Multi-agent based insurance claim settlement and liability determination method and related device

Abstract

This application provides a multi-agent-based insurance claims verification method and related equipment, applicable to the fintech business field. It includes: acquiring case information and policy terms information of the target claim case; inputting the policy terms information into a pre-trained verification rule extraction model for rule extraction, obtaining multiple verification element rules; inputting the case information and multiple verification element rules into a pre-constructed verification master agent, enabling the master agent to generate a target verification sub-agent list for the target claim case; scheduling each verification sub-agent in the target verification sub-agent list to perform verification verification on the case information, obtaining a sub-verification result for each sub-agent; and generating verification information for the target claim case based on all sub-verification results. This application can improve the processing efficiency and accuracy of insurance claims verification.

Description

Technical Field

[0001] This application pertains to the field of fintech business, and in particular relates to a method and related equipment for insurance claims settlement and liability verification based on multi-agent systems. Background Technology In the claims process of insurance companies, liability determination is the core step in determining whether compensation is due. This step requires a strict judgment based on case information, documents submitted by the client, and the insurance company's investigation records, in accordance with the insurance terms. Traditional liability determination methods often rely on manual review by claims personnel. Faced with a large volume of claims, this is not only inefficient but also prone to errors due to subjective factors.

[0002] In related technologies, to achieve automated claims processing, a pipeline model of "artificial intelligence model + engineering logic + expert rules" is usually adopted. This model heavily relies on claims business experts to manually verify and summarize the knowledge of liability assessment, and the execution logic of liability assessment points often relies on manual configuration. For each different insurance product, special rules need to be set and developed. Summary of the Invention

[0003] The main objective of this application is to propose a multi-agent-based insurance claims assessment method and related equipment. It aims to utilize the semantic understanding capabilities of large language models and the collaborative planning capabilities of multi-agents to achieve automatic extraction of assessment rules and dynamic planning of the execution process, thereby effectively improving the processing efficiency and accuracy of insurance claims assessment.

[0004] To achieve the above objectives, a first aspect of this application proposes a multi-agent-based insurance claims assessment method, the method comprising: Obtain case information and policy terms information for the target claim case; The policy terms information is input into a pre-trained accountability rule extraction model to extract rules, resulting in multiple accountability element rules. The case information and multiple accountability element rules are input into a pre-built accountability master agent, so that the accountability master agent generates a target accountability sub-agent list for the target claim case; The system schedules each sub-agent in the target accountability sub-agent list to perform accountability verification on the case information, and obtains the sub-accounting result of each sub-agent. The accountability information for the target claim case is generated based on all the aforementioned sub-accountability results.

[0005] In some embodiments, the process of inputting the policy terms information into a pre-trained liability extraction model for rule extraction yields a set of multiple liability element rules, including: The liability assessment rule extraction model is used to perform semantic parsing on each clause in the policy terms information to extract constraint values ​​corresponding to multiple liability assessment dimensions; wherein, the liability assessment dimensions include the time of the accident, the location of the accident, the cause of the accident, the object of the accident, and the exclusions. The constraint values ​​extracted from each clause are summarized and deduplicated to generate multiple accountability element rules.

[0006] In some embodiments, the step of inputting the case information and multiple accountability element rules into a pre-built accountability master agent, so that the accountability master agent generates a target accountability sub-agent list for the target claim case, includes: Extract the functional description information of each of the pre-configured core responsibility sub-agents from the core responsibility sub-agent cluster; Based on the case information, multiple accountability element rules, and the function description information, construct contextual prompt information; The contextual prompt information is input into the core responsibility agent so that the core responsibility agent can analyze the items to be verified required to execute multiple core responsibility element rules and establish a mapping relationship between the items to be verified and the functional description information. Based on the mapping relationship, a matching core responsibility sub-agent is determined from the core responsibility sub-agent cluster, and a target core responsibility sub-agent list is generated.

[0007] In some embodiments, the pre-construction process of the core responsible agent includes: Construct a cluster of core responsibility sub-intelligent agents, and determine the functional description information and calling interface of each core responsibility sub-intelligent agent in the cluster; Based on the functional description information, the call interface definition, and the preset thought chain reasoning logic, a task planning prompt instruction is constructed; The task planning prompt instruction is loaded into the core responsibility task planning model and connected to the core responsibility sub-agent cluster to obtain the core responsibility master agent.

[0008] In some embodiments, the step of scheduling each sub-agent in the target accountability sub-agent list to perform accountability verification on the case information, and obtaining a sub-accountability result for each sub-agent, includes: Call the document understanding model node of the current responsible sub-agent to extract entities from the case information and obtain the value of the element to be verified; The value of the element to be verified is matched with the effective range defined by the corresponding verification element rule to obtain the range matching result; Based on the range matching results, determine the verification conclusion and verification reason, and output the sub-verification result.

[0009] In some embodiments, before matching the value of the element to be verified with the valid range defined by the corresponding verification element rule to obtain a range matching result, the method further includes: Based on the verification logic corresponding to the current verification sub-agent, query the policy underwriting data associated with the value of the element to be verified; The value of the element to be verified is compared with the policy underwriting data to obtain the data comparison result; accordingly, the step of determining the verification conclusion and verification reason based on the range matching result includes: comprehensively determining the verification conclusion and the verification reason based on the data comparison result and the range matching result.

[0010] In some embodiments, generating the accountability information for the target claim case based on all the sub-accountability results includes: By summarizing the verification conclusions and verification reasons output by each of the aforementioned verification sub-agents, a comprehensive judgment result for the target claim case is obtained; Based on the comprehensive judgment results, a final verification conclusion is generated for the target claim case, and the verification information is output according to the preset verification dimensions.

