Multi-agent cooperation strategy generation method and device, equipment and medium

Abstract

The invention relates to a multi-agent cooperation strategy generation method, device and equipment and a medium, and the method comprises the steps: separating acoustic spectrum features and text semantic features of conference voice through environment perception processing, solving a cross-modal information conflict problem, and generating an accurate semantic understanding result; identifying the essence of the problem based on task analysis, associating the responsibility field, and constructing a classifiable problem point set; calling an agent capability library to dynamically match problem requirements, and generating a candidate agent list; quantifying a problem influence range and a decision time limit through weighted emergency scores, and generating a priority-sorted agent sequence; screening and confirming a core problem point and a primary agent; and finally, generating an executable cooperation scheme through multi-agent collaborative optimization. According to the method, the problems of incomplete feature extraction, task allocation delay and resource conflict in the prior art are solved, and the operability and decision-making efficiency of a cooperation strategy are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method, device, equipment and medium for generating a multi-agent collaboration strategy. Background Art

[0002] In the field of artificial intelligence and collaborative decision-making, multi-agent collaborative strategy generation methods have become a key technology for improving organizational decision-making efficiency. By coordinating the collaborative work of multiple AI entities with specific domain capabilities, these methods transform complex meeting discussions into actionable action plans. They are widely used in scenarios such as corporate strategy meetings, emergency command and dispatch, and intelligent manufacturing. Their core value lies in transforming multimodal information such as voice and text in human meetings into systematic collaborative instructions, enabling automated decision support from problem identification to solution generation.

[0003] However, existing technologies face significant bottlenecks when dealing with actual meeting scenarios. On the one hand, traditional methods find it difficult to effectively handle mixed voice interruptions, text ambiguities, and cross-modal information conflicts in meetings, resulting in incomplete extraction of key problem features; on the other hand, due to the lack of a dynamic evaluation mechanism for the urgency of the problem and the capabilities of the agent, high-priority task allocation delays and mismatches between professional capabilities and problem requirements often occur. What's more serious is that when generating the final strategy, existing solutions are often unable to effectively coordinate resource conflicts and logical contradictions between multiple agents, resulting in execution gaps or resource contention problems in the generated action plans. These defects make the generated collaborative strategies less operational, seriously restricting the practical application value of intelligent conference decision-making systems. Summary of the Invention

[0004] Based on this, the purpose of the present invention is to provide a multi-agent collaboration strategy generation method, device, equipment and medium that can accurately analyze meeting information, dynamically optimize agent collaboration and generate executable solutions.

[0005] The purpose of the present invention is achieved by the following scheme:

[0006] In a first aspect, the present invention provides a method for generating a multi-agent collaboration strategy, comprising the following steps:

[0007] S1: Performs environmental perception processing on the user-input conference voice and text data, separates acoustic spectrum features and text semantic features, extracts semantic key points, and generates a set of semantic understanding results;

[0008] S2: Perform task analysis on the semantic understanding result set, identify the essential characteristics of the problem and associate them with the responsibility areas, and generate a set of classified problem points;

[0009] S3: Call the agent capability library to perform resource matching on the classified problem point set, evaluate the adaptation relationship between the agent capabilities and the problem requirements, and generate a list of candidate agents. The candidate agent list is used to indicate the set of agents that can handle each problem point;

[0010] S4: Perform dynamic priority evaluation on the candidate agent list, calculate the weighted urgency score of the problem impact scope and decision time limit, and generate an evaluation agent sequence;

[0011] S5: Prioritize the evaluation agent sequence and confirm the primary decision information based on the weighted urgency score. The primary decision information is used to indicate the core problem points and the primary agent to solve the core problem points;

[0012] S6: Generate collaborative solutions for primary decision-making information, and generate collaborative strategy solutions through draft construction and multi-agent optimization integration.

[0013] In a second aspect, the present invention provides a multi-agent collaboration strategy generation device, which is configured with the following modules:

[0014] The environment perception processing module is used to perform environment perception processing on the conference voice and text data input by the user, separate the acoustic spectrum features and text semantic features, extract semantic key points, and generate a set of semantic understanding results;

[0015] The task analysis module is used to perform task analysis on the semantic understanding result set, identify the essential characteristics of the problem, associate the responsibility areas, and generate a set of classified problem points;

[0016] The agent resource matching module is used to call the agent capability library to perform resource matching processing on the classified problem point set, evaluate the adaptation relationship between the agent capability and the problem requirements, and generate a list of candidate agents. The candidate agent list is used to indicate the set of agents that can handle each problem point;

[0017] Dynamic priority evaluation module, which is used to perform dynamic priority evaluation on the candidate agent list, calculate the weighted urgency score of the problem impact scope and decision time limit, and generate the evaluation agent sequence;

[0018] The priority screening module is used to prioritize the evaluation agent sequence and confirm the primary decision information based on the weighted urgency score. The primary decision information is used to indicate the core problem points and the primary agent to solve the core problem points;

[0019] The collaborative solution generation module is used to generate collaborative solutions for primary decision-making information, and generate collaborative strategy solutions through draft construction and multi-agent optimization integration.

[0020] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements any of the above-mentioned multi-agent collaboration strategy generation methods when executing the computer program.

[0021] In a fourth aspect, the present application provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, it implements any of the above-mentioned multi-agent collaboration strategy generation methods.

[0022] In summary, the multi-agent collaborative strategy generation method provided by the present application can capture semantic key points in complex scenarios such as speech interruption and text ambiguity through a dual-modal parsing mechanism that separates acoustic spectrum features and text semantic features in the environmental perception processing stage, thereby solving the problem of incomplete feature extraction caused by cross-modal information conflicts in traditional methods; in the task analysis link, based on the in-depth analysis of the association between the identification of the essential features of the problem and the responsibility field, it can realize the dimensional decomposition and accurate classification of the problem features, ensuring zero omission of key problem elements; by dynamically calling the resource matching mechanism of the agent capability library, it can realize real-time adaptive evaluation of the agent capabilities and problem requirements, and completely avoid the waste of professional resources caused by capability mismatch; adopt The dynamic priority evaluation model with weighted urgency scoring can realize the dual-factor quantitative analysis of the scope of impact and the decision-making time limit, so as to solve the pain point of delayed allocation of high-priority tasks; the priority screening process based on scoring sorting can achieve the precise matching of core problem points and the optimal intelligent agent, ensuring the timeliness of response to key issues; finally, through the multi-agent collaborative optimization framework, it can achieve the dual guarantee of resource conflict detection and logical contradiction resolution, so as to generate an operational solution without execution faults, thereby fundamentally solving the problems of incomplete feature extraction, delayed task allocation and resource conflicts in existing technologies, achieving the dual goals of improving the efficiency of collaborative strategy generation and enhancing the success rate of execution, and providing feasible technical support for intelligent conference decision-making systems.

