A project management method and system based on multi-agent collaboration

Abstract

The embodiment of the application discloses a project management method and system based on multi-agent cooperation, the method comprises the following steps: acquiring multi-source project data, and processing the multi-source project data to generate a unified event semantic unit; constructing a unified event semantic graph based on the unified event semantic unit, and performing risk propagation reasoning based on the unified event semantic graph to generate a risk propagation chain and corresponding risk level information; controlling multiple functional agents to perform negotiation reasoning based on the risk propagation chain and the risk level information to generate a target cooperative treatment scheme; and writing feedback information corresponding to the target cooperative treatment scheme back to the unified event semantic graph to update the risk propagation strategy, the agent negotiation strategy and the subsequent scheme recommendation strategy. The embodiment of the application can improve the project risk early warning accuracy, the collaborative decision-making efficiency and the intelligent level of project management.

Description

Technical Field

[0001] This application belongs to the field of computer technology, specifically relating to a project management method and system based on multi-agent collaboration. Background Technology

[0002] Currently, project management software, R&D collaboration platforms, and AI assistants are widely used for task scheduling, defect tracking, progress reporting, and risk alerts. Common products typically use tasks, requirements, defects, code commits, and meeting minutes as management centers, improving project data recording and team collaboration efficiency.

[0003] However, existing systems often focus on data display within a single tool or rule-based reminders, resulting in inconsistent semantics among data from different sources. Events such as task delays, requirement changes, defect aggregation, and resource conflicts lack unified expression and traceable relationships. Even when large language models are introduced, they often remain at the level of summary generation, question-and-answer assistance, or simple suggestions. Summary of the Invention

[0004] The purpose of this application is to provide a project management method and system based on multi-agent collaboration to address the shortcomings of existing technologies in lacking a unified expression and traceable relationships.

[0005] To solve the above-mentioned technical problems, this application is implemented as follows: Firstly, a project management method based on multi-agent collaboration is provided, including the following steps: Acquire multi-source project data and process the multi-source project data to generate a unified event semantic unit; A unified event semantic graph is constructed based on the unified event semantic unit, and risk propagation reasoning is performed based on the unified event semantic graph to generate a risk propagation chain and corresponding risk level information. Based on the risk propagation chain and the risk level information, control multiple functional agents to perform negotiation reasoning and generate a collaborative disposal plan for the target. The feedback information corresponding to the target collaborative handling scheme is written back to the unified event semantic graph to update the risk propagation strategy, agent negotiation strategy, and subsequent scheme recommendation strategy.

[0006] Secondly, a project management system based on multi-agent collaboration is provided, including: The acquisition module is used to acquire multi-source project data and process the multi-source project data to generate a unified event semantic unit. The construction module is used to construct a unified event semantic graph based on the unified event semantic unit, and to perform risk propagation reasoning based on the unified event semantic graph to generate a risk propagation chain and corresponding risk level information. The generation module is used to control multiple functional agents to perform negotiation reasoning based on the risk propagation chain and the risk level information, and generate a target collaborative disposal plan; The update module is used to write back the feedback information corresponding to the target collaborative handling scheme to the unified event semantic graph, so as to update the risk propagation strategy, agent negotiation strategy and subsequent scheme recommendation strategy.

[0007] This application embodiment constructs a unified event semantic graph to achieve unified modeling and cross-system correlation analysis of multi-source project data. Combined with risk propagation reasoning and multi-agent collaborative decision-making, it can accurately identify risk propagation paths and impact ranges, generate targeted collaborative handling solutions, and continuously optimize risk propagation and negotiation strategies based on feedback. This can improve the accuracy of project risk early warning, the efficiency of collaborative decision-making, and the level of intelligent project management. Attached Figure Description

[0008] Figure 1 This is a flowchart of a project management method based on multi-agent collaboration provided in an embodiment of this application; Figure 2 This is a schematic diagram of the structure of a project management system based on multi-agent collaboration provided in an embodiment of this application. Detailed Implementation

[0009] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0010] To address the problems in existing technologies such as inconsistent semantics of multi-source data, biased risk warnings, lack of traceable evidence chains in AI summaries, lack of structured negotiation in multi-agent collaboration, and difficulty in accumulating user feedback into constraints for subsequent reasoning, this application proposes a multi-agent collaborative project management system and method based on unified event semantics and risk propagation reasoning.