[0011] To achieve the above objectives, a second aspect of this application proposes an insurance claims verification device based on multi-agent systems, the device comprising: The acquisition module is used to acquire case information and policy terms information for the target claim case; The extraction module is used to input the policy terms information into a pre-trained accountability rule extraction model for rule extraction, and obtain multiple accountability element rules. The input module is used to input the case information and multiple accountability element rules into a pre-built accountability master agent, so that the accountability master agent can generate a target accountability sub-agent list for the target claim case; The verification module is used to schedule each sub-intelligent agent in the target verification sub-intelligent agent list to perform verification of the case information and obtain the sub-verification result of each sub-intelligent agent; The generation module is used to generate the accountability information for the target claim case based on all the sub-accountability results.

[0012] To achieve the above objectives, a third aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0013] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0014] This application proposes a multi-agent-based insurance claims verification method and related equipment. The method includes: acquiring case information and policy terms information of the target claims case; inputting the policy terms information into a pre-trained verification rule extraction model to extract rules, obtaining multiple verification element rules; inputting the case information and multiple verification element rules into a pre-constructed verification master agent, so that the verification master agent generates a target verification sub-agent list for the target claims case; scheduling each verification sub-agent in the target verification sub-agent list to perform verification verification on the case information, obtaining the sub-verification result of each verification sub-agent; and generating verification information of the target claims case based on all sub-verification results.

[0015] According to the multi-agent-based insurance claims assessment method proposed in this invention, firstly, by acquiring case information and policy terms information of the target claims case, the necessary data foundation is provided for the subsequent automated assessment process. Next, the policy terms information is input into a pre-trained assessment rule extraction model for rule extraction, resulting in multiple assessment element rules. This step utilizes model capabilities to replace traditional manual analysis, accurately extracting key assessment knowledge from unstructured clause texts, avoiding the tediousness and potential omissions of manual processing. Finally, the case information and multiple assessment element rules are input into a pre-built assessment master intelligence. This application utilizes the main intelligent agent to generate a list of target responsibility-reviewing sub-intelligent agents for each target claim case. The main agent's intelligent decision-making enables dynamic orchestration of responsibility-reviewing tasks, eliminating the need for manual configuration for specific insurance products and enhancing the flexibility of these tasks. Then, each responsibility-reviewing sub-intelligent agent in the target responsibility-reviewing sub-intelligence list is scheduled to perform responsibility-reviewing verification on the case information, obtaining the sub-responsibility-reviewing result for each sub-intelligence agent. Through the division of labor and collaboration among multiple sub-intelligence agents, the specific verification actions are completed efficiently. Finally, based on all sub-responsibility-reviewing results, the responsibility-reviewing information for the target claim case is generated, completing the comprehensive determination of claim liability. In summary, this application leverages the semantic understanding capabilities of a large language model and the collaborative planning capabilities of multiple intelligent agents to achieve automatic extraction of responsibility-reviewing rules and dynamic planning of the execution process, thereby effectively improving the processing efficiency and accuracy of insurance claim responsibility review.

[0016] Other features and advantages of this disclosure will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the disclosure. The objectives and other advantages of this disclosure may be realized and obtained by means of the structures particularly pointed out in the description, claims and drawings. Attached Figure Description

[0017] Figure 1 This is a flowchart of the insurance claims settlement and liability assessment method based on multi-agent provided in the embodiments of this application; Figure 2 This is another flowchart of the insurance claims settlement and liability assessment method based on multi-agent provided in the embodiments of this application; Figure 3 This is another flowchart of the insurance claims settlement and liability assessment method based on multi-agent provided in the embodiments of this application; Figure 4 This is another flowchart of the insurance claims settlement and liability assessment method based on multi-agent provided in the embodiments of this application; Figure 5 This is another flowchart of the insurance claims settlement and liability assessment method based on multi-agent provided in the embodiments of this application; Figure 6 This is another flowchart of the insurance claims settlement and liability assessment method based on multi-agent provided in the embodiments of this application; Figure 7 This is another flowchart of the insurance claims settlement and liability assessment method based on multi-agent provided in the embodiments of this application; Figure 8 This is a schematic diagram of the structure of the multi-agent-based insurance claims verification device provided in the embodiments of this application; Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0021] In the claims process of insurance companies, liability determination is the core step in determining whether compensation is due. This step requires a strict review of the case information, the documents submitted by the client, and the insurance company's investigation records, against the terms of the insurance policy. Traditional liability determination methods often rely on manual review by claims personnel. Faced with a large volume of claims, this is not only inefficient but also susceptible to errors due to subjective factors.

[0022] In related technologies, to achieve automated claims processing, a pipeline model of "artificial intelligence model + engineering logic + expert rules" is usually adopted. This model heavily relies on claims business experts to manually verify and summarize the knowledge of liability assessment, and the execution logic of liability assessment points often relies on manual configuration. For each different insurance product, special rules need to be set and developed.

[0023] Based on this, embodiments of this application provide an insurance claims assessment method and related equipment based on multi-agent systems, aiming to utilize the semantic understanding capabilities of large language models and the collaborative planning capabilities of multi-agent systems to achieve automatic extraction of assessment rules and dynamic planning of the execution process, thereby effectively improving the processing efficiency and accuracy of insurance claims assessment.

[0024] The insurance claims settlement method and related equipment based on multi-agents provided in this application are specifically described through the following embodiments. First, the insurance claims settlement method based on multi-agents in this application is described.

[0025] The multi-agent-based insurance claims verification method provided in this application is applicable to the fields of fintech, healthcare, and artificial intelligence. This method can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the multi-agent-based insurance claims verification method, but is not limited to the above forms.

[0026] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0027] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0028] Figure 1 This is an optional flowchart of the multi-agent-based insurance claims verification method provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S105.

[0029] Step S101: Obtain case information and policy terms information for the target claim case.