[0023] For better understanding and implementation, the present invention is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A flowchart of a multi-agent collaboration strategy generation method provided in an embodiment of the present application;

[0025] Figure 2 A schematic diagram of a process for generating a collaborative strategy solution according to an embodiment of the present application;

[0026] Figure 3 A structural diagram of a multi-agent collaboration strategy generation device provided in another embodiment of the present application. DETAILED DESCRIPTION

[0027] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The drawings illustrate preferred embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the disclosure.

[0028] Unless otherwise defined, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which this invention pertains. The terms used in this specification are for the purpose of describing specific embodiments only and are not intended to limit the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0029] In one embodiment, Figure 1 As shown, a method for generating a multi-agent collaboration strategy is provided. This embodiment uses the method applied to a terminal as an example. It is understandable that the method can also be applied to a server, or to a system including a terminal and a server, and implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0030] S1: Perform environmental perception processing on the conference voice and text data input by the user, separate the acoustic spectrum features and text semantic features, extract semantic key points, and generate a set of semantic understanding results.

[0031] Specifically, the system uses complex acoustic signal processing algorithms to pre-process the speech signal, including removing background noise and performing speech enhancement to improve the clarity and recognizability of the speech signal. Subsequently, the system converts the processed speech signal into a spectrogram through fast Fourier transform, and then separates the acoustic spectrum features. Specifically, for text data, the system can use methods such as lexical analysis, syntactic analysis, and semantic role labeling in natural language processing technology to disassemble the text layer by layer and extract text semantic features such as keywords, phrases, and semantic structures in the text.

[0032] On this basis, the system can use deep learning algorithms based on recurrent neural networks or Transformer architectures to conduct in-depth analysis of the separated acoustic spectrum features and text semantic features to mine semantic information. Preferably, the system can extract semantic key points through the attention mechanism. These semantic key points cover core information elements such as important concepts, viewpoints, problem descriptions, and solution suggestions in the meeting content. The system integrates and organizes these semantic key points according to preset semantic association rules and data structures to generate a structured set of semantic understanding results. This set is stored in a specific format, such as JSON or XML, to facilitate calling and parsing by subsequent processing modules.

[0033] S2: Perform task parsing on the semantic understanding result set, identify the essential characteristics of the problem and associate them with the responsibility areas, and generate a set of classified problem points.

[0034] Specifically, the system uses a pre-trained deep learning model, combined with concepts and relationships in the domain knowledge graph, to perform semantic expansion and contextual understanding of semantic key points, uncovering the deeper semantic information underlying these key points. Specifically, by comparing and matching with a large amount of data in the domain knowledge base, the system identifies the essential problem characteristics associated with each semantic key point. The domain knowledge base covers multiple predefined responsibility areas, including but not limited to corporate strategy, emergency command, and intelligent manufacturing. The system accurately matches the keywords, concepts, and semantic patterns in the problem essential characteristics with the responsibility areas in the domain knowledge base. For example, if the identified problem essential characteristics involve product production process optimization or equipment fault diagnosis, the system associates them with the intelligent manufacturing responsibility area; if they involve market trend analysis or competitive strategy formulation, they are associated with the corporate strategy responsibility area. Through a series of complex matching algorithms and inference mechanisms, the system establishes a mapping between the problem essential characteristics and the responsibility areas, providing clear direction and classification for subsequent processing.

[0035] S3: Call the agent capability library to perform resource matching processing on the classified problem point set, evaluate the adaptation relationship between the agent capabilities and the problem requirements, and generate a list of candidate agents. The candidate agent list is used to indicate the set of agents that can handle each problem point.

[0036] Specifically, the agent capability library is a key resource storage unit in the system, containing various types of agents, along with detailed capability descriptions and performance parameters. Each agent in the library has a corresponding unique identifier, capability label, historical task performance records, and other information.

[0037] Specifically, after the set of classified problem points is generated, the system analyzes each problem point in the set of classified problem points one by one, and selects agents with relevant processing capabilities from the agent capability library based on the type, nature and related fields of the problem point. Preferably, the system can use an agent capability matching algorithm based on Euclidean distance or cosine similarity to quantitatively evaluate the adaptation relationship between the agent and the problem point from multiple dimensions such as agent skill matching, resource adequacy, and historical processing performance. Based on the adaptation evaluation results, the system generates a list of candidate agents for each problem point. The list is stored in a structured form and contains a set of candidate agents corresponding to each problem point and a fitness score between the agent and the problem point, providing a set of candidate agents for subsequent dynamic priority evaluation.

[0038] S4: Perform dynamic priority evaluation on the candidate agent list, calculate the weighted urgency score of the problem impact scope and decision time limit, and generate an evaluation agent sequence.

[0039] Specifically, the system analyzes multiple factors such as the areas of responsibility associated with the problem points in the candidate agent list, the business processes and personnel involved, and the possible chain reactions. For example, a problem point may involve the collaborative processes of multiple departments, affecting the work efficiency of many employees and the overall operation of the enterprise. The system calculates an impact range value based on these factors. At the same time, the system evaluates the decision-making time limit for each problem point, and determines a reasonable decision-making time limit by analyzing information such as the nature of the problem, the urgency, and the time requirements of related business processes. Preferably, the system can use a weighted algorithm to comprehensively calculate the impact range and decision-making time limit of the problem point to obtain a weighted urgency score for each problem point. The system sorts the agents in the candidate agent list according to the score and generates an evaluation agent sequence. The sequence is stored in a structured form, and the agents are arranged from high to low according to the urgency score, providing a sorted agent sequence for subsequent priority screening processing.