[0011] The purpose of this application embodiment is to: uniformly convert heterogeneous data from task systems, document systems, code systems, communication systems, and local project tables into event semantic units with event identifiers, event types, sources, subjects, objects, times, scope of impact, semantic fingerprints, and version information; construct a continuously updated project event semantic graph based on the event semantic units; identify risk sources, expand risk propagation paths, and calculate propagation scores on the event semantic graph; when a risk reaches the triggering condition, enable multiple agents to submit evidence, verify constraints, generate candidate solutions, score and rank them, and output consensus suggestions around the same risk propagation chain; and simultaneously utilize context-shared memory to save user feedback and handling results to adjust subsequent risk credibility and solution recommendation weights.

[0012] Based on the above objectives, the embodiments of this application aim to solve the following technical problems: the difficulty in automatically aligning project objects due to inconsistencies in data fields, status definitions, and object identifiers among different project management tools, document platforms, code platforms, and communication platforms; the problem that existing risk alerts, based solely on single-task or single-field rules, cannot calculate risk propagation paths along dependency, blocking, responsibility, attribution, association, and influence relationships; the problem that AI project summaries or natural language question answering lack traceable risk evidence chains and make it difficult to verify the source of risk conclusions; the problem that multiple intelligent agents or automated modules only call each other serially, lacking structured evidence exchange, constraint verification, and conflict resolution mechanisms; and the problem that user feedback and historical handling results cannot continuously influence subsequent risk scoring, alert thresholds, and solution recommendations.

[0013] This application proposes a multi-agent collaborative project management system and method based on unified event semantics and risk propagation reasoning. Its core is not to simply superimpose multiple agents, large language models or message buses, but to first establish a unified event semantic layer, then use an event semantic graph to carry project facts and relationships, and on this basis, perform risk propagation chain reasoning and multi-agent negotiation.

[0014] Specifically, the system includes a multi-source event acquisition module, a unified event semantics module, an event semantic graph module, a risk propagation inference module, a multi-agent negotiation module, a context-shared memory module, and a project management output module. The multi-source event acquisition module connects to the project management system, code repository, defect system, document system, and communication system to extract project events such as additions, changes, completions, blockages, delays, defects, comments, and meeting conclusions. The unified event semantics module transforms data from different sources into unified event objects. A unified event object includes at least an event identifier, event type, event subject, event object, occurrence time, source system, state change, context summary, associated objects, confidence level, and risk label. The event semantic graph module writes the unified event objects into a graph structure, using events, tasks, requirements, defects, personnel, milestones, and deliverables as nodes, and generation, association, dependency, blockage, impact, mitigation, and confirmation as edges, forming a traceable project event semantic graph. The risk propagation inference module identifies the risk origin, propagation path, and affected objects in the event semantic graph, and calculates the risk level based on path length, dependency strength, event confidence level, impact scope, and historical feedback. The multi-agent negotiation module organizes planning agents, risk agents, resource agents, and quality agents to propose analytical opinions on the same event facts, and generates collaborative handling suggestions through conflict identification, scheme comparison, and consensus judgment. The context-sharing memory module saves event facts, historical reasoning results, human confirmation opinions, agent negotiation conclusions, and execution feedback, enabling subsequent analyses to reuse the same context.

[0015] In this embodiment, the unified event semantic model can be represented as: E={id,type,subject,object,time,source,state,context,relations,confidence,riskTag} Here, type represents the event category, subject represents the triggering entity, object represents the affected object, relations represents the relationship with other events or project objects, and riskTag represents the type of risk that the event may cause.

