A multi-agent-based method and system for determining responsibility, a computer device and a medium

Abstract

The application provides a kind of based on multi-agent's method and system for determining responsibility, computer equipment and medium, comprising: by multiple target disassembly agent, the target of determining responsibility scene is split into multiple atomic tasks, ensure the accuracy of determining responsibility;Through multiple agents, order information, driver information, cost information, privacy number information, instant messaging information and other multi-dimensional data are collected and shared in parallel, ensure the integrity of information in the process of determining responsibility, solve the problem of low machine determining responsibility accuracy caused by relying on single-dimensional information, so as to intelligently determine responsibility for more complex business scenarios;Through multiple determining responsibility agents, according to the rules details and the target information obtained by information collection agent, determine responsibility, and output the result of determining responsibility in a structured form.The application can automatically complete the whole process from data collection, information analysis to final responsibility determination for online service platform through multiple agents, which can improve the efficiency of determining responsibility, reduce the time and cost of manual determining responsibility.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for determining responsibility based on multiple agents, as well as computer equipment and media. Background Technology

[0002] Currently, with the development of internet technology, online service platforms are proliferating. Users can place orders through smartphones, computers, and other devices. Once the platform receives the order, it provides the corresponding service. For example, taking a freight service platform, users can place orders via smartphones or computers, and the platform provides the freight service accordingly. However, as the volume of orders on online service platforms increases, a large number of problematic orders inevitably arise (e.g., cancelled orders, orders with complaints, incorrectly selected orders, orders that cannot be transported). For problematic orders, the online service platform needs to determine the responsible party, i.e., to adjudicate the issue.

[0003] However, when determining responsibility for problematic orders, the relevant technologies rely on limited information dimensions, depending solely on privacy numbers and instant messaging (IM) text messages. This insufficient information utilization prevents the comprehensive and accurate assessment of responsibility based on other relevant order data (such as order notes, highway usage, optional services, order operation records, vehicle type requirements, cost information, and driver history). Furthermore, while these technologies employ machine-based responsibility assessment, existing models suffer from low accuracy and low intelligence, lacking the ability to comprehensively analyze complex scenarios. This inability to make holistic judgments based on order information necessitates manual review and assessment for a large number of orders. Manual review requires accessing dedicated websites for audio listening and detailed analysis, resulting in inefficiency, high costs, and a cumbersome process that negatively impacts user experience and platform operational efficiency. Summary of the Invention

[0004] This invention provides a multi-agent-based method and system for determining responsibility, as well as a computer device and medium, to solve or alleviate the technical problems described above.

[0005] This invention provides a multi-agent-based accountability method, comprising the following steps: obtaining an accountability scenario target, and decomposing the accountability scenario target into multiple atomic tasks through multiple target decomposition agents, and obtaining the rule details corresponding to each atomic task; obtaining target information in parallel through multiple information collection agents, and sharing the target information obtained by each information collection agent with other information collection agents in real time; wherein, the target information includes information in an online service platform used to represent freight services; and performing accountability assessment through multiple accountability agents according to the rule details corresponding to each atomic task and the target information already obtained by each information collection agent, and outputting the accountability assessment results in a structured form.

[0006] In one embodiment of the present invention, before multiple judgment agents perform judgment according to the rule details corresponding to each atomic task and the target information already acquired by each information collection agent, the method further includes: calling a large language model to compare the target information already acquired by each information collection agent with the necessary information of the rule details to determine the information completeness of each information collection agent; when the information completeness of at least one information collection agent does not meet the necessary information of the rule details, generating a collection instruction for re-collecting or supplementing information, and responding to the collection instruction through the corresponding information collection agent to re-collect the target information or supplement the missing information, until the information completeness of each information collection agent meets the necessary information of the rule details.

[0007] In one embodiment of the present invention, the process of acquiring target information in parallel by multiple information acquisition agents includes: calling corresponding tools from the browser automation toolset to crawl page data based on the output results containing tool identifiers and parameters output by the large language model; wherein, the browser automation toolset includes tools for page navigation, button clicking, document object model snapshot crawling, and script injection execution; extracting text from the page data returned by the called tools, and summarizing the context by the large language model, retaining only the page data returned by the most recent tool, and using the page data returned by the previous tool as the target information.

[0008] In one embodiment of the present invention, the plurality of information collection agents include: an order information collection agent, a driver information collection agent, a fee information collection agent, a privacy number information collection agent, and an instant messaging information collection agent.

[0009] In one embodiment of the present invention, the process of acquiring target information in parallel through multiple information acquisition agents includes: acquiring order information representing freight services from an online service platform through an order information acquisition agent; acquiring driver information representing freight services from an online service platform through a driver information acquisition agent; acquiring fee information representing freight services from an online service platform through a fee information acquisition agent; acquiring privacy number information representing freight services from an online service platform through a privacy number acquisition agent; and acquiring instant messaging information representing freight services from an online service platform through an instant messaging information acquisition agent.

[0010] In one embodiment of the present invention, the method further includes: sharing the target information acquired by each information acquisition agent with other information acquisition agents among the plurality of information acquisition agents in real time through a preset storage mechanism, and recording the information completeness of each information acquisition agent in real time through the preset storage mechanism; and / or associating the target information acquired by each information acquisition agent with the judgment result, generating log information for judgment tracing based on the association result, and storing the log information through the preset storage mechanism.