[0030] Step S102: Input the policy terms information into the pre-trained liability extraction model to extract rules and obtain multiple liability element rules.

[0031] Step S103: Input the case information and multiple accountability element rules into the pre-built accountability master agent, so that the accountability master agent can generate a target accountability sub-agent list for the target claim case.

[0032] Step S104: Schedule each sub-agent in the target accountability sub-agent list to perform accountability verification on the case information and obtain the sub-accountability result of each sub-agent.

[0033] Step S105: Generate the accountability information for the target claim case based on all sub-accountability results.

[0034] Steps S101 to S107, as illustrated in this embodiment, firstly, obtain case information and policy terms information of the target claim case to provide the necessary data foundation for the subsequent automated liability assessment process. Next, the policy terms information is input into a pre-trained liability assessment rule extraction model for rule extraction, resulting in multiple liability assessment element rules. This step utilizes model capabilities to replace traditional manual analysis, accurately extracting key liability assessment knowledge from unstructured clause text, avoiding the tediousness and potential omissions of manual analysis. Subsequently, the case information and multiple liability assessment element rules are input into a pre-built liability assessment main intelligent agent. This method enables the main intelligent agent to generate a list of target responsibility-reviewing sub-intelligent agents for a target claim case. The main agent's intelligent decision-making allows for dynamic orchestration of responsibility-reviewing tasks, eliminating the need for manual configuration for specific insurance products and improving the flexibility of these tasks. Then, each responsibility-reviewing sub-intelligent agent in the target responsibility-reviewing sub-intelligence list is scheduled to perform responsibility-reviewing verification on the case information, obtaining the sub-responsibility-reviewing result for each sub-intelligence agent. Through the division of labor and collaboration among multiple sub-intelligence agents, the specific verification actions are completed efficiently. Finally, based on all sub-responsibility-reviewing results, the responsibility-reviewing information for the target claim case is generated, completing the comprehensive determination of claim liability. In summary, this application leverages the semantic understanding capabilities of a large language model and the collaborative planning capabilities of multiple agents to achieve automatic extraction of responsibility-reviewing rules and dynamic planning of the execution process, thereby effectively improving the processing efficiency and accuracy of insurance claim responsibility review.

[0035] In step S101 of some embodiments, the target claim case refers to a specific business matter currently accepted by the insurance company that requires determination of liability for compensation. Case information typically covers all data from the reporting to acceptance stage, including but not limited to the accident description text provided by the complainant, the time and location of the accident, a list of involved personnel, and claim documents such as ID cards, medical records, and invoices uploaded by the user. Policy terms information refers to the insurance contract text associated with the case, which details the scope of insurance liability, exclusions, compensation limits, and other legally binding provisions. The acquisition process can be achieved by reading structured fields from the database or by parsing image files using an optical character recognition (OCR) interface. This step aims to provide complete contextual input for subsequent automated processing, ensuring that the liability assessment process is based on real and sufficient business data, while supporting the processing of multiple data sources in various formats, providing a data foundation for subsequent large-scale model understanding and analysis.

[0036] In step S102 of some embodiments, the liability assessment rule extraction model is a deep learning network built on large language model technology. It possesses natural language processing and semantic understanding capabilities, and can simulate the reading habits of human experts to deeply analyze unstructured, lengthy policy terms. The rule extraction process transforms legal text into computer-executable logic. Multiple liability assessment element rules refer to specific judgment criteria extracted from the terms, such as the validity period for the time of the incident, geographical restrictions for the location of the incident, or underwriting requirements for specific occupational categories. This step automatically identifies key constraints and parameter indicators in the terms through the model, transforming them into structured rule data. This replaces the tedious work of manually sorting and entering rules by business experts in the traditional model, effectively avoiding the problems of human misunderstanding bias and knowledge lag.

[0037] Please see Figure 2 In some embodiments, step S102 may include, but is not limited to, steps S201 to S202.

[0038] Step S201: Semantic parsing of each clause in the policy terms information is performed using the accountability rule extraction model to extract the constraint values ​​corresponding to multiple accountability dimensions.

[0039] Step S202: Summarize and deduplicate the constraint values ​​extracted from each clause to generate multiple accountability element rules.

[0040] In step S201 of some embodiments, the liability extraction model performs a line-by-line scan and deep semantic understanding of each clause in the policy terms, aiming to transform the text described in natural language into structured data that can be processed by computers. The semantic parsing process not only identifies keywords but also analyzes sentence structure and contextual logic to pinpoint the specific conditions limiting liability. For example, when reading a clause in an employer's liability insurance policy regarding "injury caused by an accident during working hours and at the workplace," the model uses semantic analysis to extract working hours as the constraint value for the time of the accident, the workplace as the constraint value for the location of the accident, and the work performed by the insured as stated in the policy as the constraint value for the cause of the accident. Simultaneously, for exclusion clauses such as "intentional acts by the insured" and "driving under the influence of alcohol," these are extracted and categorized under the liability exclusion dimension. Through this multi-dimensional targeted extraction, the model can extract core business constraint information from complex policy text and map it to predefined business fields, achieving digital decomposition of legal text.

[0041] In step S202 of some embodiments, an integration and cleaning operation is performed on the constraint values ​​extracted from different clauses in the previous steps. Since insurance contracts may repeatedly mention certain general provisions in multiple chapters, or modify the main clauses in special agreements, it is necessary to deduplicate the extracted raw data and eliminate redundant information. For example, the main insurance clause may stipulate general exclusions, while the supplementary insurance clause may mention similar exclusions again. This step will merge and deduplicate these semantically similar expressions through semantic comparison, retaining the most accurate description. In addition, if the restriction on "employee age" in the special agreement conflicts with the main clause, this step will retain the value in the special agreement according to the logic of the priority of clause validity. Finally, the information such as the time, place, and object of the accident scattered in different clauses will be summarized to generate a set of non-redundant liability element rules. For example, a structured rule set containing a clear time range, a specified coverage area, and a complete list of exclusions will be generated to provide high-quality data for subsequent calls by the intelligent agent.