[0040] S5: Prioritize the evaluation agent sequence and confirm the primary decision information based on the weighted urgency score. The primary decision information is used to indicate the core problem points and the primary agent that solves the core problem points.

[0041] Specifically, the agents in the evaluation agent sequence are sorted by weighted urgency score, and the system analyzes each agent in descending order. During the analysis, the system comprehensively considers multiple factors, including the feasibility of the agent in solving the corresponding problem, resource utilization, and expected solution effect. The system assesses the evaluation agent's technical capabilities to determine whether it is sufficient to cope with the complexity of the problem. The system also assesses whether its resource requirements are within the system's tolerance. Preferably, the system analyzes the agent's performance data in historical tasks to predict its expected effect on solving the problem, including the accuracy and effectiveness of the solution, thereby determining the agents with higher priority.

[0042] The primary agent refers to the agent that is at the top of the ranking and is considered to be the most suitable agent for handling the corresponding core problem point. Specifically, the system determines the primary agent for each core problem point in order from high to low according to the ranking of weighted urgency scores, and clearly records each core problem point and its corresponding primary agent to form the primary decision information. It should be noted that there may be multiple core problem points, and each problem point has its corresponding primary agent. When determining the primary decision information, the system analyzes the urgency and importance of each problem point to ensure that the issues with the greatest impact on the business are handled first. At the same time, the system takes into account the capabilities and resource allocation of the agents to avoid affecting overall efficiency due to resource conflicts. Through comprehensive evaluation, the system can reasonably allocate tasks to ensure that each core problem point can be effectively handled.

[0043] S6: Generate collaborative solutions for primary decision-making information, and generate collaborative strategy solutions through draft construction and multi-agent optimization integration.

[0044] Specifically, the system preliminarily plans and allocates tasks for the primary agent based on the primary agent's task allocation principles, such as prioritizing task relevance and maximizing resource utilization, clarifying the roles and responsibilities of each agent in the solution. Specifically, the system uses agent-interactive optimization algorithms such as multi-agent reinforcement learning or genetic algorithms to optimize and integrate the preliminary draft.

[0045] In the application of multi-agent reinforcement learning algorithm, the system first initializes the policy parameters of each agent, which independently interacts with the environment in a virtual collaborative environment. Each agent performs corresponding actions according to the current environmental state. The system measures the effect of each agent's action through the instant reward mechanism, and the agent updates its policy parameters according to the reward signal obtained. This process usually uses Q-learning or policy gradient method to optimize the policy. The system continuously allows the agent to learn by trial and error in the environment, constantly adjusts the policy to obtain higher cumulative rewards, and the agents enhance learning efficiency through shared experience replay mechanism. Finally, after multiple rounds of training and policy updating, the policy of each agent gradually converges, forming a collaborative decision-making policy combination to optimize the execution effect of the collaborative strategy solution.

[0046] In the application of genetic algorithm, the system first initializes a population containing multiple potential solutions, each individual representing a possible collaborative strategy configuration, composed of encoded parameters. The system evaluates the fitness of each individual to calculate its effectiveness in solving the current collaboration problem. Based on the fitness evaluation results, the system selects individuals with higher fitness as parents, and through crossover operation, exchanges and combines the coding fragments of parent individuals to generate new offspring individuals. At the same time, the system randomly changes some parameter values in the offspring individual's code with a certain probability to increase the diversity of the population. After multiple generations of evolution, including repeated selection, crossover and mutation operations, the fitness of individuals in the population gradually improves, and finally converges to an optimal or approximate optimal solution, thereby optimizing the execution effect of the collaborative strategy.

[0047] In the optimization process, the system simulates the interaction and collaboration between agents, solves potential resource conflicts and logical contradictions, such as adjusting the task time arrangement of different agents to avoid resource contention, coordinating the dependency relationship between different tasks to ensure logical coherence, etc. Through multiple rounds of iterative optimization, the system finally generates a collaborative strategy solution, which is presented in a structured form, including the work content, responsibility range, mutual cooperation method, time node and expected target achievement of each agent, ensuring that the conference discussion content can be effectively transformed into an action plan with actual execution value.

[0048] In summary, the multi-agent collaboration strategy generation method provided by the application can capture semantic key points in complex scenarios such as speech interruption and text ambiguity by separating acoustic spectral features and text semantic features through a dual-modal analysis mechanism in the environment perception processing stage, thereby solving the problem of incomplete feature extraction caused by cross-modal information conflict in traditional methods; in the task analysis stage, based on the deep analysis of the problem nature feature recognition and the responsibility domain association, the dimensionality decomposition and accurate classification of the problem features can be realized, ensuring zero omission of key problem elements; through the resource matching mechanism of dynamic calling of the agent capability library, real-time adaptation evaluation of the agent capability and problem demand can be realized, completely avoiding the waste of professional resources caused by capability mismatch; the dynamic priority evaluation model using weighted emergency scoring can realize the quantitative analysis of the influence range and decision time limit, to solve the pain point of high-priority task allocation delay; based on the priority screening process of scoring and sorting, the accurate matching of core problem points and optimal agents can be achieved, ensuring the timeliness of key problem response; finally, through the multi-agent collaborative optimization framework, double protection of resource conflict detection and logic contradiction resolution can be realized, to generate an executable scheme without execution fault, thereby fundamentally solving the problems of incomplete feature extraction, task allocation delay and resource conflict in the prior art, achieving the dual goals of improving the efficiency of collaboration strategy generation and strengthening the success rate of execution, and providing technical support for intelligent conference decision-making systems.

[0049] In one of the embodiments, the S1 of the multi-agent collaboration strategy generation method provided by the application specifically includes the following steps:

[0050] S11: separating the acoustic features of the conference voice data input by the user, extracting the fundamental frequency envelope and formant distribution features, and generating an acoustic waveform feature set.