[0016] Through the above model, data such as "task delay", "requirement change", "test failure", "code rollback", "resource leave" and "meeting decision to adjust scope" from different systems are all expressed as events with the same caliber, thus providing a unified factual basis for cross-system reasoning.

[0017] Furthermore, when the system detects a risk-related event, the risk propagation inference module uses that event as a starting point and expands along edges such as dependency, blockage, and impact in the event semantic graph to obtain candidate propagation paths. Each path records the starting event, intermediate impact nodes, ending project goal, and path evidence. The risk score can be calculated based on event confidence, edge weights, importance of affected objects, results of historical similar events, and path decay coefficient. The system outputs the risk level based on the risk score and simultaneously outputs the propagation chain, allowing users to see the cause and scope of the risk.

[0018] In the multi-agent negotiation mechanism of this application embodiment, the planning agent focuses on analyzing the impact of scheduling and milestones; the risk agent focuses on analyzing risk levels and propagation paths; the resource agent focuses on analyzing personnel load and resource conflicts; and the quality agent focuses on analyzing the impact of defects, testing, and acceptance. Each agent reads the same event semantic graph and shared context memory, then outputs structured opinions. The negotiation module performs a consistency judgment on the opinions of each agent. When opinions are consistent, a recommended solution is formed; when opinions conflict, the system compares them based on project goal priority, risk level, resource constraints, and historical feedback to generate candidate solutions and reasons for recommendation.

[0019] In the context-shared memory and closed-loop update steps of this application embodiment, the shared memory records project stages, key decisions, confirmed risks, adopted solutions, and handling results. Manual confirmation or actual execution results are then converted back into unified events, which are written back into the event semantic graph and context-shared memory, thus forming a closed loop of "event input, risk reasoning, agent negotiation, execution feedback, and memory update."

[0020] The following description, in conjunction with the accompanying drawings, details the project management method based on multi-agent collaboration provided in this application through specific embodiments and application scenarios.

[0021] like Figure 1 The diagram shown is a flowchart of a project management method based on multi-agent collaboration provided in an embodiment of this application. The method includes the following steps: Step 101: Obtain multi-source project data and process the multi-source project data to generate a unified event semantic unit.

[0022] Specifically, multi-source project data can be acquired from project management systems, code repository systems, defect management systems, document systems, and communication systems; task identifiers, requirement identifiers, defect identifiers, personnel identifiers, version identifiers, and timestamp information can be extracted from the multi-source project data; the extracted results are then subjected to field mapping and semantic standardization processing based on a preset event semantic template, and a corresponding event semantic vector is generated using a semantic embedding model; the event semantic vector is combined and encoded with the task identifier, requirement identifier, defect identifier, personnel identifier, version identifier, and timestamp information to generate an event semantic fingerprint; cross-system entity alignment, duplicate event elimination, and associated object matching are performed based on the event semantic fingerprint to obtain standardized event data, and a unified event semantic unit is constructed based on the standardized event data.

[0023] The unified event semantic unit includes event identifier, event type, event subject, event object, event time, source system, status information, association information, confidence level information, and risk label information.

[0024] Step 102: Construct a unified event semantic graph based on the unified event semantic unit, and perform risk propagation reasoning based on the unified event semantic graph to generate a risk propagation chain and corresponding risk level information.

[0025] Specifically, the unified event semantic unit can be mapped to event nodes, and task objects, requirement objects, defect objects, personnel objects, version objects, milestone objects, and deliverable objects associated with the unified event semantic unit can be extracted and corresponding entity nodes can be constructed. Event entity relationship edges can be established based on the triggering relationship, ownership relationship, and association relationship between event nodes and entity nodes. Entity relationship edges can be established based on the task dependency relationship, requirement association relationship, resource ownership relationship, defect impact relationship, and milestone constraint relationship between entity nodes. Based on the association path formed by event nodes and entity nodes, the dependency relationship, blocking relationship, impact relationship, and mitigation relationship between event nodes can be deduced, and corresponding graph edge connections can be established to form a unified event semantic graph representing project objects, project events, and their semantic relationships.

