Open source intelligence analysis method and system based on dynamic role multi-agent

Abstract

The invention provides an open source intelligence analysis method and system based on a dynamic role multi-agent, and relates to the technical field of open source intelligence analysis. The method comprises the following steps: firstly, defining roles required for solving a user open source intelligence analysis task request, and constructing a collaborative topological graph among the roles; instantiating a role agent with a specific field view angle according to the role definition, and constructing a corresponding communication topological graph; a supervision agent is embedded in the communication topology to audit the output of the role agent; and finally, scheduling the role agent to execute an analysis task according to the communication topological graph, performing information circulation and error correction in combination with a check result and feedback information of the supervision agent, and finally outputting an answer to a user request. According to the method, a complex intelligence analysis task is disassembled into sub-tasks completed by a plurality of agents, and the agents are organized to perform reasoning, interaction and result integration according to a collaborative topological graph, so that accurate response to a user request is realized.

Description

Technical Field

[0001] This invention relates to the field of open-source intelligence analysis technology, and in particular to an open-source intelligence analysis method and system based on dynamic role multi-agent. Background Technology

[0002] Open Source Intelligence (OSINT) is an advanced cognitive synthesis task. Its core lies in extracting and generating insights or answers that meet the user's requirements by performing multi-dimensional in-depth analysis, correlation reasoning, and structured synthesis on a set of pre-provided reference materials that may contain complex, fragmented, or even contradictory information, in response to a specific user request.

[0003] Unlike traditional methods of extracting single information or listing facts, open-source intelligence analysis is essentially a deep semantic reconstruction and verification process that transcends fragmented data representation to uncover the underlying logic. In practical applications, the reference materials to be analyzed are often unstructured and may be riddled with fragmented information and noise. Furthermore, descriptions of the same event or entity from different sources may contain logical conflicts or factual contradictions (e.g., different news sources provide inconsistent timelines for the same breaking event). This presents a significant cognitive challenge to the analysis process; relying solely on linear interpretation from a single perspective is often insufficient. Instead, a multi-dimensional, multi-domain "expert team" approach is necessary for collaborative judgment. It requires the analyst to simulate the perspectives of experts from diverse professional backgrounds (such as geopolitics, economics and finance, and psychological profiling) to conduct cross-validation and dialectical deduction of complex information from multiple angles. By merging and integrating different viewpoints, cognitive biases and blind spots from a single perspective can be eliminated, thereby constructing objective, rigorous, and logically consistent intelligence conclusions.

[0004] Before the maturity of deep learning and Large Language Model (LLM) technologies, open-source intelligence analysis primarily adopted a "human analysis as the main method, supplemented by tools" approach. Specifically, traditional automated auxiliary tools typically relied on regular expressions, expert rule systems, or statistical machine learning models. While these technologies have some application value in specific task scenarios (such as keyword filtering, simple named entity recognition, or short text sentiment classification), they essentially remain at the level of surface-level symbol matching and statistical analysis. Limited by model architecture and parameter scale, traditional Natural Language Processing (NLP) techniques lack the ability to understand deep semantics, struggle to capture long-range dependencies in long texts, and are unable to perform complex logical reasoning and causal chain construction across multiple documents. Faced with metaphors, puns, or implicit information requiring background knowledge commonly found in open-source intelligence, traditional tools are often powerless and unable to undertake the core intelligence cognition tasks.

[0005] In recent years, with the gradual maturation of large language models based on the Transformer architecture, their powerful general semantic understanding capabilities, few-shot learning capabilities, and Chain of Thought (CoT) reasoning capabilities have brought revolutionary opportunities for breakthroughs in open-source intelligence analysis. Currently, there are attempts in the industry to utilize large language models for open-source intelligence analysis. These methods typically leverage the powerful reading comprehension and generation capabilities of large language models to summarize information, answer questions, or perform preliminary logical analysis.