[0051] Specifically, the system can use short-time Fourier transform to decompose the speech signal into multiple short-time frames, and perform spectral analysis on each frame. By calculating the fundamental frequency envelope of the speech signal, the system can capture the pitch variation characteristics in the speech, and the fundamental frequency envelope reflects the variation of the fundamental frequency component in the speech signal over time, which is an important indicator for distinguishing different pronunciation and emotional states. At the same time, the system can use cepstrum analysis method to extract formant distribution features from the speech signal. Formants are the main energy concentration areas formed in the vocal tract of the speech signal, and their distribution characteristics can reflect the timbre and vocal tract characteristics of the speaker, which plays an important role in accurately identifying speech content and speaker. The system integrates the extracted fundamental frequency envelope and formant distribution features to generate an acoustic waveform feature set.

[0052] S12: performing semantic structure analysis on the text data input by the user, identifying the entity relationship network and intent expression framework, and generating a text semantic feature set.

[0053] Specifically, the system can use natural language processing technology to deeply analyze the intrinsic semantic structure of text data and extract key semantic elements. Specifically, the system preprocesses the text, including operations such as word segmentation, part-of-speech tagging, and syntactic analysis, to identify the basic language units and grammatical structures in the text. On this basis, the system can use dependency syntactic analysis methods to identify entity relationship networks in the text. The entity relationship network covers various relationships between entities such as names of people, places, and organizations mentioned in the text. By building connection relationships between entities, the system can more comprehensively understand the scenes and situations described in the text.

[0054] The system also uses semantic role labeling technology to identify intent frameworks within text. These frameworks clarify the intent and purpose conveyed in the text, including topics discussed, questions raised, and proposed solutions. Through deep semantic analysis of the text, the system identifies key elements such as the initiator, target, action, and expected outcome of the intent, thereby constructing a complete intent framework. The system integrates the entity relationship network and the intent framework to generate a set of text semantic features.

[0055] S13: Perform multimodal alignment processing on the acoustic waveform feature set and the text semantic feature set, construct a cross-modal association matrix, and generate a semantic understanding result set.

[0056] Specifically, the system first establishes a temporal correspondence between acoustic features and text semantic features. By timestamping the speech and text data, the system matches each acoustic feature element in the speech with the corresponding semantic segment in the text, ensuring temporal consistency between the two. The system then applies a cross-modal feature association algorithm to calculate the similarity between the acoustic features and the text semantic features. This algorithm, based on a deep learning model, captures the underlying correlation patterns between acoustic features and text semantic features, assessing the degree of match between the two by calculating metrics such as cosine similarity or Euclidean distance between the features. Based on the similarity calculation results, the system constructs a cross-modal association matrix, which stores the strength of the association between acoustic features and text semantic features in matrix form. Each element represents the closeness of the association between an acoustic feature element and a text semantic segment. Finally, the system fuses the information in the cross-modal association matrix to generate a set of semantic understanding results. This set integrates the multimodal information of speech and text, providing a comprehensive and accurate semantic foundation for subsequent collaborative strategy generation.

[0057] In one embodiment, S2 of a multi-agent collaboration strategy generation method provided by the present invention specifically includes the following steps:

[0058] S21: Deconstruct the problem elements of the semantic understanding result set, disassemble the problem objectives, constraints and impact dimensions, and generate the problem essential feature vector.

[0059] Specifically, the system can use natural language processing technology to conduct in-depth analysis of semantic units and break down the problem goals, constraints, and impact dimensions. The problem goals are extracted through the intent recognition algorithm to clarify the specific results that the problem aims to achieve; the constraints are identified with the help of the rule extraction module, covering time, resources, quality and other limiting factors; the impact dimensions are determined through the association analysis module to identify the business processes, departments, and systems involved in the problem. For example, for a problem involving project schedule adjustment, the system can identify that the problem goal is "project completion on time", the constraints include "existing resource limitations" and "team members' workload saturation", and the impact dimensions may involve "project cost", "quality control", "customer satisfaction" and other aspects. The system quantifies and vectorizes these elements to generate the problem's essential feature vector, which comprehensively and accurately represents the core characteristics of the problem in numerical form.

[0060] S22: Perform domain mapping on the problem’s essential feature vector, associate it with the responsible department map in the organizational knowledge base, and generate responsibility domain labels.

[0061] Specifically, the organizational knowledge base is an information storage structure that contains detailed information such as the scope of responsibilities, business areas, and workflows of each department within the organization. The responsible department map is a mapping of the responsibility boundaries and collaborative relationships of each department in the form of a graph structure. In this embodiment, the system performs dimensional decomposition on the essential feature vector of the problem, extracts a feature subset related to the organization's business, and then uses text similarity algorithms and semantic matching technology to retrieve the responsible departments that best match these feature subsets in the organizational knowledge base. For example, when the problem involves "marketing strategy adjustment", the system maps it to relevant responsible departments such as the marketing department or brand promotion department through a matching algorithm. The system will comprehensively consider the confidence of multiple matching results and select the matching result with the highest confidence as the final responsibility area label to ensure that each problem can be accurately assigned to the corresponding department.

[0062] S23: Perform joint encoding processing on the problem essence feature vector and the responsibility area label, construct a classification problem point structure, and generate a classification problem point set.

[0063] Specifically, the system fuses the feature vectors of the problem's essence and the labels of the areas of responsibility, and integrates the two into a unified feature representation through a specific encoding algorithm. This joint encoding process aims to combine the technical characteristics and organizational characteristics of the problem to form a feature structure that can comprehensively characterize the problem's attributes and organizational affiliation. By designing a joint encoder, the system encodes the feature vectors of the problem's essence and the labels of the areas of responsibility into vectors of fixed length, and then generates a classified problem point structure through feature splicing or feature fusion. The classified problem point structure contains the core features of the problem and the corresponding organizational responsibility information. The system stores and organizes these structures according to predefined data structures and formats to generate a set of classified problem points. Each element in the set is a complete classified problem point, which contains both a specific feature description of the problem and a clear description of the responsible department for the problem.

[0064] In one embodiment, S3 of a multi-agent collaboration strategy generation method provided by the present invention specifically includes the following steps:

[0065] S31: Perform capability requirement modeling on the set of classified problem points, abstract the core capability dimensions required for problem solving, and generate a capability requirement vector.