[0026] Furthermore, risk-triggered events are identified, and the event nodes corresponding to the risk-triggered events are used as starting nodes. Graph traversal reasoning is performed along the dependency, blocking, and influence edges between the event nodes to generate a risk propagation chain that includes the risk origin, propagation nodes, affected objects, and propagation evidence. A risk propagation score is calculated based on event confidence, relationship weight, importance of affected objects, path decay factor, and historical feedback information, and the corresponding risk level information is determined based on the risk propagation score.

[0027] Step 103: Based on the risk propagation chain and the risk level information, control multiple functional agents to perform negotiation reasoning and generate a target collaborative disposal plan.

[0028] Specifically, the risk propagation chain and the risk level information can be sent to multiple functional agents, and the multiple functional agents can be controlled to generate risk analysis results, resource analysis results, quality analysis results and plan analysis results respectively based on the unified event semantic graph and context shared memory; based on the evidence exchange, constraint verification, conflict detection and scheme evaluation process between agents, negotiation reasoning is performed to generate a target collaborative disposal scheme.

[0029] Step 104: Write back the feedback information corresponding to the target collaborative handling scheme to the unified event semantic graph to update the risk propagation strategy, agent negotiation strategy, and subsequent scheme recommendation strategy.

[0030] Specifically, the execution results, user feedback information, risk confirmation results, and agent negotiation results corresponding to the target collaborative handling scheme can be written into the context shared memory; the feedback information in the context shared memory can be converted into new unified event semantic units and written back to the unified event semantic graph to update the relationship weights between event nodes, risk propagation scoring parameters, agent negotiation weights, and subsequent scheme recommendation strategies.

[0031] This application embodiment constructs a unified event semantic graph to achieve unified modeling and cross-system correlation analysis of multi-source project data. Combined with risk propagation reasoning and multi-agent collaborative decision-making, it can accurately identify risk propagation paths and impact ranges, generate targeted collaborative handling solutions, and continuously optimize risk propagation and negotiation strategies based on feedback. This can improve the accuracy of project risk early warning, the efficiency of collaborative decision-making, and the level of intelligent project management.

[0032] In this application embodiment, the following implementation process is specifically involved: I. Data Access and Event Generation The system integrates with the project management system, code repository, defect system, documentation system, and communication tools via interfaces, webhooks, database synchronization, or log monitoring. When changes in task status, scope adjustments, new defects, failed code submissions, or changes in meeting conclusions are detected, raw event logs are generated. These raw event logs are then transformed into unified event objects after field mapping, entity recognition, and context summarization. For example, "high-priority defect reopened" in the defect system, "test task postponed" in the task system, and "scope confirmation temporarily suspended" in communication logs are all transcribed into unified events, while retaining evidence of their origin.

[0033] II. Event Semantic Graph Construction The system writes unified event objects into an event semantic graph. If event objects involve the same requirement, task, personnel, version, or milestone, a related edge is created; if an event hinders another task, a blocking edge is created; if an event may affect delivery dates, quality goals, or resource allocation, an impact edge is created. Each edge in the event semantic graph can have its type, weight, source, and update time set. Thus, the system not only stores the project objects themselves but also the semantic relationships and temporal order between project changes.

[0034] III. Risk Propagation Reasoning Process When a risk-triggered event occurs, the system executes propagation reasoning starting from that event. First, it filters tasks, requirements, defects, personnel, and milestones directly related to the event. Then, it continues searching along dependency edges, blocking edges, and impact edges, obtaining one or more risk propagation chains. The system calculates a risk score for each risk propagation chain. Calculation factors include event confidence, priority of affected objects, number of blocking relationships in the path, time to key milestones, historical similar risk outcomes, and human confirmation feedback. Output results include risk level, propagation path, key evidence, and recommended action window.