[0011] In one embodiment of the present invention, the process of outputting the accountability result in a structured form includes: taking the results output by the plurality of accountability agents in a structured form, which include whether there is accountability, intermediate accountability results, and accountability basis, as the accountability result of the agents; taking the accountability result of the agents as the final accountability result of the accountability scenario target; or, performing manual sampling inspection on the accountability result of the agents, and taking the manual sampling inspection result as the final accountability result of the accountability scenario target.

[0012] This invention also provides a multi-agent-based accountability system, comprising: a scenario decomposition module for acquiring an accountability scenario target, decomposing the accountability scenario target into multiple atomic tasks through multiple target decomposition agents, and acquiring rule details corresponding to each atomic task; an information acquisition module for acquiring target information in parallel through multiple information acquisition agents, and sharing the target information acquired by each information acquisition agent with other information acquisition agents in real time; wherein the target information includes information representing freight services in an online service platform; and an accountability module for performing accountability assessment through multiple accountability assessment agents according to the rule details corresponding to each atomic task and the target information acquired by each information acquisition agent, and outputting the accountability assessment results in a structured form.

[0013] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the multi-agent-based responsibility determination method described in any one of the above.

[0014] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the multi-agent-based responsibility determination method described in any one of the above.

[0015] The beneficial effects of this invention are as follows: This invention proposes a multi-agent-based method and system for determining liability, along with computer equipment and media. By using multiple target decomposition agents, the liability determination scenario is broken down into multiple atomic tasks, ensuring accuracy. Simultaneously, multiple agents collect and share multi-dimensional data such as order information, driver information, fee information, privacy information, and instant messaging information in parallel, ensuring information integrity during the liability determination process. This overcomes the problem of low accuracy in machine-based liability determination caused by relying on only a single dimension of information in existing technologies, thus enabling intelligent liability determination for more complex business scenarios. Finally, multiple liability determination agents perform liability determination according to the rule details corresponding to each atomic task and the target information acquired by each information collection agent, outputting the determination results in a structured form, ensuring the efficiency of the liability determination process. Therefore, this invention, through multiple clearly defined and collaborative agents, can automatically complete the entire process from data collection and information analysis to final liability determination for online service platforms, thereby transforming the liability determination process of online service platforms from manual to intelligent and automated, improving efficiency, and reducing the time and cost of manual liability determination. Furthermore, this invention enables multiple intelligent agents to perform information collection tasks in parallel, achieving comprehensive collection of multi-source information such as order notes, driver information, and communication records. During the information collection process, multiple intelligent agents share data, which not only overcomes the problem of low machine judgment accuracy caused by relying on only a single dimension of information in existing technologies, but also breaks through the technical bottlenecks of insufficient information utilization, low level of automation and intelligence, poor adaptability to complex rules, and low efficiency of manual operation in existing accountability systems. Thus, it reduces reliance on manual labor and operating costs while improving the accuracy of accountability, processing efficiency, and system integration. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0017] In the attached diagram: Figure 1This is a flowchart illustrating a multi-agent-based responsibility determination method provided in an embodiment of the present invention. Figure 2 This is a flowchart illustrating a multi-agent-based responsibility determination method provided in another embodiment of the present invention. Figure 3 This is a flowchart illustrating a multi-agent-based responsibility determination method provided in another embodiment of the present invention. Figure 4 This is a flowchart illustrating a multi-agent-based responsibility determination method provided in another embodiment of the present invention. Figure 5 This is a schematic diagram of the hardware structure of a multi-agent-based accountability system provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the hardware structure of a computer device suitable for implementing one or more embodiments of the present invention. Detailed Implementation

[0018] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0019] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. The drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0020] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0021] The inventors discovered that determining liability for order cancellations is a highly complex and time-sensitive business process in internet freight service platforms. In existing liability assessment processes, machine-based systems typically initiate automated assessment immediately upon receiving a cancellation request. However, this traditional approach relies heavily on pre-defined, rigid rules and has extremely limited information access channels, often only able to parse private number translation records and instant messaging text messages. This single-dimensional data collection method causes existing machine-based assessment systems to completely lose their ability to comprehensively analyze complex scenarios. For example, in actual business operations, order remarks, vehicle type requirements, cost information, and driver historical behavior are all crucial evidence for determining liability. Due to the lack of this multi-dimensional information, the accuracy of existing machine-based assessment systems remains consistently low, forcing freight service platforms to transfer a large number of disputed orders to human assessors for review. Human review requires assessors to access dedicated websites for audio listening and detailed analysis, resulting in high manpower costs and low efficiency, significantly extending the processing cycle and impacting user experience and platform operational efficiency. Based on this, the present invention proposes a novel automated accountability scheme based on multi-agent collaboration.