[0066] Specifically, the system performs capability requirement modeling on the set of classified problem points. By deeply analyzing the core elements of each problem point, it abstracts the core capability dimensions required to solve the problem and generates a capability requirement vector. In this process, the system decomposes each problem point in the set of classified problem points into multiple dimensions to identify the key components of the problem, including technical requirements, business logic, decision complexity, etc. The system uses predefined capability dimension templates to match the elements of the problem point and identify the core capability dimensions required to solve the problem, such as data analysis capabilities, logical reasoning capabilities, and domain knowledge application capabilities. By quantifying the importance weight of each core capability dimension, the system integrates these capability dimensions into a vector-shaped capability requirement vector, which comprehensively and accurately describes the specific requirements of each problem point at the capability level, providing a precise target reference for subsequent intelligent agent capability matching.

[0067] S32: Call the agent capability library to perform feature retrieval processing, extract the skill matrix and domain expertise parameters, and generate the agent capability vector.

[0068] Specifically, the agent capability library is a database that integrates information on the capabilities of various agents, including the skill matrix and domain expertise parameters of each agent. The skill matrix describes in detail the performance level of the agent in different skill dimensions, such as the ability level in data processing, pattern recognition, decision support, etc. The domain expertise parameters reflect the degree of expertise of the agent in a specific business field, such as the depth of knowledge and application experience in industries such as finance, healthcare, and manufacturing. The system uses a retrieval algorithm to query the agents related to the classification problem points in the agent capability library and extract their skill matrix and domain expertise parameters. Through specific encoding and normalization processing, these features are integrated into an agent capability vector, which comprehensively represents the ability characteristics of the agent in numerical form.

[0069] S33: Perform fitness calculation on the capability requirement vector and the agent capability vector, calculate the cosine similarity measurement between the problem requirement characteristics and the agent capability characteristics, and generate a list of candidate agents.

[0070] Specifically, the system can use the cosine similarity measurement algorithm to calculate the similarity between the problem requirement characteristics and the agent capability characteristics. The system first standardizes the capability requirement vector and the agent capability vector to eliminate dimensional differences. Then, the cosine similarity value between the two vectors is calculated by vector dot product and modulus length. The value range of cosine similarity is between -1 and 1, and the closer the value is to 1, the higher the matching degree. For each problem point, the system calculates the cosine similarity of multiple agents that match it, and sorts them according to the similarity value to generate a list of candidate agents. The candidate agent list is stored in a structured form, including agent identification, similarity score, and capability matching details.

[0071] In one embodiment, S4 of a multi-agent collaboration strategy generation method provided by the present invention specifically includes the following steps:

[0072] S41: Quantify the impact range of the corresponding problems in the candidate agent list, calculate the hierarchical penetration depth of the problem in the organizational structure and the proportion of affected departments, and generate an impact range coefficient. The impact range coefficient is used to indicate the degree of spread of the problem.

[0073] Specifically, the system extracts detailed information on each level and department from the organizational chart, and establishes a departmental hierarchical relationship diagram and an inter-departmental collaboration network model. For each problem point, the system identifies the initial set of departments that may be affected by the problem by analyzing the nature of the problem, the business processes involved, and the data flows. Then, the system uses a depth-first search or breadth-first search algorithm to simulate the problem diffusion path in the organizational chart, and calculates the potential path length of the problem from the initial department to the upper and lower levels, that is, the hierarchical penetration depth. At the same time, the system determines the set of affected departments by analyzing the correlation between the problem and the business of each department, combining historical data and predefined impact rules, and calculates its proportion in the total number of departments in the organization. The system weightedly integrates the hierarchical penetration depth and the proportion of affected departments to generate an impact range coefficient.

[0074] S42: Perform time limit feature extraction processing on the problem description data in the semantic understanding result set, parse the time-sensitive keywords in the problem context and calculate the remaining response time window to generate decision time limit parameters.

[0075] Specifically, the system leverages named entity recognition and temporal expression parsing algorithms within natural language processing to identify all key time-related information from the problem description text, including deadlines, scheduled event times, and periodic task times. Simultaneously, the system analyzes the typical processing cycle and priority requirements of the problem, combining the organization's internal business process specifications and historical data, to determine the expected response time window. The system compares the current time with the identified time-sensitive keywords, calculates the length of the remaining response time window, and normalizes the remaining response time window, taking into account the urgency of the time and the rationality of the processing cycle, to generate decision-making time limit parameters.

[0076] S43: Dynamically weight the influence range coefficient and decision time limit parameter, calculate the weighted emergency score of the problem-agent combination through adaptive weight adjustment, and generate an evaluation agent sequence.

[0077] Specifically, the system dynamically adjusts the weights of the impact coefficient and decision time limit parameters based on the organization's historical data and current business conditions. In certain business environments, the spread of the problem may have a more significant impact on the organization. In this case, the system will increase the weight of the impact coefficient; in time-sensitive scenarios, the system will increase the weight of the decision time limit parameter. The system uses machine learning algorithms, such as linear regression or neural networks, to learn from the organization's historical decision-making cases and establish a weight adjustment model. When calculating the weighted urgency score, the system multiplies the impact coefficient and decision time limit parameter of each problem-agent combination by its corresponding dynamic weight, and then adds them together to obtain the weighted urgency score. The system sorts the agents in the candidate agent list according to the weighted urgency score to generate an evaluation agent sequence.

[0078] In one embodiment, S5 of a multi-agent collaboration strategy generation method provided by the present invention specifically includes the following steps:

[0079] S51: Sorting the evaluation agent sequence by scores, arranging the problem-agent combinations in descending order of weighted urgency scores, and generating a sorted priority queue.

[0080] Specifically, the system reads the weighted urgency score of each agent in the evaluation agent sequence. This score combines the impact range coefficient and decision time limit parameters of the problem, reflecting the urgency of the problem and the priority of the agent in handling it. The system uses sorting algorithms such as quick sort or heap sort to sort the problem-agent combinations from high to low according to the weighted urgency score. During the sorting process, the system ensures that the score of each combination is accurately compared and positioned to determine its position in the queue. The sorted priority queue not only retains the detailed information of each problem-agent combination, but also intuitively displays the priority of processing through the sequential relationship.