[0035] For example, if the "core interface requirement change" event is dependent on "backend development tasks", "integration testing tasks", "test case updates", and "version release milestones", the system can infer that the change may propagate along the development, integration testing, and testing chain, and create a risk of release delay.

[0036] IV. Multi-agent negotiation process After the risk propagation results are generated, the planning agent proposes scheduling adjustments based on milestones and task dependencies; the resource agent proposes resource reallocation based on personnel workload; the quality agent proposes quality assurance recommendations based on defects and test coverage; and the risk agent proposes warning levels and handling priorities based on the propagation chain and risk scores. The negotiation module organizes the opinions of each agent into structured negotiation messages, with fields including suggested actions, underlying events, affected parties, expected benefits, costs, and confidence levels. If there are conflicts between suggestions, such as the planning agent suggesting shortening the testing cycle while the quality agent suggests increasing testing verification, the system will provide a compromise solution or escalate to manual confirmation based on risk level, project goal priority, and historical feedback.

[0037] V. Output and Feedback Closed Loop The system ultimately outputs risk warnings, propagation chain diagrams, a list of affected entities, recommended handling solutions, and conflict points requiring manual confirmation. After users confirm, reject, or modify suggestions, the system transforms the processing results into feedback events, writing them back to the event semantic graph and shared context memory. Through this approach, the system can continuously accumulate event facts and handling experience during project operation, ensuring that subsequent risk predictions do not rely on isolated indicators but rather on unified event semantics, risk propagation chains, and multi-agent negotiation results.

[0038] Compared with existing technologies, the embodiments of this application have at least the following beneficial effects: By unifying event semantic units and event semantic graphs, multi-source heterogeneous project data is transformed into a reasonable data structure, solving the problems of inconsistent field definitions and semantic misalignment in different systems; by reasoning through risk propagation chains, risk identification is upgraded from single-point task warning to chain-like impact analysis based on dependency, blocking, responsibility, attribution, and influence relationships, enabling early detection of critical path risks and cross-project chain risks; through a negotiated multi-agent mechanism, agents submit evidence, generate solutions, verify constraints, and form consensus around the same risk chain, which is different from ordinary multi-module serial calls or simple large-model question and answer; through context-shared memory, the system can accumulate historical risks, user feedback, negotiation records, and handling results, reducing duplicate warnings and improving the adaptability of risk scoring and solution recommendations; through risk evidence chains and propagation path outputs, the interpretability of project risk warnings and decision-making suggestions is improved, facilitating communication between project managers and business stakeholders, R&D teams, and management; through the closed loop of event flow, graph structure, and agent negotiation, a project management mechanism of continuous updating, continuous reasoning, and continuous optimization is formed. The aforementioned effects stem from the interaction between unified intermediate representation, time-series projection, computable propagation paths, structured negotiation, and feedback updates. In real-world projects, the risk lead time, reduction in repeated warnings, and shortening of response time can be further verified after deployment using historical project data.

[0039] like Figure 2 The diagram shown is a structural schematic of a project management system based on multi-agent collaboration provided in an embodiment of this application, including: The acquisition module 210 is used to acquire multi-source project data and process the multi-source project data to generate a unified event semantic unit.

[0040] Specifically, the acquisition module 210 is used to acquire multi-source project data from the project management system, code repository system, defect management system, document system, and communication system; extract task identifiers, requirement identifiers, defect identifiers, personnel identifiers, version identifiers, and timestamp information from the multi-source project data; perform field mapping and semantic standardization processing on the extraction results based on a preset event semantic template; and generate corresponding event semantic vectors using a semantic embedding model; combine and encode the event semantic vectors with the task identifiers, requirement identifiers, defect identifiers, personnel identifiers, version identifiers, and timestamp information to generate event semantic fingerprints; perform cross-system entity alignment, duplicate event elimination, and associated object matching based on the event semantic fingerprints to obtain standardized event data; and construct a unified event semantic unit based on the standardized event data; wherein, the unified event semantic unit includes event identifier, event type, event subject, event object, event time, source system, status information, association information, confidence information, and risk label information.