[0022] Figure 1 A flowchart illustrating a multi-agent-based responsibility determination method is shown. Specifically, in an exemplary embodiment, as follows... Figure 1 As shown, this embodiment provides a multi-agent-based responsibility determination method, including the following steps: S110: Obtain the judgment scenario target, and decompose the judgment scenario target into multiple atomic tasks through multiple target decomposition agents, and obtain the rule details corresponding to each atomic task. In some examples, if the judgment scenario is related to roads, the judgment scenario target can also be called the judgment scenario route. In some examples, the judgment scenario can be a scenario of judging responsibility for problematic orders on an online service platform, and the judgment scenario target can be the judgment scenario as the task processing target of multiple agents. In some examples, the online service platform includes, but is not limited to, freight service platforms, and problematic orders include, but are not limited to, canceled orders, complained orders, incorrectly selected orders, and orders that cannot be transported due to personal reasons. Among them, the scenario of judging responsibility for canceled orders can also be called the no-show judgment scenario, the scenario of judging responsibility for incorrectly selected orders can also be called the wrongly picked-up order judgment scenario, and the scenario of judging responsibility for orders that cannot be transported due to personal reasons can also be called the personal reason inability to transport judgment scenario. In some examples, atomic tasks include, but are not limited to: verifying whether the driver arrived at the loading point within the specified time, verifying whether the driver actively called the shipper and the call lasted longer than fifteen seconds, and verifying whether the shipper explicitly refused the driver's service within the application. This method decomposes the liability assessment scenario objectives into multiple atomic tasks by breaking down the intelligent agent into multiple objectives, transforming complex legal and business logic into Boolean logic operations that are easily handled by the intelligent agent. In some examples, the intelligent agent can be called an Agent. An intelligent agent can be a proxy or entity capable of perceiving the environment and taking actions to achieve specific goals. The intelligent agent possesses autonomy, adaptability, and interactivity; it can be software, hardware, or a system. The intelligent agent can perceive changes in the environment, make judgments and decisions based on its learned knowledge and algorithms, and then execute actions to influence the environment or achieve predetermined goals.

[0023] S120 involves multiple information collection agents acquiring target information in parallel, and sharing the acquired target information with other information collection agents in real time. The target information includes information representing freight services from the online service platform. Since data acquisition in traditional microservice architectures is often chain-dependent, with subsequent interface calls waiting for the previous interface's return result, this method designs each information collection agent as an independent asynchronous execution unit. Each agent, upon being awakened, simultaneously initiates probe and collection actions from different data sources, thereby reducing the overall information collection time complexity from linear accumulation to a single time consumption dependent on the slowest node. Furthermore, by sharing the target information acquired by each information collection agent with other information collection agents in real time, this method not only overcomes the problem of low accuracy in machine judgment caused by relying on only a single dimension of information in existing technologies, enabling intelligent judgment for more complex business scenarios, but also reduces the concurrent query pressure on the database by sharing the target information collected by multiple information collection agents. This ensures that the target information used by each agent when processing the same order is consistent, avoiding inference conflicts caused by data asynchrony.

[0024] In some examples, a pre-defined storage mechanism can be used to share the target information acquired by each information-collecting agent with other information-collecting agents in real time. This mechanism also records the completeness of information acquired by each agent in real time to confirm whether the target information acquired by each agent meets the necessary requirements of the rule details. In some examples, the pre-defined storage mechanism can be a Memory mechanism. This method introduces a Memory mechanism to provide a shared memory space for all information-collecting agents. When an agent acquires target information such as a driver's real phone number or vehicle license plate, and writes it into the Memory mechanism, other running agents can directly obtain this target information by listening to the Memory mechanism in real time or periodically polling it, without needing to collect it repeatedly. This solves the problems of information silos and duplicate collection during parallel data collection by multiple agents.

[0025] S130: Multiple adjudicating agents perform adjudication according to the rule details corresponding to each atomic task and the target information acquired by each information collection agent, and output the adjudication results in a structured form. In some examples, the process of outputting the adjudication results in a structured form may include: using the results output by multiple adjudicating agents in a structured form, including whether there is liability, intermediate adjudication results, and adjudication basis, as the adjudication results of the agents, and using the adjudication results of the agents as the final adjudication results of the adjudication scenario target. In other examples, the process of outputting the adjudication results in a structured form may include: using the results output by multiple adjudicating agents in a structured form, including whether there is liability, intermediate adjudication results, and adjudication basis, as the adjudication results of the agents, performing manual sampling inspection of the adjudication results, and using the manual sampling inspection results as the final adjudication results of the adjudication scenario target. Therefore, compared with direct manual adjudication, manually sampling the adjudication results of the agents can significantly improve adjudication efficiency and reduce labor costs. The process of manual sampling inspection is as follows: Figure 2 As shown, in Figure 2 In this model, the BERT model consists of multiple agents, including multiple target decomposition agents, multiple information collection agents, and multiple accountability agents. Therefore, the output accountability results not only determine whether relevant personnel (such as drivers and cargo owners) are liable, but also fully preserve intermediate results and the basis for accountability during the process. This highly transparent output method not only provides sufficient evidence for potential subsequent manual appeals, but also provides valuable data samples for freight service platforms to optimize their accountability rule base.

[0026] In some exemplary embodiments, multiple information collection agents include, but are not limited to: an order information collection agent, a driver information collection agent, a fare information collection agent, a privacy number information collection agent, and an instant messaging information collection agent. Therefore, the process of acquiring target information in parallel through multiple information collection agents may include: acquiring order information representing freight services from the online service platform through the order information collection agent; acquiring driver information representing freight services from the online service platform through the driver information collection agent; acquiring fare information representing freight services from the online service platform through the fare information collection agent; acquiring privacy number information representing freight services from the online service platform through the privacy number information collection agent; and acquiring instant messaging information representing freight services from the online service platform through the instant messaging information collection agent. The order information collection agent, driver information collection agent, fare information collection agent, privacy number information collection agent, and instant messaging information collection agent acquire the corresponding target information in parallel and share the acquired target information with each other. For example, after the order information obtained by the order information collection agent is written into the memory mechanism, the driver information collection agent, the fee information collection agent, the privacy number information collection agent, and the instant messaging information collection agent can directly obtain the order information obtained by the order information collection agent by listening to the memory mechanism in real time or periodically polling it. The information sharing process of the driver information collection agent, the fee information collection agent, the privacy number information collection agent, and the instant messaging information collection agent is the same as that of the order information collection agent, and will not be elaborated here.