[0081] S52: Based on the performance records in the historical task database, the priority queue is verified, the matching degree between the agent's historical task success rate and the current problem characteristics is evaluated, and a verification report is generated.

[0082] Specifically, the system accesses the historical task database and retrieves the performance records of each agent when it handled similar problems in the past, including key indicators such as task completion rate, solution quality, and resource utilization. Through a feature matching algorithm, the system compares the characteristics of the current problem with the characteristics of problems that the agent has historically handled, and identifies the performance of the agent when handling problems with similar characteristics. The system uses statistical analysis methods to calculate the degree of match between the agent's historical task success rate and the current problem characteristics, and generates quantitative capability verification indicators. These indicators reflect the ability and reliability of the agent in handling similar problems in actual applications. The system integrates the capability verification indicators of all agents into a capability verification report, which records in detail the evaluation results of each agent, including its strengths and potential shortcomings.

[0083] S53: Perform decision optimization processing on the capability verification report, select the question-agent combination with the best comprehensive score, and confirm the primary decision information.

[0084] Specifically, the system uses a multi-dimensional comprehensive scoring model to weightedly integrate the agent's historical task success rate, its matching degree with the current problem, and its weighted urgency score to calculate a comprehensive score for each combination. The system selects the optimal problem-agent combination based on the comprehensive scores from high to low, and generates primary decision information. The weight coefficient is determined based on the organization's historical decision-making experience and current business priorities. Usually, the weighted urgency score is given a higher weight to ensure that urgent issues are given priority. The system sorts the problem-agent combinations according to the comprehensive score and selects the combination with the highest comprehensive score as the primary decision information. The primary decision information not only contains a detailed description of the core problem points, but also clearly specifies the primary agent responsible for handling these problem points.

[0085] In one embodiment, Figure 2 As shown, S6 of the multi-agent collaboration strategy generation method provided by the present invention specifically includes the following steps:

[0086] S61: Perform domain template call processing on the primary decision information, activate the solution framework of the corresponding problem type in the knowledge base, and generate an initial strategy draft for the core problem points.

[0087] Specifically, the domain templates in the knowledge base cover a variety of common business scenarios and problem types. Each template has been carefully designed and verified, and includes the basic process of problem solving, key task nodes and recommended action steps. The system uses a feature matching algorithm to compare the primary decision-making information with the templates in the knowledge base one by one to identify the solution framework that best matches the current problem type. During the matching process, the system analyzes the keywords, context and feature parameters of the problem to ensure that the selected template can best meet the needs of the actual problem. Once the match is successful, the system will embed the specific information of the core problem point into the solution framework and generate an initial strategy draft through parameterized filling. The draft includes a background description of the problem, a preliminary solution, the expected execution steps and a preliminary estimate of resource requirements.

[0088] S62: Perform multi-agent collaborative annotation processing on the initial strategy draft, collect domain expertise suggestions from non-primary agents, detect resource allocation conflicts and logical contradictions, and generate an optimized strategy framework.

[0089] Specifically, the system sends the initial strategy draft to the non-primary agents participating in the collaboration. These agents each have different domain expertise and knowledge reserves. Each agent will now analyze the draft and make improvement suggestions based on its own capabilities and experience. These suggestions may include improving the details of the solution, optimizing the order of task execution, and warning of potential risks. The system will integrate and analyze these suggestions to determine which improvement suggestions have higher value and feasibility. At the same time, the system will activate the resource conflict detection mechanism to analyze the resource allocation involved in the draft to check whether there is any over-allocation of resources or resource competition.

[0090] The system also uses logic verification algorithms to check the draft's task flow and decision logic, ensuring there are no contradictions or inconsistencies between steps. After integrating the suggestions of all agents and any conflicts and inconsistencies detected, the system modifies and supplements the initial draft strategy to generate an optimized strategy framework. This framework builds on the initial solution and fully incorporates the expertise of multiple agents, correcting resource and logic issues and improving the strategy's feasibility and effectiveness.

[0091] S63: Execute executable links on the optimization strategy framework, integrate the time window allocation of the task execution sequence and the binding scheme of key resources, and generate the final collaborative strategy solution. The collaborative strategy solution is used to indicate a multi-agent collaborative action plan that can be executed immediately.

[0092] Specifically, the system schedules the task sequence within the optimization strategy framework. By analyzing task priorities, dependencies, and estimated execution times, the system employs a scheduling algorithm to allocate appropriate time windows for each task. This time window allocation not only considers the urgency of the task but also ensures sufficient buffer time between tasks to accommodate unexpected situations. Furthermore, the system designs a binding scheme for key resources to ensure that each task receives the necessary resource support, including human resources, equipment resources, and data resources. The system uses a resource allocation algorithm to optimize resource utilization and avoid resource waste or overallocation. To verify the rationality of the time window allocation and resource binding scheme, the system simulates the task execution process and adjusts and optimizes the scheme based on the simulation results. The resulting collaborative strategy solution includes a detailed execution plan that clearly defines each task's execution time, responsible agent, required resources, and expected outcomes.

[0093] The multi-agent collaborative strategy generation method provided above processes domain templates for primary decision information. Based on the pre-stored solution framework in the knowledge base, the system can quickly generate an initial strategy draft for core problem points, providing a basic framework for subsequent optimization, thereby achieving efficient transformation from complex meeting content to preliminary action plans. Then, using multi-agent collaborative annotation processing, the system can not only collect professional advice from non-primary agents and enrich the content of the strategy draft, but also detect and resolve resource allocation conflicts and logical contradictions, effectively avoiding the common resource contention and execution fault problems in existing technologies, and improving the rationality and feasibility of the strategy. Finally, through executable link construction processing, the system integrates the time window allocation of the task execution sequence and the binding scheme of key resources to generate the final collaborative strategy solution, which can be directly used to guide multi-agent collaborative action plans, ensuring the consistency and efficiency from strategy generation to execution, thereby achieving the purpose of improving organizational decision-making efficiency and ensuring the efficient execution of complex tasks.