[0006] However, while existing methods based on large language models have improved the efficiency and automation of information processing to some extent, they still face severe technical challenges when dealing with highly complex and dynamically changing intelligence scenarios. Specifically, most existing technologies use a unidirectional linear reasoning architecture, lacking effective self-correction and multi-dimensional perspective mechanisms. This makes it impossible to effectively suppress the hallucination phenomenon inherent in large language models in intelligence analysis, and the model is prone to generating seemingly plausible false facts with a highly confident tone. Once there is a factual bias or logical fallacy at the beginning of the reasoning chain, the error will be amplified cascaded along the linear process, ultimately leading to the generated analysis report being based on false premises. In addition, limited by the single perspective of a single intelligent agent, existing methods struggle to balance "breadth-based search" across the entire network with "depth mining" of specific clues, easily getting bogged down in localized information and failing to construct rigorous intelligence conclusions from a multidisciplinary and multi-dimensional objective perspective. Summary of the Invention

[0007] The technical problem to be solved by the present invention is to address the shortcomings of the prior art by providing an open-source intelligence analysis method and system based on dynamic role multi-agent, so as to realize the analysis of open-source intelligence.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0009] On the one hand, this invention provides an open-source intelligence analysis method based on dynamic role multi-agent systems, including:

[0010] Define the roles required to resolve user requests for open-source intelligence analysis tasks and construct a collaboration topology graph among the roles;

[0011] Instantiate role agents with specific domain perspectives based on role definitions and construct corresponding communication topology graphs;

[0012] An oversight agent is embedded in the communication topology to audit the output of the role agent, and output audit results and feedback information.

[0013] The scheduling agent performs analysis tasks according to the communication topology diagram, and combines the verification results and feedback information of the supervisory agent to carry out information flow and error correction, and finally produces a response to the user's request.

[0014] Furthermore, the specific method for defining the roles required to resolve user open-source intelligence analysis task requests and constructing a collaborative topology graph among these roles is as follows:

[0015] A planner based on a large language model receives open-source intelligence analysis task requests from users. and a collection of reference materials and information. As input, through reasoning The set of role definitions required to resolve user requests is then provided. Collaboration topology between roles ,Right now .

[0016] Furthermore, the set of role definitions Each element in the set Each tuple is a character name. and role description Role Responsibilities Description The information dimensions and specific reasoning tasks that this role needs to focus on during the analysis process are defined;

[0017] The collaboration topology diagram between roles A directed acyclic graph describing the collaborative relationships between roles is constructed for a planner based on a large language model, denoted as . , Let be the set of nodes in the graph. Let be the set of edges of the graph; let be the set of nodes of the graph. Starting from a preset node Termination node It consists of a set of n execution nodes, where For the role The corresponding execution node; the set of edges in the graph. This defines the flow path of information between nodes, and each directed edge in the edge set... Represents a node The output data will be used as nodes The input context.

[0018] Furthermore, the specific method for instantiating a role agent with a specific domain perspective based on the role definition and constructing the corresponding communication topology graph is as follows:

[0019] For the set of role definitions Each role definition Construct corresponding role-based intelligent agent instances based on large language models one by one, denoted as... Specifically:

[0020] Character Name and role description Fill into the character's intelligent agent prompt template In the process, generate system prompts specific to this intelligent agent. By loading this prompt word, Initialized as a having by Reasoning nodes for defined domain-specific thinking methods;

[0021] After instantiating all the role agents, based on the collaboration topology graph between the roles... Establish the communication topology between intelligent agents, specifically as follows:

[0022] For the collaboration topology diagram between roles Each directed edge in Configure message routing rules so that nodes Corresponding intelligent agent The output is used as a node. Corresponding intelligent agent Input.

[0023] Furthermore, the specific method for embedding a supervisory agent in the communication topology to verify the output of the role agent is as follows: For the communication topology graph excluding the starting node... and termination node Any execution node other than those that undertake actual analysis tasks Instantiate a corresponding supervisory agent And embed it into the role-based intelligent agent node. In the execution process; supervising the intelligent agent Verification rules have been configured. and the maximum corrected retry threshold Receive the role's intelligent agent The output, combined with the character's intelligent agent Received input and its role responsibilities Verify that the output does not contain factual errors or indicate a failure to fulfill responsibilities.