[0027] In some exemplary embodiments, before multiple adjudicating agents perform adjudication based on the rule details corresponding to each atomic task and the target information already acquired by each information acquisition agent, the process may further include: calling a large language model to compare the target information acquired by each information acquisition agent with the required information of the rule details to determine the information completeness of each information acquisition agent; when the information completeness of at least one information acquisition agent does not meet the required information of the rule details, generating an acquisition instruction for re-acquiring or supplementing information acquisition, and having the corresponding information acquisition agent respond to the acquisition instruction to re-acquire the target information or supplement the missing information until the information completeness of each information acquisition agent meets the required information of the rule details. In some examples, the information completeness of each information acquisition agent can be recorded in real time through a memory mechanism. In some examples, the required information of the rule details corresponding to each atomic task can be determined according to different adjudication scenarios, can be determined by the large language model based on its natural language understanding ability, or can be preset manually; no specific restrictions are imposed here. For example, in a "vehicle model mismatch" liability determination scenario, the necessary information in the rule details must simultaneously possess two data dimensions: "the vehicle model requested by the cargo owner" and "the vehicle model actually driven by the driver." In some examples, when a large language model is invoked to compare the target information already acquired by each information collection agent with the necessary information in the rule details, the large language model can leverage its natural language understanding capabilities to accurately extract entities from the target information already acquired by the information collection agents and semantically align them with the necessary information in the rule details corresponding to each atomic task. This allows for a completeness assessment of the target information already acquired by the information collection agents, avoiding the omission of key information during the liability determination process and preventing inference conflicts caused by different information collection agents having different target information. In some examples, if the information completeness of one or more information collection agents in the comparison results does not meet the necessary information in the rule details, a first collection instruction can be generated for re-collection of information. The corresponding information collection agent then re-collects all target information, thereby achieving completeness in information collection. In some examples, if the information completeness of one or more information-collecting agents in the comparison results does not meet the required information details of the rule, a second collection instruction can be generated for supplementary information collection. The corresponding information-collecting agent then supplements the missing target information, thereby achieving complete information collection. Therefore, this method uses the information completeness of the information-collecting agents to re-collect or supplement information. This dynamic retry strategy based on a reflective mechanism can change the traditional program's running mode of simply continuing until an error occurs.

[0028] In some exemplary embodiments, the process of acquiring target information in parallel through multiple information-gathering agents may further include: calling corresponding tools from a browser automation toolset to crawl page data based on the output results containing tool identifiers and parameters from the large language model; wherein the browser automation toolset includes tools for page navigation, button clicking, Document Object Model (DOM) snapshot capture, and script injection execution; extracting text from the page data returned by the called tools, and summarizing the context by the large language model, retaining only the page data returned by the most recent tool, and using the page data returned by the previous tool as the target information. In some examples, the browser automation toolset may be based on open-source web developer tools built into the browser. In some examples, the browser automation toolset may include navigate_page, click, take_snapshot, evaluate_script, etc., to simulate manual operation of the web page interface. In some examples, when calling a tool, the tool to be called can be determined by the output of the large model containing tool identifiers and parameters, such as calling take_snapshot to capture the current web page content. Meanwhile, text extraction can be performed on the page data returned by the invoked tools. The LLM (Large Language Model) can then summarize the context, retaining only the page data returned by the most recent tool and extracting the web page data returned by the previous tool into target information relevant to the requirements.

[0029] In some exemplary embodiments, after outputting the accountability results in a structured form, the method may further include: associating the target information acquired by each information-collecting agent with the accountability results, generating log information for accountability tracing based on the association results, and storing the log information through a preset storage mechanism. Therefore, by generating and storing log information for accountability tracing, this method allows technicians or business experts to manually review and trace back these log information entries containing complete thought processes when the accountability results output by the accountability agent are questioned. This clearly reproduces the agent's decision-making logic, quickly locates the root cause of the problem, and greatly improves the maintainability and credibility of the agent's accountability results.

[0030] In some exemplary embodiments, a multi-agent-based responsibility determination method is provided, which can be applied to the freight transportation field, including the following steps: Phased Task Planning and Execution: A phased strategy is adopted for task processing. First, information collection and completeness assessment are performed, followed by as-needed judgment and accountability. The first phase uses a fixed process, with multiple agents possessing autonomous thinking (reflection or ReAct) capabilities, allowing for parallel task execution. The second phase is led by autonomous planning agents, dynamically scheduling functional agents to achieve intelligent planning and execution of the overall process. Specifically, the first phase: Information Collection and Completeness Assessment. The fixed information collection phase is executed first. Each functional agent (e.g., order information collection agent, driver information collection agent, etc.) operates autonomously based on the ReAct (Reasoning-Acting) mode, evaluating the completeness of the acquired information through a reflection mechanism. Each functional agent executes tasks in parallel; for example, the order information collection agent extracts basic order data, the driver information collection agent obtains driver vehicle type and historical behavior data, and the IM information collection agent captures communication records. This phase uses a memory mechanism to record the information collection progress in real time. If key information is missing (e.g., vehicle type or arrival time not obtained), a retry or supplementary collection process is automatically triggered. The second phase: Dynamic Planning and Judgment and Execution. The autonomous planning agent dynamically schedules accountability decisions based on the information completeness assessment results from the first phase. If the information is complete, it directly proceeds to the accountability decision-making stage; if the information is incomplete, it plans the next data collection action, such as calling specific tools to complete the data, or generating collection instructions for re-collecting or supplementing information, and the corresponding functional agent responds to the collection instructions to re-collect information or supplement the missing information. Then, it calls the rule base according to scenario requirements. For example, in the "wrong order grabbing" scenario, it prioritizes scheduling the privacy number information collection agent and the IM record collection agent to ensure the sufficiency of the accountability basis. The entire process achieves intelligent process execution through multi-agent collaboration, avoiding the rigidity of fixed processes and improving adaptability and efficiency.