[0094] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0095] Based on the same inventive concept, the present application also provides a multi-agent collaborative strategy generation device for implementing the multi-agent collaborative strategy generation method involved above. The implementation solution provided by the device is similar to the implementation solution described in the above method, so the specific limitations of one or more multi-agent collaborative strategy generation device embodiments provided below can be found in the above-mentioned limitations of the multi-agent collaborative strategy generation method, and will not be repeated here.

[0096] Preferably, if Figure 3 As shown, the present invention provides a multi-agent collaboration strategy generation device 700, which is configured with the following modules:

[0097] The environment perception processing module 710 is used to perform environment perception processing on the conference voice and text data input by the user, separate the acoustic spectrum features and text semantic features, extract semantic key points, and generate a semantic understanding result set;

[0098] The task analysis module 720 is used to perform task analysis on the semantic understanding result set, identify the essential characteristics of the problem, associate them with the responsibility areas, and generate a set of classified problem points;

[0099] The agent resource matching module 730 is used to call the agent capability library to perform resource matching processing on the classified problem point set, evaluate the adaptation relationship between the agent capabilities and the problem requirements, and generate a candidate agent list. The candidate agent list is used to indicate the set of agents that can handle each problem point;

[0100] Dynamic priority evaluation module 740, for performing dynamic priority evaluation on the candidate agent list, calculating a weighted urgency score of the problem impact scope and decision time limit, and generating an evaluation agent sequence;

[0101] Priority screening module 750 is used to perform priority screening on the evaluation agent sequence and confirm the primary decision information based on the weighted urgency score sorting. The primary decision information is used to indicate the core problem point and the primary agent to solve the core problem point;

[0102] The collaborative solution generation module 760 is used to generate collaborative solutions for primary decision-making information, and generate collaborative strategy solutions through draft construction and multi-agent optimization integration.

[0103] Preferably, the environment perception processing module 710 provided in the embodiment of the present application is configured with the following units:

[0104] The acoustic feature extraction unit is used to perform acoustic feature separation on the conference voice data input by the user, extract the fundamental frequency envelope and formant distribution features, and generate an acoustic waveform feature set;

[0105] The text semantic parsing unit is used to parse the semantic structure of the text data input by the user, identify the entity relationship network and the intention expression framework, and generate a set of text semantic features;

[0106] The multimodal alignment processing unit is used to perform multimodal alignment processing on the acoustic waveform feature set and the text semantic feature set, construct a cross-modal association matrix, and generate a semantic understanding result set.

[0107] Preferably, the task parsing module 720 provided in the embodiment of the present application is configured with the following units:

[0108] The problem element deconstruction unit is used to deconstruct the problem elements of the semantic understanding result set, dismantle the problem objectives, constraints and impact dimensions, and generate the problem essential feature vector;

[0109] The responsibility domain mapping unit is used to perform domain mapping processing on the problem essence feature vector, associate it with the responsibility department map in the organizational knowledge base, and generate responsibility domain labels;

[0110] The problem point encoding unit is used to jointly encode the problem essence feature vector and the responsibility area label, construct a classified problem point structure, and generate a classified problem point set.

[0111] Preferably, the agent resource matching module 730 provided in the embodiment of the present application is configured with the following units:

[0112] The capability requirement modeling unit is used to perform capability requirement modeling on the set of classified problem points, abstract the core capability dimensions required for problem solving, and generate a capability requirement vector;

[0113] The agent feature extraction unit is used to call the agent capability library to perform feature retrieval processing, extract the skill matrix and domain expertise parameters, and generate the agent capability vector;

[0114] The fitness calculation unit is used to perform fitness calculation on the capability requirement vector and the agent capability vector, calculate the cosine similarity measurement between the problem requirement characteristics and the agent capability characteristics, and generate a list of candidate agents.

[0115] Preferably, the dynamic priority evaluation module 740 provided in the embodiment of the present application is configured with the following units:

[0116] The impact quantification unit is used to quantify the impact of the corresponding issues in the candidate agent list, calculate the depth of the problem's hierarchical penetration in the organizational structure and the proportion of affected departments, and generate an impact coefficient that indicates the degree of spread of the problem;

[0117] The decision time limit extraction unit is used to extract time limit features from the problem description data in the semantic understanding result set, parse the time-sensitive keywords in the problem context, calculate the remaining response time window, and generate decision time limit parameters;

[0118] The dynamic weighted evaluation unit is used to dynamically weight the influence range coefficient and decision time limit parameters, calculate the weighted emergency score of the problem-agent combination through adaptive weight adjustment, and generate an evaluation agent sequence.

[0119] Preferably, the priority screening module 750 provided in the embodiment of the present application is configured with the following units:

[0120] A priority sorting unit is used to sort the evaluation agent sequence by scores, arranging the problem-agent combinations in descending order of weighted urgency scores to generate a sorted priority queue;

[0121] The capability verification evaluation unit is configured to perform capability verification processing on the priority queue based on the performance records in the historical task database, evaluate the matching degree between the historical task success rate of the agent and the current problem characteristics, and generate a capability verification report;

[0122] The decision optimization confirmation unit is configured to perform decision optimization processing on the capability verification report, select a problem-agent combination with the optimal comprehensive score, and confirm the primary decision information.

[0123] Preferably, the collaborative scheme generation module 760 provided by the embodiment of the present application is configured with the following units:

[0124] The strategy draft generation unit is configured to perform field template calling processing on the primary decision information, activate a solution framework corresponding to the problem type in the knowledge base, and generate an initial strategy draft for the core problem point.

[0125] The multi-agent collaborative optimization unit is configured to perform multi-agent collaborative labeling processing on the initial strategy draft, collect field expertise suggestions of non-primary agents and detect resource allocation conflicts and logical contradictions, and generate an optimized strategy framework.

[0126] The executable link construction unit is configured to perform executable link construction processing on the optimized strategy framework, integrate time window allocation of the task execution sequence and binding schemes of key resources, and generate a final collaborative strategy solution indicating a multi-agent collaborative action plan that can be immediately executed.

[0127] In one embodiment, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the multi-agent collaborative strategy generation method described above when executing the computer program.