[0024] Furthermore, the scheduling agent performs analysis tasks according to the communication topology diagram, and combines the verification results and feedback information from the supervisory agent to perform information flow and error correction, ultimately producing a response to the user's request, specifically including:

[0025] Step 1: Initialize the input information and the queue of tasks to be processed required for topology scheduling;

[0026] From the starting node Begin, prepare the initial input information. And update all nodes related to the starting node. Find the in-degree of connected nodes; then, find all nodes with a current in-degree of 0 and add them to the task queue; for multiple tasks in the queue, use multi-threading to compute in parallel.

[0027] Step 2: The active agents in the task queue perform reasoning based on the input context and output intermediate analysis results;

[0028] When the role agent is activated from the queue of pending tasks. At that time, first from The upstream node receives the output to form Input context ; Based on its input context Perform inference and output intermediate analysis results. ,in, Represents the reasoning operations of a large language model;

[0029] Step 3: The supervising agent reviews the intermediate analysis results output by the role agent and provides the review results and feedback.

[0030] Supervisory agent receives role agent Received input context Intermediate analysis results output and the role description of the intelligent agent. As input, perform the audit operation. And ultimately produce two outputs. ,in For the review results, This indicates that the review has been approved. This indicates that the review failed. For feedback;

[0031] Step 4: Based on the verification results and feedback information from the supervising agent, perform information flow and error correction;

[0032] If the audit result of the supervisory agent is This indicates that the current character's output meets the requirements, thus identifying the character's intelligent agent. All downstream nodes intermediate analysis results Pass it to these nodes and update the nodes. The dependency state; when a node Add it to the task queue when all its prerequisites are satisfied.

[0033] If the audit result of the supervisory agent is This will prevent the intermediate analysis results of the blocking agent from flowing downstream, and will include feedback containing specific error information. Return to the character's intelligent agent ; After receiving feedback, it will be combined The description in the text is then used for further reasoning to generate a revised output. This correction process constitutes an iterative loop until the output passes the review of the supervising agent or reaches the preset maximum correction retry threshold of the supervising agent. If the maximum correction retry threshold is reached... If the error remains unresolved, a prompt will appear asking whether manual intervention is required or if the analysis process should be stopped.

[0034] Step 5: The termination node summarizes all collected outputs and feeds them back to the user as the final analysis result;

[0035] When any role intelligent agent node The process is completed and passes supervision, and its downstream node is the termination node. At that time, the final analysis results generated by this node are aggregated to the termination node. When all nodes are connected to the termination node After all the preceding nodes have completed their tasks, the terminating node... All collected outputs are summarized and fed back to the user as the final analysis result.

[0036] On the other hand, the present invention also provides an open-source intelligence analysis system based on dynamic role multi-agent, including a role definition module, a role agent instantiation module, a supervisory agent embedding module, and a task execution module;

[0037] The role definition module defines the roles required to resolve user requests for open-source intelligence analysis tasks and constructs a collaborative topology graph between roles.

[0038] The role agent instantiation module instantiates role agents with specific domain perspectives based on the role definition and constructs the corresponding communication topology graph.

[0039] The supervisory agent embedding module embeds a supervisory agent into the communication topology to audit the output of the role agent, and outputs audit results and feedback information;

[0040] The task execution module enables the role-based intelligent agent to perform tasks according to the communication topology diagram, and combines the verification results and feedback information of the supervisory intelligent agent to carry out information flow, ultimately producing a response to the user's request.

[0041] Thirdly, this application proposes an electronic device comprising: one or more processors, and a memory for storing instructions that, when executed by the one or more processors, cause the one or more processors to execute the open-source intelligence analysis method based on dynamic role multi-agent.

[0042] Fourthly, this application proposes a computer-readable storage medium storing executable instructions that, when executed, cause a processor to perform the open-source intelligence analysis method based on dynamic role multi-agent.

[0043] Fifthly, this application proposes a computer program product, including a computer program or instructions that, when executed by a processor, implement the aforementioned open-source intelligence analysis method based on dynamic role multi-agent.