[0031] Multi-Agent Collaboration and Information Sharing: Multiple functional agents are employed, such as order information collection agents, driver information collection agents, fee information collection agents, privacy number information collection agents, and IM information collection agents. These agents can work in parallel and share information and track progress through a memory mechanism. A task planning agent is responsible for overall scheduling, invoking the appropriate information collection agents and accountability agents based on the input scenario ID or the overall scenario's accountability requirements. Specifically, functional agents are divided and processed in parallel: Multiple functional agents are deployed, including: an order information collection agent (which can extract order notes, vehicle type requirements, etc.), a driver information collection agent (which can obtain driver vehicle type, arrival time, etc.), a fee information collection agent (which can capture fee details, etc.), a privacy number information collection agent (which can parse private call records, etc.), and an IM information collection agent (which can collect text chat content, etc.). Each functional agent works in parallel: Data is shared through a central memory (such as sharing key identifiers like order ID and driver ID) to avoid duplicate collection. Unified scheduling of task planning agents: The task planning agent dynamically calls the appropriate agent combination based on the input scenario ID or the overall scenario accountability requirements. The scheduling process is based on a rule base and real-time information status: For example, in the "wrong order grabbing" scenario, the order information agent, privacy number information agent, and IM information collection agent are called first to ensure that core evidence is obtained quickly.

[0032] Dynamic Target Decomposition and Rule Application: The Target Decomposition Agent is responsible for analyzing target rules for each scenario. It can query rule details via API (Application Programming Interface) or retrieve rule details from the knowledge base. It calls functional agents such as the Order Information Collection Agent, Privacy Number Information Collection Agent, and IM Information Collection Agent, and supplements this information with knowledge base experience to ensure information completeness and accuracy of liability determination. Specifically, rule retrieval and target decomposition: The target decomposition agent obtains scenario rule details through API queries or knowledge base searches. For example, for "wrongly accepted orders," it needs to verify whether the driver's vehicle model matches and whether there are proactive communication records. The macro-level liability determination scenario target or route is decomposed into atomic tasks. For example, for a wrongly accepted order, the consistency of the vehicle model is first verified. If they match, it is directly determined as no liability; if they do not match, the liability statement in the communication records is checked. Information Integrity Guarantee: For mandatory information required by the rules (such as vehicle model and arrival time), if any missing information is found, the information collection agent can retry multiple times to complete it.

[0033] Tool Invocation and Information Extraction Optimization: Multiple tool libraries are integrated, including those for webpage access, module location, webpage scrolling, button clicking, information extraction, and internal database / API calls, to support automated operations. For information whose accuracy cannot be confirmed, a knowledge base can be used to supplement it, ensuring the reliability of information collection. After information collection is completed, the completeness of the information is evaluated. If any information is missing, the corresponding functional agent is invoked to complete it.

[0034] Multimodal Tool Library Integration: Integrates browser automation tools including navigate_page, click, take_snapshot, and evaluate_script to simulate manual web interface operations. Tool Invocation: The output results of the large model determine the tools and their parameters, automatically deciding which tool to use, such as calling take_snapshot to fetch the current webpage content. Optimization Measures: For asynchronous loading issues, a "delayed retry" mechanism is adopted (fetching again after 5 seconds); for element recognition issues, the href (hyperlink) attribute is temporarily added through script injection to ensure clickability. Context Summarization and Hot Caching Mechanism: After text extraction of the raw data returned by the tools (e.g., DOM snapshots), the LLM performs context summarization, retaining only the page results returned by the most recent tool, and extracting the webpage content returned by the previous tool into information relevant to the requirements.

[0035] Judgment Decision and Output: The judgment agent, based on information collection results and a rule base, performs scenario-based judgments (e.g., wrongful robbery, no-show, inability to deliver due to personal reasons, etc.). Finally, it summarizes and evaluates the outputs of all judgment agents, outputting a structured judgment result. The entire process maintains contextual coherence through a memory mechanism, ensuring the efficiency and interpretability of the judgment process. Specifically, multi-scenario judgment logic: The judgment agent applies the rule base based on collected information, using specific prompts and order information provided by other information collection agents to perform targeted judgments for each scenario. Structured Result Generation: The judgment result is output in JSON format, including whether liability exists, intermediate judgment results, and the basis for each judgment field. The structured result generation process maintains contextual coherence through a memory mechanism and is saved in log information, ensuring the traceability of the judgment process.

[0036] According to the description of some of the above embodiments, in one example, such as Figure 3 As shown, a flowchart of a method for determining liability based on multiple agents, using order ID, driver ID, and liability determination scenario ID as core input information, and employing the methods described in the above embodiments, is presented. The specific liability determination process will not be elaborated here.