[0128] In one embodiment, the present application further provides a computer readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the multi-agent collaborative strategy generation method described above.

[0129] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example" or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples without contradiction.

[0130] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to the partial description of the method embodiments. The device embodiments described above are merely illustrative, wherein the components described as separate parts may or may not be physically separated, and the parts displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the disclosed solution. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0131] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any person skilled in the art can easily conceive of various modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should be included within the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for generating a multi-agent collaboration strategy, characterized in that: The following steps are involved: S1: Performs environmental perception processing on the user-input conference voice and text data, separates acoustic spectrum features and text semantic features, extracts semantic key points, and generates a set of semantic understanding results; S2: Perform task analysis on the semantic understanding result set, identify the essential characteristics of the problem and associate them with the responsibility areas, and generate a set of classified problem points; S3: Calling the agent capability library to perform resource matching processing on the classified problem point set, evaluating the adaptation relationship between the agent capabilities and the problem requirements, and generating a candidate agent list, wherein the candidate agent list is used to indicate the set of agents that can handle each problem point; S4: Performing dynamic priority evaluation on the candidate agent list, calculating a weighted urgency score of the problem impact scope and decision time limit, and generating an evaluation agent sequence; S5: Prioritize the evaluation agent sequence and confirm primary decision information based on weighted urgency score sorting. The primary decision information is used to indicate the core problem point and the primary agent that solves the core problem point; S6: Perform collaborative solution generation processing on the primary decision information, and generate collaborative strategy solutions through draft construction and multi-agent optimization integration.

2. The method according to claim 1, characterized in that Said S1 comprises: S11: Perform acoustic feature separation on the conference voice data input by the user, extract the fundamental frequency envelope and formant distribution features, and generate an acoustic waveform feature set; S12: Analyze the semantic structure of the text data input by the user, identify the entity relationship network and intent expression framework, and generate a set of text semantic features; S13: Perform multimodal alignment processing on the acoustic waveform feature set and the text semantic feature set, construct a cross-modal association matrix, and generate a semantic understanding result set.

3. The method according to claim 1, characterized in that The S2 includes: S21: Deconstructing the problem elements on the semantic understanding result set, breaking down the problem objectives, constraints, and impact dimensions, and generating a problem essential feature vector; S22: Performing domain mapping processing on the essential feature vector of the problem, associating it with the responsible department map in the organizational knowledge base, and generating a responsible domain label; S23: Perform joint encoding processing on the problem essential feature vector and the responsibility area label, construct a classification problem point structure, and generate a classification problem point set.

4. The method according to claim 1, wherein The S3 includes: S31: Perform capability requirement modeling on the set of classified problem points, abstract the core capability dimensions required for problem solving, and generate a capability requirement vector; S32: Call the agent capability library to perform feature retrieval processing, extract the skill matrix and domain expertise parameters, and generate the agent capability vector; S33: Performing a degree of compatibility calculation on the capability requirement vector and the agent capability vector, calculating a cosine similarity measure between the problem requirement characteristics and the agent capability characteristics, and generating a candidate agent list.

5. The method according to claim 1, wherein The S4 includes: S41: Quantifying the impact range of the corresponding issues in the candidate agent list, calculating the hierarchical penetration depth of the issue in the organizational structure and the proportion of affected departments, and generating an impact range coefficient, which is used to indicate the degree of spread of the issue; S42: performing time limit feature extraction processing on the question description data in the semantic understanding result set, parsing the time-sensitive keywords in the question context and calculating the remaining response time window to generate a decision time limit parameter; S43: Dynamically weight the influence range coefficient and the decision time limit parameter, calculate the weighted urgency score of the problem-agent combination through adaptive weight adjustment, and generate an evaluation agent sequence.

6. The method according to claim 1, characterized in that The S5 includes: S51: Sorting the evaluation agent sequence by scores, arranging the problem-agent combinations in descending order of weighted urgency scores, and generating a sorted priority queue; S52: Perform capability verification on the priority queue based on the performance records in the historical task database, evaluate the matching degree between the agent's historical task success rate and the current problem characteristics, and generate a capability verification report; S53: Perform decision optimization processing on the capability verification report, select the question-agent combination with the best comprehensive score, and confirm the primary decision information.

7. The method according to any one of claims 1 to 6, characterized in that The S6 includes: S61: Performing domain template call processing on the primary decision information, activating the solution framework of the corresponding problem type in the knowledge base, and generating an initial strategy draft for the core problem point; S62: Perform multi-agent collaborative annotation processing on the initial strategy draft, collect domain expertise suggestions from non-primary agents, detect resource allocation conflicts and logical contradictions, and generate an optimization strategy framework; S63: Perform executable link construction processing on the optimization strategy framework, integrate the time window allocation of the task execution sequence and the binding scheme of key resources, and generate a final collaborative strategy solution, which is used to indicate a multi-agent collaborative action plan that can be executed immediately.

8. A multi-agent collaboration strategy generation device, characterized in that: The device comprises: The environment perception processing module is used to perform environment perception processing on the conference voice and text data input by the user, separate the acoustic spectrum features and text semantic features, extract semantic key points, and generate a set of semantic understanding results; A task analysis module is used to perform task analysis on the semantic understanding result set, identify the essential characteristics of the problem, associate them with the responsibility areas, and generate a set of classified problem points; An agent resource matching module is used to call the agent capability library to perform resource matching processing on the set of classified problem points, evaluate the adaptation relationship between the agent capabilities and the problem requirements, and generate a list of candidate agents. The candidate agent list is used to indicate the set of agents that can handle each problem point; A dynamic priority evaluation module is used to perform dynamic priority evaluation on the candidate agent list, calculate the weighted urgency score of the problem impact scope and decision time limit, and generate an evaluation agent sequence; A priority screening module is used to perform priority screening on the evaluation agent sequence and confirm the primary decision information based on the weighted urgency score sorting, wherein the primary decision information is used to indicate the core problem points and the primary agent for solving the core problem points; The collaborative solution generation module is used to generate collaborative solutions for the primary decision-making information, and generate collaborative strategy solutions through draft construction and multi-agent optimization integration.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images