[0044] The beneficial effects of adopting the above technical solution are as follows: The open-source intelligence analysis method and system based on dynamic role multi-agent provided by this invention breaks through the limitation of the fixed single perspective in traditional intelligence analysis based on a single large language model, and realizes dynamic adaptation of analysis dimensions. By dynamically generating customized "expert role teams" for each user request through a planner, the multi-agent system can simulate the collaborative thinking of a team of human analysts. This mechanism breaks through the bottleneck of cognitive breadth and depth of a single large language model, thereby avoiding cognitive bias caused by a single perspective.

[0045] Furthermore, this invention addresses the issue of illusions arising in long-chain reasoning in large language models by establishing a process-level real-time supervision system, effectively suppressing the cascading effect of errors. This invention achieves real-time verification during the reasoning process by introducing a supervisory agent in the intermediate links of the communication topology. This mechanism can intercept and correct information that is inconsistent with the facts or has logical conflicts in real time during the reasoning chain, ensuring the correctness of the overall intermediate reasoning results, thereby significantly improving the accuracy and credibility of the final result.

[0046] Meanwhile, by leveraging a structured multi-agent collaborative architecture in the form of a directed acyclic graph, this invention enhances the ability to analyze unstructured and contradictory information. By breaking down complex cognitive tasks into structured multi-agent collaborative processes, different roles can perform dialectical deduction and cross-validation within the collaborative graph, thereby enabling the identification of truth from falsehood among a vast array of reference materials and solving the cross-document reasoning problem. Attached Figure Description

[0047] Figure 1 A flowchart of the open-source intelligence analysis method based on dynamic role multi-agent provided in Embodiment 1 of the present invention;

[0048] Figure 2 This is a schematic diagram illustrating the operational logic of the open-source intelligence analysis method based on dynamic role multi-agent provided in Embodiment 1 of the present invention. Detailed Implementation

[0049] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0050] Example 1:

[0051] In this embodiment, an open-source intelligence analysis method based on dynamic role multi-agent is used to dynamically construct a role-based intelligent agent and a multi-agent collaborative network. This decomposes complex intelligence analysis tasks into sub-tasks to be completed by several intelligent agents. Based on the generated collaborative topology, the intelligent agents are organized to perform orderly reasoning, interaction and result integration, thereby achieving accurate response to user requests.

[0052] like Figure 1 , 2 As shown, the specific steps include:

[0053] S1: Define the roles required to resolve user requests for open-source intelligence analysis tasks and construct a collaboration topology graph between roles, then transform user requests into a structured multi-agent collaboration scheme.

[0054] Specifically, the planner based on a large language model receives user requests for open-source intelligence analysis tasks. and a collection of reference materials and information. As input, through reasoning The set of role definitions required to resolve user requests is then provided. Collaboration topology between roles ,Right now .

[0055] The following details the character definition collection. Collaboration topology between roles The composition of.

[0056] For the set of role definitions A planner based on a large language model deeply analyzes user requests. and a collection of reference materials and information. Give the set of n role definitions required to resolve this user request, denoted as . Each element in the set Each tuple is a character name. and role description Role Responsibilities Description The information dimensions and specific reasoning tasks that the role needs to focus on during the analysis process are defined in detail, providing semantic constraints for the subsequent construction of role-based intelligent agents with specific expert perspectives.

[0057] For the collaboration topology diagram between roles The planner based on the large language model constructed a directed acyclic graph describing the collaborative relationships among the aforementioned roles, denoted as . , Let be the set of nodes in the graph. Let be the set of edges of the graph. Let be the set of nodes of the graph. Starting from a preset node Termination node It consists of a set of n execution nodes, where For the role The corresponding execution node; the set of edges in the graph. This defines the flow path of information between nodes, and each directed edge in the edge set... Represents a node The output data will be used as nodes The input context. Through this topology, the planner clarifies the order in which each role processes information and the logic of information flow.

[0058] In this embodiment, the specific content of the prompt words used by the planner based on the large language model is shown in Table 1;

[0059] Table 1. Prompt words used by the planner based on the large language model

[0060] S2: Instantiate agents and build cooperative topology

[0061] This step aims to instantiate the plan generated in phase S1. In this phase, not only are large language model agents with specific domain perspectives instantiated based on role definitions and corresponding communication topologies constructed, but a supervisory agent is also embedded to verify the output, thereby addressing the vulnerability of large language models to illusions during reasoning.