[0042] Through steps S201 to S202 described above, this embodiment first uses a liability assessment rule extraction model to semantically parse the policy terms information. Combined with specific claims scenarios (such as employer's liability insurance, accident insurance, etc.), the complex text is broken down into specific constraints based on key dimensions such as the time, location, and cause of the incident. Subsequently, through aggregation and deduplication, the potential for duplicate descriptions or logical conflicts between clauses is resolved, transforming scattered clause information into structured and standardized liability assessment element rules. This process achieves automated conversion from unstructured text to machine-executable rules, ensuring not only the completeness and accuracy of the liability assessment rules but also effectively solving the problems of time-consuming, labor-intensive, and error-prone traditional manual rule compilation, laying a data foundation for efficient planning and execution by multiple agents.

[0043] In step S103 of some embodiments, the main intelligent agent responsible for verification is a high-level decision-making unit built based on a large language model, possessing logical reasoning and task planning capabilities. After receiving case information and structured verification element rules, the main intelligent agent responsible for verification will dynamically analyze the specific context of the case to determine which specific key points need to be verified in the current case. The target list of sub-intelligent agents responsible for verification is the output of this planning process, which includes several functional units that must be invoked for this specific case. This process realizes the mapping from static rules to dynamic execution paths, enabling the verification process to be flexibly adjusted according to the case, without having to hardcode a fixed execution tree for each new product, thus improving the versatility and adaptability of the solution.

[0044] Please see Figure 3 In some embodiments, step S103 may include, but is not limited to, steps S301 to S304.

[0045] Step S301: Extract the functional description information of each core responsibility sub-agent from the pre-configured core responsibility sub-agent cluster.

[0046] Step S302: Construct contextual prompt information based on case information, multiple accountability element rules, and functional description information.

[0047] Step S303: Input the context prompt information into the main intelligent agent of the responsibility, so that the main intelligent agent of the responsibility can analyze the items to be verified required to execute multiple responsibility element rules, and establish a mapping relationship between the items to be verified and the functional description information.

[0048] Step S304: Determine the matching core responsible sub-agent from the core responsible sub-agent cluster according to the mapping relationship, and generate a target core responsible sub-agent list.

[0049] In step S301 of some embodiments, the pre-configured cluster of claim verification sub-agents constitutes a toolkit with various specialized capabilities, including independent agent units for handling different claims verification tasks. Extracting the functional description information of each claim verification sub-agent refers to obtaining metadata or natural language descriptions that define the specific capabilities, input / output parameters, and applicable scenarios of each sub-agent. For example, for a sub-agent specifically used to verify personnel identity, its functional description information might explicitly state that the agent can call optical character recognition technology to read ID card information and call the backend interface to compare the insured's list; for a sub-agent used to verify time validity, its description information might indicate that it can extract the consultation time from medical record text and compare it with the policy validity period. This step aims to transform the encapsulated code logic into textual tool descriptions that a large language model can read and understand, providing a pool of alternative resources for subsequent intelligent planning.

[0050] In step S302 of some embodiments, constructing contextual prompts is a process of structurally integrating business background, rule constraints, and available tools. This process requires concatenating unstructured case information (such as case descriptions and lists of supporting documents), structured rules for multiple liability assessment elements (such as mandatory compensation standards), and functional description information extracted in the previous step according to a preset template. For example, the prompts can be constructed as an instruction containing role settings, informing the large language model that the current task is to conduct liability assessment planning for a specific employer's liability insurance case, and listing all relevant judgment criteria and descriptions of all currently available sub-agent tools. This structured information construction method can effectively transform complex business scenarios into a reasoning context that the large language model can process, ensuring that the model understands the task objective during planning.

[0051] In step S303 of some embodiments, the main intelligent agent responsible for verification utilizes the logical reasoning and semantic understanding capabilities of the large language model to perform in-depth analysis of the input contextual prompts. Analyzing the unverified items required to execute multiple verification element rules means that the model, through thought chain reasoning, transforms abstract rule clauses into specific action requirements. For example, for the rule "the accident time must be within the validity period of the labor contract," the model analyzes that two specific unverified items need to be executed: "obtain the start and end dates of the labor contract" and "obtain the specific time of the accident." Subsequently, establishing a mapping relationship between the unverified items and functional description information means that after understanding the requirements of the unverified items, the model automatically retrieves and matches the sub-agent most suitable for completing the task from the functional description information. For example, the model finds that the requirement of "obtaining the start and end dates of the labor contract" matches the functional description of the "contract element extraction sub-agent," thereby establishing a logical connection between the two.

[0052] In step S304 of some embodiments, based on the established mapping relationship, the responsible intelligent agent selects the sub-intelligent agents actually needed for the current task from the sub-intelligent agent cluster and generates a target responsible sub-intelligent agent list. This list is not only a collection of sub-intelligent agents, but may also include the logical order or parallel relationship of execution. For example, if the case involves personal injury and property loss, the mapping relationship may point to a personal injury verification sub-intelligent agent and a loss assessment and pricing sub-intelligent agent, and the generated target list will mark these two agents as objects to be scheduled. If certain verification items have dependencies (such as identity verification must be performed before insurance coverage verification), the generated list will also reflect this order. This step completes the implementation from logical planning to execution scheduling, ensuring that the selected combination of sub-intelligent agents is the optimal solution for the current case and eliminating interference from irrelevant intelligent agents.