[0037] In some exemplary embodiments, such as Figure 4 As shown, a multi-agent accountability process based on the Model Context Protocol (MCP) is provided, wherein... Figure 4 The chat template is a core tool in the large language model that converts multi-turn dialogue messages into a format that the model can understand. The system prompt is the core instruction set for running the large language model; it defines the model's identity, behavioral rules, capability boundaries, and interaction methods. The user request represents the client's requirements for running the large language model. The next step prompt is the action guide for running the large language model. The ReAct framework includes the Think tool for reflection, the Action execution tool for execution, and the Observation analysis results for information loss detection.

[0038] In summary, this invention proposes a multi-agent-based liability determination method. By using multiple target decomposition agents, the liability determination scenario is broken down into multiple atomic tasks, ensuring accuracy. Simultaneously, multiple agents collect and share multi-dimensional data such as order information, driver information, fee information, privacy information, and instant messaging information in parallel, ensuring information integrity during the liability determination process. This overcomes the problem of low accuracy in machine-based liability determination caused by relying on only a single dimension of information in existing technologies, thus enabling intelligent liability determination for more complex business scenarios. Finally, multiple liability determination agents perform liability determination according to the rule details corresponding to each atomic task and the target information acquired by each information collection agent, outputting the determination results in a structured form to ensure the efficiency of the liability determination process. Therefore, this method, through multiple clearly defined and collaborative agents, can automatically complete the entire process from data collection and information analysis to final liability determination for online service platforms, thereby transforming the liability determination of online service platforms from manual to intelligent and automated, improving efficiency, and reducing the time and cost of manual liability determination. Furthermore, this method utilizes multiple intelligent agents to execute information collection tasks in parallel, achieving comprehensive collection of multi-source information such as order remarks, driver information, and communication records. During the information collection process, multiple agents share data, overcoming the low accuracy of machine-based judgments due to reliance on single-dimensional information in existing technologies. It also overcomes the technical bottlenecks of existing accountability systems, such as insufficient information utilization, low levels of automation and intelligence, poor adaptability to complex rules, and low efficiency of manual operations. This improves accuracy, processing efficiency, and system integration while reducing reliance on manual labor and operational costs. Simultaneously, this method utilizes open-source web developer tools built into the browser to create an automated browser tool, solving the problem of operating various web systems without API integration and fully simulating manual operation processes, thus overcoming data silos between systems. Additionally, it optimizes the stability of asynchronous web page loading, resolving issues of incomplete page element loading and misidentification of StaticText, and improves page loading accuracy through delayed retries and script injection schemes. By employing contextual summarization and a memory mechanism for hot caching management, the problem of high token consumption and inference confusion caused by the verbose DOM information returned by the take_snapshot tool is resolved, improving the success rate of complex tasks. LLM decision-making based on the ReAct framework addresses the issue of a single-direction LLM task process, enabling large models to have a reflective mechanism when acquiring information. Multimodal information fusion decision-making solves the problem of inaccurate accountability based on a single information source, achieving comprehensive analysis and accurate accountability based on multi-source information such as order data, driver information, and communication records. Private LLM deployment and secure isolation address sensitive data security issues, ensuring that all driver and passenger data is processed in a localized environment, meeting the highest level of compliance requirements.

[0039] In an exemplary embodiment of the present invention, such as Figure 5 As shown, a multi-agent-based accountability system is provided, including: The scenario decomposition module 510 is used to obtain the judgment scenario target, and decompose the judgment scenario target into multiple atomic tasks through multiple target decomposition intelligent agents, and obtain the rule details corresponding to each atomic task; The information acquisition module 520 is used to acquire target information in parallel through multiple information acquisition agents, and to share the target information acquired by each information acquisition agent with other information acquisition agents in real time; wherein, the target information includes information in the online service platform used to represent freight services; The accountability module 530 is used to perform accountability assessments by multiple accountability agents according to the rule details corresponding to each atomic task and the target information acquired by each information collection agent, and output the accountability results in a structured form.

[0040] It is understood that the multi-agent-based accountability system and the multi-agent-based accountability method provided in the above embodiments belong to the same concept. The specific execution method of the multi-agent-based accountability method has been described in detail in the above method embodiments and will not be repeated here. In practical applications, the multi-agent-based accountability system provided in the above embodiments can allocate the above functions to different functional modules as needed. That is, the internal structure of the multi-agent-based accountability system is divided into different functional modules, and then all or part of the functions of the corresponding functional modules are implemented through the multi-agent-based accountability method described in the above embodiments. For example, all or part of the functions of the scene decomposition module 510 can be implemented through the relevant execution process of step S110, all or part of the functions of the information collection module 520 can be implemented through the relevant execution process of step S120, and all or part of the functions of the accountability module 530 can be implemented through the relevant execution process of step S130. For specific implementation processes, please refer to the above embodiments, and no specific limitations are imposed here.