[0062] This step includes the following three sub-steps:

[0063] S201: Instantiation of the role-based intelligent agent;

[0064] For the set of role definitions Each role definition (in ), construct corresponding role agent instances based on large language models for each, denoted as .

[0065] Specifically, the character name generated in step S1 and role description Fill into the character's intelligent agent prompt template In the process, generate system prompts specific to this intelligent agent. By loading this prompt word, Initialized as a having by Reasoning nodes for a specific domain of thinking.

[0066] The above-mentioned character intelligent agent prompt word template The specific details are shown in Table 2, where the input context placeholder {input_context} will be replaced by the actual input data in the subsequent S302 step.

[0067] Table 2. Character Agent Prompt Templates

[0068] S202: Construct the communication topology between role-based intelligent agents;

[0069] After instantiating all role agents, the collaborative topology graph output in step S1 is used as a basis. Establish a communication topology between agents.

[0070] Specifically, for the graph Each directed edge in Configure message routing rules so that nodes Corresponding intelligent agent The output is used as a node. Corresponding intelligent agent Input.

[0071] S203: Embed a supervisory agent in the communication topology;

[0072] To achieve quality control over the inference process, enhancements are made to the basic communication topology built in S202. This step aims to introduce a conditional flow mechanism based on a "generate-review-release" logic, ensuring the correctness of the analysis results of each role agent by embedding a supervisory agent into the existing communication link.

[0073] Specifically, the communication topology diagram constructed in S202 Except for the starting node and termination node Any execution node other than those that undertake actual analysis tasks Instantiate a corresponding supervisory agent And embed it into the role-based intelligent agent node. In the execution process. Supervising the intelligent agent. Verification rules have been configured. and the maximum corrected retry threshold Its core responsibility is to receive information from the role's intelligent agent. The output, and combined with Received input and its role responsibilities Verify that the output does not contain factual errors (i.e., hallucinations) or instances of dereliction of duty.

[0074] By embedding a supervisory agent, this invention transforms the originally unconstrained flow process into a flow process with error correction capabilities while maintaining the original multi-agent collaborative flow direction. This ensures that only analysis results that have passed quality review can be received and processed by downstream nodes.

[0075] In this embodiment, in order to achieve the above-mentioned supervision function, the supervisory agent uses the prompt word templates shown in Table 3:

[0076] Table 3. Cue word templates used by the supervisory agent

[0077] S3: The role-based intelligent agent performs analysis tasks according to the communication topology diagram, and combines the verification results and feedback information of the supervisory intelligent agent to carry out information flow between the role-based intelligent agents, and finally produces a response to the user's request;

[0078] S301: Initialize the queues required for input information and topology scheduling;

[0079] From the starting node Begin, prepare the initial input information. And update all nodes related to the starting node. The in-degree of connected nodes is determined. Then, all nodes with a current in-degree of 0 (i.e., agents that meet the execution conditions) are found and added to the task queue. For multiple tasks in the queue, multi-threading is used for parallel computation to improve the overall processing speed; at the same time, the dependencies of the collaborative topology graph are strictly followed to ensure that a node can only be executed after all upstream nodes are ready.

[0080] S302: The intelligent agent performs inference based on the input context and outputs intermediate analysis results;

[0081] When the role agent is activated from the queue of pending tasks. At that time, first from The upstream node receives the output to form Input context .then, Based on its input context Perform inference and output intermediate analysis results. ,in, This represents the reasoning operation of a large language model. In this process, the agent, based on its defined role, focuses on processing, extracting, and logically deducing information transmitted from upstream, ultimately completing the task defined in its role.

[0082] S303: The supervisory agent reviews the intermediate analysis results output by the role agent and provides the review results and feedback.

[0083] Intermediate analysis results generated It will not be immediately passed to downstream nodes, but rather to the supervising agent corresponding to that role's agent. The review will be conducted.

[0084] Supervisory agent receives role agent Received input context Intermediate analysis results output and the role description of the intelligent agent. As input, perform the audit operation. And ultimately produce two outputs. ,in For the review results, This indicates that the review has been approved. This indicates that the review failed. For feedback (only) Sometimes this has practical significance, including suggestions for correcting hallucinations or pointing out areas of negligence. The review process aims to assess the role of the intelligent agent. Intermediate analysis results output Was it completed correctly and completely? The task described in the text.