[0053] Through steps S301 to S304 described above, this embodiment first constructs contextualized accountability tasks by building contextual prompts. Then, it utilizes the reasoning capabilities of the accountability agent to transform business rules into specific execution actions and match them with appropriate tools, finally generating an execution list. This process achieves dynamic orchestration and automated planning of the accountability process, flexibly addressing the differentiated needs of different insurance types and case details. It eliminates the need for manual rewriting of the calling logic for each new rule, improving the system's versatility and intelligence.

[0054] Please see Figure 4 In some embodiments, the pre-construction process of the core responsible agent may include, but is not limited to, steps S401 to S403.

[0055] Step S401: Construct a cluster of responsible sub-intelligent agents and determine the functional description information and calling interface of each responsible sub-intelligent agent in the cluster.

[0056] Step S402: Based on the functional description information, the call interface definition, and the preset thought chain reasoning logic, construct the task planning prompt instruction.

[0057] Step S403: Load the task planning prompts into the core responsibility task planning model and connect it to the core responsibility sub-agent cluster to obtain the core responsibility master agent.

[0058] In step S401 of some embodiments, constructing the claim verification sub-agent cluster refers to building a set of functional units focused on handling a single claim verification task. Each claim verification sub-agent typically consists of a large language model node and a database query interface. Determining the functional description information and calling interface definition of each claim verification sub-agent is a prerequisite for implementing model tool calls. The functional description information is a natural language description of the specific capabilities, applicable scenarios, and meaning of input parameters of the agent. For example, for an "employee list verification sub-agent," its functional description information might be defined as "used to query the company's latest list of insured personnel to verify whether they are insured after receiving their names and ID numbers." The calling interface definition standardizes the data structure for interaction with the agent, clearly specifying the names, types, and whether input fields are required. This step, through standardized encapsulation, transforms program code or algorithm logic into standardized tools that the large language model can understand and call, establishing a unified communication protocol for subsequent automated scheduling.

[0059] In step S402 of some embodiments, constructing task planning prompts is a key process of injecting business logic into the large model using prompting engineering techniques. These prompts not only include functional descriptions and interface definitions of all available sub-agents identified in the previous step, but also incorporate pre-defined thought chain reasoning logic. Thought chain reasoning logic is a strategy that guides the model to think step-by-step. It instructs the model, when faced with complex claims cases, not to directly provide conclusions, but to first analyze the risk points involved in the case information, then compare these risk points with the accountability element rules, thereby identifying missing evidence or data that needs verification, and finally matching the most suitable sub-agent from the tool library based on these specific needs. For example, a prompt might include the following logic: "First, analyze the case description provided by the user; second, compare with the rule set to find the time, location, and personnel elements that need verification; third, for each element, check if there is a matching verification tool in the tool description; fourth, output the tool call sequence." In this way, static tool definitions are combined with dynamic reasoning strategies to generate task planning prompts that guide the model's actions.

[0060] In step S403 of some embodiments, loading the task planning prompts into the core responsibility task planning model and connecting it to the core responsibility sub-agent cluster is the assembly process for initializing the core responsibility master agent. The core responsibility task planning model is a pre-trained generative large language model with long text understanding and instruction following capabilities. The loading process typically inputs the aforementioned prompts as system-level instructions or pre-context into the model, setting it to a specific role (i.e., core responsibility planning expert). Connecting it to the core responsibility sub-agent cluster grants the model actual execution permissions. By registering a tool library, the function call requests generated by the model during inference can be correctly routed to the corresponding sub-agents for execution. This step transforms the general basic large model into an agent entity with specific business planning and tool scheduling capabilities, enabling it to decompose and assign complex core responsibility tasks.

[0061] Through steps S401 to S403 described above, this embodiment first establishes a cluster of sub-agents for claims handling, encompassing various specialized capabilities, through standardized encapsulation. Next, by constructing task planning prompts that include tool definitions and thought chain logic, it endows the general-purpose model with domain-specific business reasoning capabilities. Finally, it instantiates the main agent for claims handling by loading instructions and accessing the tool library. This process transforms discrete tool code into an autonomous agent with intelligent planning capabilities, enabling the main agent for claims handling not only to know which tools are available but also to understand how to combine and use these tools under different case circumstances. This allows it to flexibly respond to complex and ever-changing claims scenarios, achieving automated orchestration and efficient scheduling of claims handling tasks.

[0062] In step S104 of some embodiments, scheduling refers to starting corresponding computing resources according to a predetermined logical order or parallel strategy. Each sub-agent responsible for verification is a pre-built, independent execution unit focused on a single-dimensional verification task, such as an agent specifically responsible for verifying the validity of time or an agent specifically responsible for comparing personnel identities. Verification refers to the process by which the sub-agent responsible for verification compares and analyzes the actual data in the case information with the corresponding verification element rules. After completing the verification, each sub-agent responsible for verification outputs a sub-verification result containing a conclusion of whether it passes or fails, as well as the specific judgment basis. This step decouples the complex comprehensive verification task into multiple sub-tasks. Through the division of labor and cooperation among multiple agents, not only is the efficiency of parallel processing improved, but the verification logic of each dimension is also made traceable.

[0063] Please see Figure 5 In some embodiments, step S104 may include, but is not limited to, steps S501 to S503.

[0064] Step S501: Call the document understanding model node of the current verification sub-agent to extract entities from the case information and obtain the value of the element to be verified.

[0065] Step S502: Match the value of the element to be verified with the valid range defined by the corresponding verification element rule to obtain the range matching result.

[0066] Step S503: Determine the verification conclusion and verification reason based on the range matching result, and output the sub-verification result.