[0041] In summary, this invention proposes a multi-agent-based accountability system. By using multiple target decomposition agents, the accountability scenario objective is broken down into multiple atomic tasks, ensuring accuracy. Simultaneously, multiple agents collect and share multi-dimensional data such as order information, driver information, fee information, privacy information, and instant messaging information in parallel, ensuring information integrity during the accountability process. This overcomes the problem of low accuracy in machine-based accountability due to reliance on single-dimensional information in existing technologies, enabling intelligent accountability for more complex business scenarios. Finally, multiple accountability agents perform accountability based on the rule details corresponding to each atomic task and the target information acquired by each information collection agent, outputting the results in a structured form to ensure efficiency. Therefore, this system, through multiple clearly defined and collaborative agents, can automatically complete the entire process from data collection and information analysis to final accountability determination for online service platforms. This transforms the accountability process of online service platforms from manual to intelligent and automated, improving efficiency and reducing the time and cost of manual accountability. Furthermore, this system utilizes multiple intelligent agents to perform information collection tasks in parallel, achieving comprehensive collection of multi-source information such as order notes, driver information, and communication records. During the information collection process, multiple agents share data, overcoming the low accuracy of machine-based judgments due to reliance on single-dimensional information in existing technologies. It also overcomes the technical bottlenecks of existing accountability systems, such as insufficient information utilization, low levels of automation and intelligence, poor adaptability to complex rules, and low efficiency of manual operations. This improves accuracy, processing efficiency, and system integration while reducing reliance on manual labor and operating costs. Simultaneously, this system utilizes open-source web developer tools built into the browser to create an automated browser tool, solving the problem of operating various web systems without API integration and fully simulating manual operation processes, thus overcoming data silos between systems. Additionally, by optimizing the stability of asynchronous web page loading, it resolves issues of incomplete page element loading and misidentification of StaticText. Furthermore, it improves page loading accuracy through delayed retries and script injection schemes. By employing contextual summarization and a memory mechanism for hot caching management, the problem of high token consumption and inference confusion caused by the verbose DOM information returned by the take_snapshot tool is resolved, improving the success rate of complex tasks. LLM decision-making based on the ReAct framework addresses the issue of a single-direction LLM task process, enabling large models to have a reflective mechanism when acquiring information. Multimodal information fusion decision-making solves the problem of inaccurate accountability based on a single information source, achieving comprehensive analysis and accurate accountability based on multi-source information such as order data, driver information, and communication records. Private LLM deployment and secure isolation address sensitive data security issues, ensuring that all driver and passenger data is processed in a localized environment, meeting the highest level of compliance requirements.

[0042] In an exemplary embodiment of the present invention, a computer device is also provided. The computer device may include a memory, a processor, and a computer program stored in the memory. The processor can execute the computer program to cause the computer device to perform actions such as... Figure 1 The steps of the multi-agent-based accountability method are shown. Figure 6 A schematic diagram of the structure of a computer device 1000 is shown. (See attached diagram.) Figure 6 As shown, the computer device 1000 includes: a processor 1010, a memory 1020, a power supply 1030, a display unit 1040, and an input unit 1060.

[0043] The processor 1010 is the control center of the computer device 1000. It connects various components via interfaces and lines, and performs various functions of the computer device 1000 by running or executing computer programs / instructions stored in the memory 1020, thereby providing overall monitoring of the computer device 1000. In some embodiments, when the processor 1010 calls a computer program stored in the memory 1020, it can execute, for example... Figure 1 The steps of the multi-agent-based responsibility determination method are shown. Optionally, the processor 1010 may include one or more processing units; preferably, the processor 1010 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. In some embodiments, the processor 1010 and the memory 1020 may be implemented on a single chip; in other embodiments, they may be implemented on separate chips.

[0044] The memory 1020 mainly includes a program storage area and a data storage area. The program storage area can store the operating system, various applications, etc.; the data storage area can store instruction data created according to the use of the computer device 1000. In addition, the memory 1020 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0045] The computer device 1000 also includes a power supply 1030 (such as a battery) that supplies power to various components. The power supply can be logically connected to the processor 1010 through a power management system, thereby enabling the management of functions such as charging, discharging, and power consumption through the power management system.

[0046] The display unit 1040 can be used to display information input by the user or information provided to the user, and can also be used to display various menus of the computer device 1000, etc. In this embodiment of the invention, it is mainly used to display the display interfaces of various applications in the computer device 1000, as well as text, pictures, and other objects displayed in the display interfaces. The display unit 1040 may include a display panel 1050. The display panel 1050 may be configured in the form of a liquid crystal display (LCD), an organic light-emitting diode (OLED), or the like.

[0047] The input unit 1060 can be used to receive information such as numbers or characters input by the user. The input unit 1060 may include a touch panel 1070 and other input devices 1080. The touch panel 1070 can also be referred to as a touch screen, and the touch panel 1070 can collect touch operations on or near the user (such as operations performed by the user using a finger, stylus, or any suitable object or accessory on or near the touch panel 1070).

[0048] Specifically, the touch panel 1070 can detect user touch operations and the signals generated by these operations, convert these signals into touch point coordinates and send them to the processor 1010, and receive and execute commands transmitted by the processor 1010. Furthermore, the touch panel 1070 can employ various input methods such as resistive, capacitive, infrared, and surface acoustic waves to achieve interaction. Other input devices 1080 include, but are not limited to, one or more of the following: physical keyboard, function keys (such as volume control buttons, power buttons, etc.), trackball, mouse, and joystick.

[0049] Of course, the touch panel 1070 can also cover the display panel 1050. When the touch panel 1070 detects a touch operation on or near it, it can transmit the information to the processor 1010 to determine the type of touch event. Subsequently, the processor 1010 provides corresponding visual output on the display panel 1050 based on the type of touch event. Although in Figure 6 In this embodiment, the touch panel 1070 and the display panel 1050 are two separate components to realize the input and output functions of the computer device 1000. However, in some embodiments, the touch panel 1070 and the display panel 1050 can be integrated to realize the input and output functions of the computer device 1000.