[0085] S304: Based on the verification results and feedback information of the supervising agent, perform information flow and error correction between role agents;

[0086] Based on the audit results of the supervisory agent's output The next action is determined dynamically:

[0087] If the audit result is This indicates that the current character's output meets the requirements, thus identifying the character's intelligent agent. All downstream nodes intermediate analysis results Pass it to these nodes and update the nodes. The dependency state; when When all preceding dependencies are satisfied (i.e. When the in-degree is updated to 0, add it to the task queue.

[0088] If the audit result is This will block the intermediate analysis results of the blocking agent from flowing downstream, and will include specific error information. Return to the character's intelligent agent . After receiving feedback, it will be combined The description in the text is then used for further reasoning to generate a revised output. This correction process constitutes an iterative loop until the output passes the review of the supervising agent, or reaches the preset maximum correction retry threshold of the supervising agent. If the maximum correction retry threshold is reached... If the error remains unresolved, a prompt will appear asking whether manual intervention is required or if the analysis process should be stopped.

[0089] In this embodiment, during the return transmission When providing prompts to the character's intelligent agent, use the prompt word template shown in Table 4:

[0090] Table 4 Return Prompt templates used when providing prompts to the character's intelligent agent

[0091] S305: Output of results;

[0092] As each role-based intelligent agent node executes, when any role-based intelligent agent node... The process is completed and passes supervision, and its downstream node is the termination node. At that time, the final analysis results generated by this node are aggregated to the termination node. When all nodes are connected to the termination node After all the preceding nodes have completed their tasks, the terminating node... All collected outputs are summarized and fed back to the user as the final analysis result.

[0093] Through the steps S1 to S3 described above, the present invention can decompose complex open-source intelligence analysis tasks into a controllable, self-correcting multi-agent reasoning process, thereby improving the depth and accuracy of the analysis.

[0094] Example 2:

[0095] An open-source intelligence analysis system based on dynamic role multi-agent systems includes a role definition module, a role agent instantiation module, a supervisory agent embedding module, and a task execution module.

[0096] The role definition module defines the roles required to resolve user requests for open-source intelligence analysis tasks and constructs a collaborative topology graph between roles.

[0097] The role agent instantiation module instantiates role agents with specific domain perspectives based on the role definition and constructs the corresponding communication topology graph.

[0098] The supervisory agent embedding module embeds a supervisory agent into the communication topology to audit the output of the role agent, and outputs audit results and feedback information;

[0099] The task execution module enables the role-based intelligent agent to perform tasks according to the communication topology diagram, and combines the verification results and feedback information of the supervisory intelligent agent to carry out information flow, ultimately producing a response to the user's request.

[0100] Example 3:

[0101] This embodiment proposes an electronic device, including: one or more processors, and a memory, wherein the memory is used to store instructions, and when the instructions are executed by the one or more processors, the one or more processors execute the open-source intelligence analysis method based on dynamic role multi-agent.

[0102] The electronic device may be a mobile phone, computer, or tablet computer, etc., and includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, implements the open-source intelligence analysis method based on dynamic role multi-agent as described in the embodiments. It is understood that the electronic device may also include input / output (I / O) interfaces and communication components.

[0103] The processor is used to execute all or part of the steps in the open-source intelligence analysis method based on dynamic role multi-agent as described in the above embodiments. The memory is used to store various types of data, which may include, for example, instructions for any application or method in an electronic device, as well as application-related data.

[0104] The processor can be implemented as an Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), controller, microcontroller, microprocessor, or other electronic components, and is used to execute the open-source intelligence analysis method based on dynamic role multi-agent described in the above embodiments.

[0105] Example 4:

[0106] This embodiment proposes a computer-readable storage medium that stores executable instructions. When these instructions are executed, if they are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0107] The computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the open-source intelligence analysis method based on dynamic role multi-agent described in the various embodiments of this application.