[0067] In step S501 of some embodiments, the document understanding model node in this step possesses dual capabilities of image text recognition and natural language understanding, enabling it to read original documents such as medical records, accident reports, invoices, or employment contracts. Entity extraction refers to accurately extracting key business data fields from the aforementioned original materials, such as extracting the diagnosis time from a hospital diagnosis certificate, the liability ratio from a traffic accident report, or the person's name from an ID card. These extracted specific data are defined as elements to be verified, serving as a crucial intermediate product for transforming ambiguous case materials into structured data that can be logically processed by a computer.

[0068] In step S502 of some embodiments, matching the value of the element to be verified with the valid range defined by the corresponding liability element rule is the process of executing the core verification logic. The valid range is the set or interval of values ​​allowed by the liability element as determined by the policy terms, such as the valid time period consisting of the insurance start and end dates, the coverage area list stipulated in the policy, or the set of specific circumstances listed in the exclusions. The matching process is to mathematically or logically compare the actual occurrence value extracted in the previous step with this theoretically allowed value. For example, if the value of the element to be verified is the time of the incident "July 1, 2025", and the valid range defined by the rule is "January 1, 2024 to December 31, 2024", then the matching operation will result in a negative result that the value falls outside the valid range; if the value of the element to be verified is the cause of the incident "accidental fall", and the valid range defined by the rule excludes "death due to illness", then the matching operation will result in a positive result that it meets the coverage scope.

[0069] Please see Figure 6 In some embodiments, steps S601 to S602 may be included before step S502.

[0070] Step S601: Based on the verification logic corresponding to the current verification sub-agent, query the policy underwriting data associated with the value of the element to be verified. Step S602: Compare the value of the element to be verified with the policy underwriting data to obtain the data comparison result.

[0071] In step S601 of some embodiments, querying the policy underwriting data associated with the value of the element to be verified, according to the verification logic corresponding to the current verification sub-agent, means that the agent actively initiates a retrieval request to the core insurance business database to obtain structured contract details when performing a specific verification task. The verification logic defines the types of external data sources that the agent needs to rely on during the verification process. For example, for the verification task of the insured object in employee liability insurance, the logic stipulates that the latest list of insured persons must be referenced. Policy underwriting data refers to structured information stored in the database used to accurately define the boundaries of insurance liability, including but not limited to the list of insured persons, specially agreed deductibles, endorsement change records, or specific underwriting area codes. For example, when the agent extracts the name of the injured person from the claim materials, it will call the relevant interface to query the list of all insured employees under that policy number, or query whether there are any specially agreed limits for that position. This step extends the static clause rule verification to dynamic real-time database interaction, ensuring that the data on which the verification process is based is kept in real-time synchronized with the insurance company's core underwriting records.

[0072] In step S602 of some embodiments, consistency comparison typically involves judging set inclusion relationships or matching precise values, such as judging whether the injured person's name extracted from the medical record is included in the employee list returned by the query, or judging whether the claimed compensation item belongs to the list of covered items stipulated in the policy. Accordingly, determining the verification conclusion and verification reason based on the data comparison result and the scope matching result means that the verification sub-agent uses multi-dimensional logic and operations to generate the final judgment. It not only requires that the value of the element to be verified meets the valid range set by the general rules (i.e., the scope matching result is true, such as the time of the accident being within the insurance period), but also requires that it must be consistent with the core underwriting record (i.e., the data comparison result is true, such as the person must be on the list). Only when both conditions are met simultaneously is the verification conclusion passed. If either condition is not met, the verification reason will clearly indicate whether it is a violation of the general rules or a discrepancy with the underwriting record. For example, even if the accident occurred within the insurance period, if the injured person is not on the underwriting list, the conclusion of the comprehensive judgment is still a denial of the claim.

[0073] Through steps S601 to S602 described above, this embodiment introduces a real-time query and comparison mechanism for core underwriting data, constructing a dual verification logic. First, by querying policy underwriting data, it achieves coverage of dynamic underwriting information (such as personnel changes and endorsement adjustments). Second, by combining consistency comparison and scope matching results, it achieves verification of the compliance and authenticity of claims cases. This composite verification mechanism improves the accuracy of liability verification for complex policies and effectively prevents erroneous payouts due to delayed underwriting information updates or mismatches of key elements.

[0074] In step S503 of some embodiments, determining the verification conclusion and verification reason based on the range matching result is a process of converting the logical operation result into business decision language. The verification conclusion is usually a clear Boolean value or classification label, used to indicate whether the verification point has passed the verification, such as "yes / no" or "pass / fail". The verification reason is a natural language explanation of the conclusion, used to explain in detail the judgment basis and enhance the interpretability of the result. For example, when the time matching fails, the generated reason will clearly state that "the time of the accident is not within the policy validity period"; when the matching is successful, the reason will state that "the location of the accident is within the coverage area". These two parts of information together constitute the sub-verification result, as the complete deliverable after the current sub-agent has completed the execution of this single verification point, for the subsequent master agent to summarize and judge.

[0075] Through steps S501 to S503 described above, this embodiment of the application first utilizes document understanding model nodes to transform unstructured claims documents into structured data to be verified, overcoming the shortcomings of traditional rule engines in handling images or long texts. Next, by matching with the rules of the accountability elements, the objectivity and compliance of the verification logic are ensured. Finally, an accountability result containing clear conclusions and detailed reasons is output. This process achieves closed-loop execution of a single accountability task, ensuring that the judgment at each accountability point is based on evidence and improving the transparency and traceability of the accountability process.

[0076] In step S105 of some embodiments, generating accountability information is a comprehensive judgment process aimed at integrating fragmented verification results into a final conclusion with business decision-making value. This step summarizes the sub-accountability results fed back by all accountability sub-agents and analyzes whether there are logical conflicts or missing key elements. The accountability information for the target claim case not only includes the final conclusion on whether compensation should be paid, but also includes a detailed explanation of the reasoning basis, such as indicating which specific clause led to the denial of the claim, or what materials confirmed the liability for compensation. Through logical induction and comprehensive judgment of multi-source results, the completeness and interpretability of the final output result are ensured, providing intuitive and reliable auxiliary decision support for the subsequent operations of claims personnel.