[0050] The computer device 1000 may also include one or more sensors, such as pressure sensors, gravity acceleration sensors, proximity sensors, etc. Of course, depending on the specific application scenario, the computer device 1000 may also include other components such as cameras.

[0051] In an exemplary embodiment of the present invention, a computer-readable storage medium is also provided, which stores a computer program / instructions. When executed by a processor, the computer program / instructions enable the computer device to perform the functions described in the present invention. Figure 1 The steps of the multi-agent-based accountability method are shown.

[0052] It will be understood by those skilled in the art that Figure 6 This is merely an example of a computer device and does not constitute a limitation on the device. The device may include more or fewer components than illustrated, or a combination of certain components, or different components. For ease of description, the above parts are divided into modules (or units) according to their functions and described separately. Of course, in implementing this invention, the functions of each module (or unit) can be implemented in one or more software or hardware components.

[0053] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A method for assigning responsibility based on multiple agents, characterized in that, The method includes: The system obtains the target of the judgment scenario, and decomposes the target of the judgment scenario into multiple atomic tasks through a multi-target decomposition intelligent agent, and obtains the rule details corresponding to each atomic task; Multiple information-collecting agents acquire target information in parallel, and the target information acquired by each agent is shared in real time with other agents in the multiple information-collecting agents; wherein, the target information includes information in the online service platform used to represent freight services; Multiple adjudicating agents perform adjudication according to the rule details corresponding to each atomic task and the target information acquired by each information collection agent, and output the adjudication results in a structured form.

2. The multi-agent-based responsibility determination method according to claim 1, characterized in that, Before multiple adjudicating agents perform adjudication according to the rule details corresponding to each atomic task and the target information acquired by each information collection agent, the method further includes: The large language model is invoked to compare the target information acquired by each information collection agent with the necessary information of the rule details to determine the information completeness of each information collection agent. When the information completeness of at least one information collection agent does not meet the required information of the rule details, a collection instruction for re-collecting or supplementing information is generated, and the corresponding information collection agent responds to the collection instruction to re-collect the target information or supplement the missing information until the information completeness of each information collection agent meets the required information of the rule details.

3. The multi-agent-based responsibility determination method according to claim 2, characterized in that, The process of acquiring target information in parallel through multiple information-gathering agents includes: Based on the output results containing tool identifiers and parameters from the large language model, the corresponding tools are called from the browser automation toolset to crawl page data; wherein, the browser automation toolset includes tools for page navigation, button clicks, document object model snapshot crawling, and script injection execution; The page data returned by the invoked tools is extracted into text, and the large language model is used to summarize the context. Only the page data returned by the most recent tool is retained, and the page data returned by the previous tool is used as the target information.

4. The multi-agent-based responsibility determination method according to any one of claims 1 to 3, characterized in that, The multiple information collection agents include: an order information collection agent, a driver information collection agent, a fee information collection agent, a privacy number information collection agent, and an instant messaging information collection agent.

5. The multi-agent-based responsibility determination method according to claim 4, characterized in that, The process of acquiring target information in parallel through multiple information acquisition agents includes: acquiring order information representing freight services from the online service platform through the order information acquisition agent; acquiring driver information representing freight services from the online service platform through the driver information acquisition agent; acquiring fee information representing freight services from the online service platform through the fee information acquisition agent; acquiring privacy number information representing freight services from the online service platform through the privacy number acquisition agent; and acquiring instant messaging information representing freight services from the online service platform through the instant messaging information acquisition agent.

6. The multi-agent-based responsibility determination method according to claim 1, characterized in that, The method further includes: The target information acquired by each information acquisition agent is shared in real time with other information acquisition agents among the multiple information acquisition agents through a preset storage mechanism, and the information integrity of each information acquisition agent is recorded in real time through the preset storage mechanism. And / or, associate the target information acquired by each information collection agent with the judgment result, generate log information for judgment tracing based on the association result, and store the log information through the preset storage mechanism.

7. The multi-agent-based responsibility determination method according to claim 1, characterized in that, The process of outputting the judgment results in a structured form includes: The results output by the multiple accountability-judging agents in a structured form, including whether there is accountability, intermediate accountability results, and the basis for accountability, are taken as the accountability results of the agents. The intelligent agent's accountability result is used as the final accountability result for the accountability scenario target; or, the intelligent agent's accountability result is manually sampled, and the result of the manual sampling is used as the final accountability result for the accountability scenario target.

8. A multi-agent-based accountability system, characterized in that, The system includes: The scenario decomposition module is used to obtain the judgment scenario target, and decompose the judgment scenario target into multiple atomic tasks through multiple target decomposition intelligent agents, and obtain the rule details corresponding to each atomic task; An information acquisition module is used to acquire target information in parallel through multiple information acquisition agents, and to share the target information acquired by each information acquisition agent with other information acquisition agents in real time; wherein, the target information includes information in the online service platform used to represent freight services; The accountability module is used to perform accountability assessments by multiple accountability agents according to the rule details corresponding to each atomic task and the target information acquired by each information acquisition agent, and output the accountability results in a structured form.

9. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the multi-agent-based liability determination method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the steps of the multi-agent-based accountability method as described in any one of claims 1 to 7.

Images