[0108] The aforementioned storage media include: flash memory, hard disks, multimedia cards, card-type memory (e.g., SD (Secure Digital Memory Card) or DX (Memory Data Register, MDR) memory), random access memory (RAM), static random-access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic storage, disks, optical discs, servers, APP (Application) app stores, and other media capable of storing program verification codes. These media store computer programs, which, when executed by a processor, can implement the various steps of the aforementioned open-source intelligence analysis method based on dynamic role multi-agent systems.

[0109] Example 5:

[0110] This embodiment proposes a computer program product, including a computer program or instructions, which, when executed by a processor, implements the open-source intelligence analysis method based on dynamic role multi-agent.

[0111] Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a computer program product.

[0112] The various embodiments in this application are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope defined by the present invention.

Claims

1. An open-source intelligence analysis method based on dynamic role multi-agent systems, characterized in that, include: Define the roles required to resolve user requests for open-source intelligence analysis tasks and construct a collaboration topology graph among the roles; Instantiate role agents with specific domain perspectives based on role definitions and construct corresponding communication topology graphs; An oversight agent is embedded in the communication topology to audit the output of the role agent, and output audit results and feedback information. The scheduling agent performs analysis tasks according to the communication topology diagram, and combines the verification results and feedback information of the supervisory agent to carry out information flow and error correction, and finally produces a response to the user's request.

2. The open-source intelligence analysis method based on dynamic role multi-agent as described in claim 1, characterized in that, The specific method for defining the roles required to resolve user requests for open-source intelligence analysis tasks and constructing a collaborative topology graph among these roles is as follows: A planner based on a large language model receives open-source intelligence analysis task requests from users. and a collection of reference materials and information. As input, through reasoning The set of role definitions required to resolve user requests is then provided. Collaboration topology between roles ,Right now .

3. The open-source intelligence analysis method based on dynamic role multi-agent as described in claim 2, characterized in that, The set of role definitions Each element in the set Each tuple is a character name. and role description Role Responsibilities Description The information dimensions and specific reasoning tasks that this role needs to focus on during the analysis process are defined; The collaboration topology diagram between roles A directed acyclic graph describing the collaborative relationships between roles is constructed for a planner based on a large language model, denoted as . , Let be the set of nodes in the graph. Let be the set of edges of the graph; let be the set of nodes of the graph. Starting from a preset node Termination node It consists of a set of n execution nodes, where For the role The corresponding execution node; the set of edges in the graph. This defines the flow path of information between nodes, and each directed edge in the edge set... Represents a node The output data will be used as nodes The input context.

4. The open-source intelligence analysis method based on dynamic role multi-agent as described in claim 3, characterized in that, The specific method for instantiating a role agent with a specific domain perspective based on the role definition and constructing the corresponding communication topology graph is as follows: For the set of role definitions Each role definition Construct corresponding role-based intelligent agent instances based on large language models one by one, denoted as... Specifically: Character Name and role description Fill into the character's intelligent agent prompt template In the process, generate system prompts specific to this intelligent agent. By loading this prompt word, Initialized as a having by Reasoning nodes for defined domain-specific thinking methods; After instantiating all the role agents, based on the collaboration topology graph between the roles... Establish the communication topology between intelligent agents, specifically as follows: For the collaboration topology diagram between roles Each directed edge in Configure message routing rules so that nodes Corresponding intelligent agent The output is used as a node. Corresponding intelligent agent Input.

5. The open-source intelligence analysis method based on dynamic role multi-agent as described in claim 4, characterized in that, The specific method for embedding a supervisory agent in the communication topology to verify the output of the agent is as follows: Except for the starting node, the communication topology graph is processed by... and termination node Any execution node other than those that undertake actual analysis tasks Instantiate a corresponding supervisory agent And embed it into the role-based intelligent agent node. In the execution process; supervising the intelligent agent Verification rules have been configured. and the maximum corrected retry threshold Receive the role's intelligent agent The output, combined with the character's intelligent agent Received input and its role responsibilities Verify that the output does not contain factual errors or indicate a failure to fulfill responsibilities.