[0077] Please see Figure 7 In some embodiments, step S105 may include, but is not limited to, steps S701 to S702.

[0078] Step S701: Summarize the verification conclusions and verification reasons output by each responsible sub-agent to obtain the comprehensive judgment result of the target claim case.

[0079] Step S702: Based on the comprehensive judgment results, generate the final verification conclusion for the target claim case and output the verification information according to the preset verification dimensions.

[0080] In step S701 of some embodiments, since the various sub-agents responsible for verification in the preceding steps perform verification on a single dimension in parallel and independently, this step is responsible for collecting the status data returned by all participating agents, including Boolean value conclusions representing pass or fail and specific factual evidence supporting the conclusions. The resulting comprehensive judgment result is not merely a simple text concatenation, but a structured dataset containing full verification details. For example, in a complex case involving personal injury and vehicle damage, the comprehensive judgment result will simultaneously include the insured status returned by the time verification sub-agent, confirmation within the insured area returned by the location verification sub-agent, and an anomaly marker returned by the object verification sub-agent indicating that the injured party is not on the list. This process ensures that no risk point in any dimension is overlooked, providing a complete and mutually exclusive chain of evidence for subsequent global decision-making.

[0081] In step S702 of some embodiments, generating the final verification conclusion for the target claim case based on the comprehensive judgment result is a decision-making step in executing global business rules. This step can adopt a veto strategy; if any key dimension in the comprehensive judgment result fails verification, the final verification conclusion will be determined as a claim rejection or transferred to manual review; conversely, if all dimensions pass, it will be determined as normal payment. Outputting verification information according to preset verification dimensions means formatting the above decision logic and basis according to a standard framework such as the time, place, and cause of the incident. For example, generating a structured JSON report or verification list, which clearly lists the verification status of each dimension, the specific judgment reason, and the final case handling recommendation. This standardized output method eliminates the data format differences between different cases, making the output results not only intuitively readable by claims personnel but also automatically parsed and processed by downstream payment or case closure systems.

[0082] Through steps S701 to S702 described above, this embodiment first integrates the local verification results scattered across various sub-agents into a global view through a aggregation mechanism; then, it generates a final conclusion through global decision logic and outputs the accountability information in a standardized dimensional format. This process realizes the transformation from fragmented verification to holistic decision-making, ensuring not only the logical rigor and interpretability of the accountability conclusion, but also providing a decision-making basis for the final automated closed loop of the claims process, effectively improving the transparency and interaction efficiency of case handling.

[0083] Please see Figure 8 This application also provides a multi-agent-based insurance claims verification device, which can implement the above-mentioned multi-agent-based insurance claims verification method. The device includes: The acquisition module is used to acquire case information and policy terms information for the target claim case; The extraction module is used to input policy terms information into a pre-trained liability rule extraction model for rule extraction, resulting in multiple liability element rules. The input module is used to input case information and multiple accountability element rules into the pre-built accountability master agent, so that the accountability master agent can generate a list of target accountability sub-agents for the target claim case; The verification module is used to schedule each sub-agent in the target accountability sub-agent list to perform accountability verification on the case information and obtain the sub-accountability result of each sub-agent. The generation module is used to generate accountability information for the target claim case based on all sub-accountability results.

[0084] The specific implementation of this multi-agent-based insurance claims verification device is basically the same as the specific implementation of the multi-agent-based insurance claims verification method described above, and will not be repeated here.

[0085] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described multi-agent-based insurance claims verification method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0086] Please see Figure 9 , Figure 9 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 901 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 902 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and is called and executed by the processor 901 using the multi-agent-based insurance claims assessment method of the embodiments of this application. The input / output interface 903 is used to implement information input and output; The communication interface 904 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 905 transmits information between various components of the device (e.g., processor 901, memory 902, input / output interface 903, and communication interface 904); The processor 901, memory 902, input / output interface 903, and communication interface 904 are connected to each other within the device via bus 905.

[0087] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described multi-agent-based insurance claims verification method.

[0088] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0089] This application proposes a multi-agent-based insurance claims verification method and related equipment. The method includes: acquiring case information and policy terms information of the target claims case; inputting the policy terms information into a pre-trained verification rule extraction model to extract rules, obtaining multiple verification element rules; inputting the case information and multiple verification element rules into a pre-constructed verification master agent, so that the verification master agent generates a target verification sub-agent list for the target claims case; scheduling each verification sub-agent in the target verification sub-agent list to perform verification verification on the case information, obtaining the sub-verification result of each verification sub-agent; and generating verification information of the target claims case based on all sub-verification results.

[0090] According to the multi-agent-based insurance claims assessment method proposed in this invention, firstly, by acquiring case information and policy terms information of the target claims case, the necessary data foundation is provided for the subsequent automated assessment process. Next, the policy terms information is input into a pre-trained assessment rule extraction model for rule extraction, resulting in multiple assessment element rules. This step utilizes model capabilities to replace traditional manual analysis, accurately extracting key assessment knowledge from unstructured clause texts, avoiding the tediousness and potential omissions of manual processing. Finally, the case information and multiple assessment element rules are input into a pre-built assessment master intelligence. This application utilizes the main intelligent agent to generate a list of target responsibility-reviewing sub-intelligent agents for each target claim case. The main agent's intelligent decision-making enables dynamic orchestration of responsibility-reviewing tasks, eliminating the need for manual configuration for specific insurance products and enhancing the flexibility of these tasks. Then, each responsibility-reviewing sub-intelligent agent in the target responsibility-reviewing sub-intelligence list is scheduled to perform responsibility-reviewing verification on the case information, obtaining the sub-responsibility-reviewing result for each sub-intelligence agent. Through the division of labor and collaboration among multiple sub-intelligence agents, the specific verification actions are completed efficiently. Finally, based on all sub-responsibility-reviewing results, the responsibility-reviewing information for the target claim case is generated, completing the comprehensive determination of claim liability. In summary, this application leverages the semantic understanding capabilities of a large language model and the collaborative planning capabilities of multiple intelligent agents to achieve automatic extraction of responsibility-reviewing rules and dynamic planning of the execution process, thereby effectively improving the processing efficiency and accuracy of insurance claim responsibility review.