[0041] The construction module 220 is used to construct a unified event semantic graph based on the unified event semantic unit, and to perform risk propagation reasoning based on the unified event semantic graph to generate a risk propagation chain and corresponding risk level information.

[0042] Specifically, the construction module 220 is used to map the unified event semantic unit to event nodes, extract task objects, requirement objects, defect objects, personnel objects, version objects, milestone objects, and deliverable objects associated with the unified event semantic unit, and construct corresponding entity nodes; establish event entity relationship edges based on the triggering relationship, ownership relationship, and association relationship between event nodes and entity nodes; establish entity relationship edges based on the task dependency relationship, requirement association relationship, resource ownership relationship, defect impact relationship, and milestone constraint relationship between entity nodes; deduce the dependency relationship, blocking relationship, impact relationship, and mitigation relationship between event nodes based on the association path formed by event nodes and entity nodes, and establish corresponding graph edge connections to form a unified event semantic graph representing project objects, project events, and their semantic relationships.

[0043] Furthermore, the construction module 220 is also specifically used to identify risk-triggered events, and with the event node corresponding to the risk-triggered event as the starting node, perform graph traversal reasoning along the dependency relationship edges, blocking relationship edges, and influence relationship edges between the event nodes to generate a risk propagation chain containing the risk starting point, propagation nodes, affected objects, and propagation evidence; calculate the risk propagation score based on the event confidence, relationship weight, importance of affected objects, path decay factor, and historical feedback information, and determine the corresponding risk level information based on the risk propagation score.

[0044] The generation module 230 is used to control multiple functional agents to perform negotiation reasoning based on the risk propagation chain and the risk level information to generate a target collaborative disposal plan.

[0045] Specifically, the generation module 230 is used to send the risk propagation chain and the risk level information to multiple functional agents, control the multiple functional agents to generate risk analysis results, resource analysis results, quality analysis results and plan analysis results respectively based on the unified event semantic graph and context shared memory; and perform negotiation reasoning based on the evidence exchange, constraint verification, conflict detection and scheme evaluation process between agents to generate a target collaborative disposal scheme.

[0046] The update module 240 is used to write back the feedback information corresponding to the target collaborative handling scheme to the unified event semantic graph, so as to update the risk propagation strategy, agent negotiation strategy and subsequent scheme recommendation strategy.

[0047] Specifically, the update module 240 is used to write the execution results, user feedback information, risk confirmation results, and agent negotiation results corresponding to the target collaborative handling scheme into the context shared memory; and to convert the feedback information in the context shared memory into new unified event semantic units and write them back to the unified event semantic graph to update the relationship weights between event nodes, risk propagation scoring parameters, agent negotiation weights, and subsequent scheme recommendation strategies.

[0048] This application embodiment constructs a unified event semantic graph to achieve unified modeling and cross-system correlation analysis of multi-source project data. Combined with risk propagation reasoning and multi-agent collaborative decision-making, it can accurately identify risk propagation paths and impact ranges, generate targeted collaborative handling solutions, and continuously optimize risk propagation and negotiation strategies based on feedback. This can improve the accuracy of project risk early warning, the efficiency of collaborative decision-making, and the level of intelligent project management.

