the fluctuation variance, smooth the execution status based on the time decay weight, and generate task execution feedback data; A multi-layer risk status propagation network is constructed, in which the bottom-layer nodes correspond to risk factors, the middle-layer nodes represent risk cluster categories, and the top-layer nodes represent the comprehensive risk level. The risk propagation coefficient between nodes is calculated based on the task execution feedback data. Based on the risk propagation coefficient, the node status of each layer is updated through bottom-up risk accumulation to obtain an initial risk state. The node status of each layer is then corrected through top-down constraint propagation to generate a risk state vector that includes task progress risk, resource utilization risk, and quality compliance risk. The risk state vector and the task execution feedback data are subjected to feature combination, the correlation coefficient matrix between the features is calculated, a combination weight is assigned to each feature based on the correlation coefficient matrix, and the risk state vector and the task execution feedback data are weighted and fused according to the combination weight to generate a task risk assessment score.

6. The method for decomposing and arranging conference activity execution tasks based on AI technology according to claim 1 is characterized in that: Generate task adjustment instructions based on the task status data and resource status data in the task execution feedback data, update the task execution directed graph based on the task adjustment instructions, and generate the final task scheduling data including: The task status data and resource status data in the task execution feedback data are combined with the task risk assessment score to form a state input, and a task priority adjustment strategy network and a resource allocation strategy network are constructed based on the state input; the task priority adjustment strategy network generates a task priority adjustment coefficient according to the task status data, and the resource allocation strategy network generates a resource allocation ratio according to the resource status data; The state input, task priority adjustment coefficient, resource allocation ratio and its corresponding task risk assessment score are stored in the experience replay pool; the policy network is trained based on the experience replay pool, the network parameters are optimized according to the changes in the task risk assessment score, and the optimized task priority adjustment coefficient and resource allocation ratio are combined into task adjustment instructions; In the task execution directed graph, task nodes that need to be adjusted are identified based on the task priority adjustment coefficient in the task adjustment instruction; the degree of influence between nodes is calculated based on the task status data and resource status data; and the set of nodes whose influence exceeds the influence threshold is used to construct a subgraph to be adjusted. Establishing a constraint network in the subgraph to be adjusted, updating dependencies between nodes based on task status data and resource status data, adjusting task priorities and resource allocation according to task adjustment instructions, and updating the task execution directed graph; Task adjustment instructions and task execution feedback data are combined into training samples for storage; similar samples are retrieved based on current task status data and resource status data, the strategy network is continuously optimized, and the final task scheduling data is generated.

7. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the conference activity execution task decomposition and orchestration management method based on AI technology as described in any one of claims 1 to 6.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method for decomposing and arranging conference activity execution tasks based on AI technology as described in any one of claims 1 to 6 is implemented.

Images