[0091] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0092] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0093] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0094] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0095] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0096] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0097] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0098] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0099] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0100] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0101] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for insurance claims settlement and liability assessment based on multi-agent systems, characterized in that, The method includes: Obtain case information and policy terms information for the target claim case; The policy terms information is input into a pre-trained accountability rule extraction model to extract rules, resulting in multiple accountability element rules. The case information and multiple accountability element rules are input into a pre-built accountability master agent, so that the accountability master agent generates a target accountability sub-agent list for the target claim case; The system schedules each sub-agent in the target accountability sub-agent list to perform accountability verification on the case information, and obtains the sub-accounting result of each sub-agent. The accountability information for the target claim case is generated based on all the aforementioned sub-accountability results.

2. The insurance claims settlement and liability assessment method based on multi-agent technology according to claim 1, characterized in that, The policy terms information is input into a pre-trained accountability rule extraction model for rule extraction, resulting in a set of multiple accountability element rules, including: The liability assessment rule extraction model is used to perform semantic parsing on each clause in the policy terms information to extract constraint values ​​corresponding to multiple liability assessment dimensions; wherein, the liability assessment dimensions include the time of the accident, the location of the accident, the cause of the accident, the object of the accident, and the exclusions. The constraint values ​​extracted from each clause are summarized and deduplicated to generate multiple accountability element rules.

3. The insurance claims settlement and liability assessment method based on multi-agent technology according to claim 1, characterized in that, The step of inputting the case information and multiple accountability element rules into a pre-built accountability master agent, so that the accountability master agent generates a target accountability sub-agent list for the target claim case, includes: Extract the functional description information of each of the pre-configured core responsibility sub-agents from the core responsibility sub-agent cluster; Based on the case information, multiple accountability element rules, and the function description information, construct contextual prompt information; The contextual prompt information is input into the core responsibility agent so that the core responsibility agent can analyze the items to be verified required to execute multiple core responsibility element rules and establish a mapping relationship between the items to be verified and the functional description information. Based on the mapping relationship, a matching core responsibility sub-agent is determined from the core responsibility sub-agent cluster, and a target core responsibility sub-agent list is generated.

4. The insurance claims settlement and liability assessment method based on multi-agent technology according to claim 3, characterized in that, The pre-construction process of the responsible intelligent agent includes: Construct a cluster of core responsibility sub-intelligent agents, and determine the functional description information and calling interface of each core responsibility sub-intelligent agent in the cluster; Based on the functional description information, the call interface definition, and the preset thought chain reasoning logic, a task planning prompt instruction is constructed; The task planning prompt instruction is loaded into the core responsibility task planning model and connected to the core responsibility sub-agent cluster to obtain the core responsibility master agent.

5. The insurance claims settlement and liability assessment method based on multi-agent technology according to claim 1, characterized in that, The process involves scheduling each sub-agent in the target accountability sub-agent list to perform accountability verification on the case information, obtaining a sub-accountability result for each sub-agent, including: Call the document understanding model node of the current responsible sub-agent to extract entities from the case information and obtain the value of the element to be verified; The value of the element to be verified is matched with the effective range defined by the corresponding verification element rule to obtain the range matching result; Based on the range matching results, determine the verification conclusion and verification reason, and output the sub-verification result.

6. The insurance claims settlement method based on multi-agents as described in claim 5, characterized in that, Before matching the value of the element to be verified with the valid range defined by the corresponding verification element rule to obtain the range matching result, the process further includes: Based on the verification logic corresponding to the current verification sub-agent, query the policy underwriting data associated with the value of the element to be verified; The value of the element to be verified is compared with the policy underwriting data to obtain the data comparison result; accordingly, the step of determining the verification conclusion and verification reason based on the range matching result includes: comprehensively determining the verification conclusion and the verification reason based on the data comparison result and the range matching result.

7. The insurance claims settlement method based on multi-agents as described in claim 5, characterized in that, The generation of accountability information for the target claim case based on all the sub-accountability results includes: By summarizing the verification conclusions and verification reasons output by each of the aforementioned verification sub-agents, a comprehensive judgment result for the target claim case is obtained; Based on the comprehensive judgment results, a final verification conclusion is generated for the target claim case, and the verification information is output according to the preset verification dimensions.

8. An insurance claims verification device based on multi-agent systems, characterized in that, The device includes: The acquisition module is used to acquire case information and policy terms information for the target claim case; The extraction module is used to input the policy terms information into a pre-trained accountability rule extraction model for rule extraction, and obtain multiple accountability element rules. The input module is used to input the case information and multiple accountability element rules into a pre-built accountability master agent, so that the accountability master agent can generate a target accountability sub-agent list for the target claim case; The verification module is used to schedule each sub-intelligent agent in the target verification sub-intelligent agent list to perform verification of the case information and obtain the sub-verification result of each sub-intelligent agent; The generation module is used to generate the accountability information for the target claim case based on all the sub-accountability results.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the multi-agent-based insurance claims assessment method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, characterized in that, when the computer program is executed by a processor, it implements the multi-agent-based insurance claims assessment method according to any one of claims 1 to 7.

Images