[0049] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, this computer program implements the various processes of the above-described multi-agent collaborative project management method embodiments and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0050] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0051] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0052] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A project management method based on multi-agent collaboration, characterized in that, Includes the following steps: Acquire multi-source project data and process the multi-source project data to generate a unified event semantic unit; A unified event semantic graph is constructed based on the unified event semantic unit, and risk propagation reasoning is performed based on the unified event semantic graph to generate a risk propagation chain and corresponding risk level information. Based on the risk propagation chain and the risk level information, control multiple functional agents to perform negotiation reasoning and generate a collaborative disposal plan for the target. The feedback information corresponding to the target collaborative handling scheme is written back to the unified event semantic graph to update the risk propagation strategy, agent negotiation strategy, and subsequent scheme recommendation strategy.

2. The method according to claim 1, characterized in that, The process of acquiring multi-source project data and processing the multi-source project data to generate a unified event semantic unit specifically includes: Acquire project data from multiple sources, including project management systems, code repository systems, defect management systems, document systems, and communication systems; Extract task identifiers, requirement identifiers, defect identifiers, personnel identifiers, version identifiers, and timestamp information from the multi-source project data. Based on a preset event semantic template, perform field mapping and semantic standardization on the extracted results, and use a semantic embedding model to generate corresponding event semantic vectors. The event semantic vector is combined and encoded with the task identifier, requirement identifier, defect identifier, personnel identifier, version identifier, and timestamp information to generate an event semantic fingerprint; Based on the event semantic fingerprint, cross-system entity alignment, duplicate event elimination, and associated object matching are performed to obtain standardized event data, and a unified event semantic unit is constructed based on the standardized event data. The unified event semantic unit includes event identifier, event type, event subject, event object, event time, source system, status information, association information, confidence level information, and risk label information.

3. The method according to claim 1, characterized in that, The construction of a unified event semantic graph based on the unified event semantic unit specifically includes: The unified event semantic unit is mapped to an event node, and the task object, requirement object, defect object, personnel object, version object, milestone object and deliverable object associated with the unified event semantic unit are extracted, and the corresponding entity node is constructed. Event-entity relationship edges are established based on the triggering, attribution, and association relationships between event nodes and entity nodes; entity relationship edges are established based on the task dependency, requirement association, resource attribution, defect impact, and milestone constraint relationships between entity nodes. Based on the association paths formed by event nodes and entity nodes, the dependencies, blocking relationships, influence relationships, and mitigation relationships between event nodes are deduced, and corresponding graph edge connections are established to form a unified event semantic graph that represents project objects, project events, and their semantic relationships.

4. The method according to claim 1, characterized in that, The step of performing risk propagation reasoning based on the unified event semantic graph to generate risk propagation chains and corresponding risk level information specifically includes: Identify risk-triggered events, and use the event node corresponding to the risk-triggered event as the starting node. Perform graph traversal reasoning along the dependency, blocking, and influence edges between the event nodes to generate a risk propagation chain that includes the risk starting point, propagation nodes, affected objects, and propagation evidence. The risk propagation score is calculated based on the event confidence level, relationship weight, importance of the affected object, path decay factor, and historical feedback information, and the corresponding risk level information is determined based on the risk propagation score.

5. The method according to claim 1, characterized in that, The process of controlling multiple functional agents to perform negotiation reasoning based on the risk propagation chain and risk level information to generate a collaborative response plan for the target specifically includes: The risk propagation chain and the risk level information are sent to multiple functional agents, and the multiple functional agents are controlled to generate risk analysis results, resource analysis results, quality analysis results and plan analysis results respectively based on the unified event semantic graph and context shared memory; Based on the evidence exchange, constraint verification, conflict detection and scheme evaluation process between intelligent agents, negotiation reasoning is performed to generate a collaborative solution for the target. The step of writing back the feedback information corresponding to the target collaborative handling scheme to the unified event semantic graph to update the risk propagation strategy, agent negotiation strategy, and subsequent scheme recommendation strategy specifically includes: Write the execution results, user feedback information, risk confirmation results, and agent negotiation results corresponding to the target collaborative handling scheme into the context shared memory; The feedback information in the shared context memory is converted into new unified event semantic units and written back to the unified event semantic graph to update the relationship weights between event nodes, risk propagation scoring parameters, agent negotiation weights, and subsequent solution recommendation strategies.