6. The open-source intelligence analysis method based on dynamic role multi-agent as described in claim 5, characterized in that, The scheduling agent performs analysis tasks according to the communication topology diagram, and combines the verification results and feedback information from the supervisory agent to perform information flow and error correction, ultimately producing a response to the user's request, specifically including: Step 1: Initialize the input information and the queue of tasks to be processed required for topology scheduling; From the starting node Begin, prepare the initial input information. And update all nodes related to the starting node. Find the in-degree of connected nodes; then, find all nodes with a current in-degree of 0 and add them to the task queue; for multiple tasks in the queue, use multi-threading to compute in parallel. Step 2: The active agents in the task queue perform reasoning based on the input context and output intermediate analysis results; When the role agent is activated from the queue of pending tasks. At that time, first from The upstream node receives the output to form Input context ; Based on its input context Perform inference and output intermediate analysis results. ,in, Represents the reasoning operations of a large language model; Step 3: The supervising agent reviews the intermediate analysis results output by the role agent and provides the review results and feedback. Supervisory agent receives role agent Received input context Intermediate analysis results output and the role description of the intelligent agent. As input, perform the audit operation. And ultimately produce two outputs. ,in For the review results, This indicates that the review has been approved. This indicates that the review failed. For feedback; Step 4: Based on the verification results and feedback information from the supervising agent, perform information flow and error correction; If the audit result of the supervisory agent is This indicates that the current character's output meets the requirements, thus identifying the character's intelligent agent. All downstream nodes intermediate analysis results Pass it to these nodes and update the nodes. The dependency state; when a node Add it to the task queue when all its prerequisites are satisfied. If the audit result of the supervisory agent is This will prevent the intermediate analysis results of the blocking agent from flowing downstream, and will include feedback containing specific error information. Return to the character's intelligent agent ; After receiving feedback, it will be combined The description in the text is then used for further reasoning to generate a revised output. This correction process constitutes an iterative loop until the output passes the review of the supervising agent or reaches the preset maximum correction retry threshold of the supervising agent. If the maximum correction retry threshold is reached... If the error remains unresolved, a prompt will appear asking whether manual intervention is required or if the analysis process should be stopped. Step 5: The termination node summarizes all collected outputs and feeds them back to the user as the final analysis result; When any role intelligent agent node The process is completed and passes supervision, and its downstream node is the termination node. At that time, the final analysis results generated by this node are aggregated to the termination node. When all nodes are connected to the termination node After all the preceding nodes have completed their tasks, the terminating node... All collected outputs are summarized and fed back to the user as the final analysis result.

7. An open-source intelligence analysis system based on dynamic role multi-agent, implemented based on the open-source intelligence analysis method based on dynamic role multi-agent as described in claim 1, characterized in that, It includes a role definition module, a role agent instantiation module, a supervisory agent embedding module, and a task execution module; The role definition module defines the roles required to resolve user requests for open-source intelligence analysis tasks and constructs a collaborative topology graph between roles. The role agent instantiation module instantiates role agents with specific domain perspectives based on the role definition and constructs the corresponding communication topology graph. The supervisory agent embedding module embeds a supervisory agent into the communication topology to audit the output of the role agent, and outputs audit results and feedback information; The task execution module enables the role-based intelligent agent to perform tasks according to the communication topology diagram, and combines the verification results and feedback information of the supervisory intelligent agent to carry out information flow, ultimately producing a response to the user's request.

8. An electronic device for executing the open-source intelligence analysis method based on dynamic role multi-agent as described in any one of claims 1-6, characterized in that, include: One or more processors, and a memory for storing instructions that, when executed by the one or more processors, cause the one or more processors to perform the open-source intelligence analysis method based on dynamic role multi-agents.

9. A computer-readable storage medium for executing the open-source intelligence analysis method based on dynamic role multi-agent as described in any one of claims 1-6, characterized in that, It stores executable instructions that, when executed, cause the processor to perform the open-source intelligence analysis method based on dynamic role multi-agents.

10. A computer program product for executing the open-source intelligence analysis method based on dynamic role multi-agent as described in any one of claims 1-6, characterized in that, This includes computer programs or instructions that, when executed by a processor, implement the described open-source intelligence analysis method based on dynamic role multi-agents.

Images