6. A project management system based on multi-agent collaboration, characterized in that, include: The acquisition module is used to acquire multi-source project data and process the multi-source project data to generate a unified event semantic unit. The construction module is used to construct a unified event semantic graph based on the unified event semantic unit, and to perform risk propagation reasoning based on the unified event semantic graph to generate a risk propagation chain and corresponding risk level information. The generation module is used to control multiple functional agents to perform negotiation reasoning based on the risk propagation chain and the risk level information, and generate a target collaborative disposal plan; The update module is used to write back the feedback information corresponding to the target collaborative handling scheme to the unified event semantic graph, so as to update the risk propagation strategy, agent negotiation strategy and subsequent scheme recommendation strategy.

7. The system according to claim 6, characterized in that, The acquisition module is specifically used to acquire multi-source project data from the project management system, code repository system, defect management system, document system, and communication system; extract task identifiers, requirement identifiers, defect identifiers, personnel identifiers, version identifiers, and timestamp information from the multi-source project data; perform field mapping and semantic standardization processing on the extraction results based on a preset event semantic template; and generate corresponding event semantic vectors using a semantic embedding model; combine and encode the event semantic vectors with the task identifiers, requirement identifiers, defect identifiers, personnel identifiers, version identifiers, and timestamp information to generate event semantic fingerprints; perform cross-system entity alignment, duplicate event elimination, and associated object matching based on the event semantic fingerprints to obtain standardized event data; and construct a unified event semantic unit based on the standardized event data; wherein, the unified event semantic unit includes event identifier, event type, event subject, event object, event time, source system, status information, association information, confidence information, and risk label information.

8. The system according to claim 6, characterized in that, The construction module is specifically used to map the unified event semantic unit to event nodes, extract task objects, requirement objects, defect objects, personnel objects, version objects, milestone objects, and deliverable objects associated with the unified event semantic unit, and construct corresponding entity nodes; establish event entity relationship edges based on the triggering relationship, ownership relationship, and association relationship between event nodes and entity nodes; establish entity relationship edges based on the task dependency relationship, requirement association relationship, resource ownership relationship, defect impact relationship, and milestone constraint relationship between entity nodes; deduce the dependency relationship, blocking relationship, impact relationship, and mitigation relationship between event nodes based on the association path formed by event nodes and entity nodes, and establish corresponding graph edge connections to form a unified event semantic graph representing project objects, project events, and their semantic relationships.

9. The system according to claim 6, characterized in that, The construction module is specifically used to identify risk-triggered events, and take the event node corresponding to the risk-triggered event as the starting node, perform graph traversal reasoning along the dependency, blocking, and influence edges between the event nodes to generate a risk propagation chain containing the risk starting point, propagation nodes, affected objects, and propagation evidence; calculate the risk propagation score based on the event confidence, relationship weight, importance of affected objects, path decay factor, and historical feedback information, and determine the corresponding risk level information based on the risk propagation score.

10. The method according to claim 1, characterized in that, The generation module is specifically used to send the risk propagation chain and the risk level information to multiple functional agents, control the multiple functional agents to generate risk analysis results, resource analysis results, quality analysis results and plan analysis results respectively based on the unified event semantic graph and context shared memory; and perform negotiation reasoning based on the evidence exchange, constraint verification, conflict detection and scheme evaluation process between agents to generate a target collaborative disposal scheme. The update module is specifically used to write the execution results, user feedback information, risk confirmation results, and agent negotiation results corresponding to the target collaborative handling scheme into the context shared memory; and to convert the feedback information in the context shared memory into new unified event semantic units and write them back to the unified event semantic graph to update the relationship weights between event nodes, risk propagation scoring parameters, agent negotiation weights, and subsequent scheme recommendation strategies.

Images