Business handling method and device, computer equipment and storage medium

Abstract

The invention discloses a business handling method and device, computer equipment and a storage medium, and the method comprises the steps: receiving a user business request, and extracting business information containing a user identity, a target business and an association condition from the business request; based on the service request, searching a service rule, a mutual exclusion relation and a constraint condition related to the service information in a pre-constructed service knowledge graph; inputting the business rule, the mutual exclusion relation and the constraint condition into a local fine-tuning large model, and generating a business handling path containing a conflict judgment and processing strategy through thinking chain reasoning; according to the business handling path, generating a structured operation instruction which can be identified and executed by a business system; and sending the operation instruction to the service system to promote the service system to execute a service corresponding to the operation instruction. And the business handling efficiency and success rate are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent business, and in particular to a business handling method and device, computer equipment and a storage medium. BACKGROUND

[0002] In recent years, in the field of traditional intelligent business handling, the existing systems and technologies have many defects. Traditional business handling systems usually rely on fixed rule engines, which are difficult to adapt to rapidly changing business environments and complex business rules. Due to the delay in updating rules, the system is prone to handling failure or inaccurate results when dealing with complex businesses involving contract restrictions, mutual exclusion relationships, etc. In addition, when traditional systems face frequent changes in business rules, a large amount of manual intervention is required to maintain and update the rule base, which not only increases operating costs, but also may cause system downtime or service interruption. The existing technology lacks efficient transaction management and rollback mechanisms when dealing with multi-business system collaboration and complex business processes. This leads to difficulties in restoring the system to a consistent state when a certain link fails, which can cause data inconsistency or incomplete business handling. SUMMARY

[0003] The purpose of the present application is to provide a business handling method, device, equipment and medium to at least solve the problems of low reliability, low accuracy and low explainability in current intelligent business handling.

[0004] To solve the above technical problems, the present application provides a business handling method, comprising: receiving a user business request, extracting business information including user identity, target business and associated conditions from the business request; Based on the business request, retrieving business rules, mutual exclusion relationships and constraint conditions related to the business information in the pre-constructed business knowledge graph; Input the business rules, mutual exclusion relationships and constraint conditions into the locally fine-tuned large model, and generate a business handling path containing conflict judgment and processing strategy through thought chain reasoning; According to the business handling path, generate structured operation instructions that can be recognized and executed by the business system; Send the operation instruction to the business system to prompt the business system to execute the operation instruction corresponding to the business.

[0005] Optionally, the business rules, mutual exclusion relationships and constraint conditions related to the business information in the pre-constructed business knowledge graph are retrieved, comprising: adopting a graph convolutional neural network to embed and represent the nodes in the pre-constructed business knowledge graph, generating node vectors containing semantic information; Based on the node vector, the semantic similarity between the service request and the graph node is calculated, and an initial associated node set is located; Starting from the initial associated node set, subgraph retrieval based on shortest path and adjacency matrix analysis is performed to extract a subgraph structure containing complete rule chains and mutual exclusion relationships; From the subgraph structure, retrieve the business rules, mutual exclusion relationships, and constraint conditions related to the business information.

[0006] Optionally, the business handling path containing conflict judgment and processing strategy generated by the thought chain reasoning includes: Based on the retrieved constraint conditions, gradually reason and explicitly output the business conflict detection logic and potential risk warning; For the identified conflicts, simulate multiple processing strategies and evaluate their cascading effects on user status and contract terms; Integrate the evaluation results to generate an interpretable business handling path containing specific business operation sequences.

[0007] Optionally, the structured operation instructions that can be recognized and executed by the business system are generated according to the business handling path, including: Parse the business handling path and convert each atomic operation step in it into corresponding system call logic; Map the system call logic to specific structured instructions that can be directly executed by the business system, including API call commands, SQL statements, or workflow tasks; Sort and dependency check the structured instructions according to business logic to form an ordered operation instruction sequence.

[0008] Optionally, the business handling path includes an alternative handling path, and the alternative handling path generation method includes: Detect business conflicts between the user's target business and existing businesses; When detecting that the user's target business and existing businesses have mutual exclusion relationships, extract the existing business identifiers that cause mutual exclusion and the corresponding contract terms; Confirm the mutual exclusion business represented by the mutual exclusion relationship according to the existing business identifiers and corresponding contract terms, and generate an alternative handling path to cancel the mutual exclusion business and handle the target business.

[0009] Optionally, after confirming the mutual exclusion business represented by the mutual exclusion relationship according to the existing business identifiers and corresponding contract terms, and generating an alternative handling path to cancel the mutual exclusion business and handle the target business, it further includes: Calculate the cost information of the alternative handling path; establish the association of the alternative handling path and the cost information to integrate the alternative handling path and the cost information into the finally generated structured operation instruction.

[0010] Optionally, after the operation instruction is sent to the business system to prompt the business system to perform the business corresponding to the operation instruction, the method further includes: monitoring the execution of the operation instruction; when the execution of the operation instruction fails, parsing an error code returned by the execution failure to determine whether it is a temporary system exception or a permanent business rule conflict; if it is a temporary exception, generating a retry instruction and adding it to a delay task queue to wait for retriggering; if it is a permanent conflict or retry failure, terminating the automated process and pushing a case context and a reasoning chain to a manual processing queue.

[0011] To solve the above technical problems, the application further provides a business handling device, which includes: a business request module configured to receive a user business request and extract business information including user identity, target business and associated conditions from the business request; a graph retrieval module configured to retrieve business rules, mutual exclusion relationships and constraint conditions related to the business information in a pre-constructed business knowledge graph based on the business request; a business path module configured to input the business rules, mutual exclusion relationships and constraint conditions into a locally fine-tuned large model to generate a business handling path including conflict judgment and processing strategy through thought chain reasoning; a business instruction module configured to generate a structured operation instruction recognizable and executable by a business system according to the business handling path; an instruction execution module configured to send the operation instruction to the business system to prompt the business system to perform the business corresponding to the operation instruction.

[0012] To solve the above technical problems, the application further provides a computer device including a memory and a processor, the memory storing computer readable instructions, and the computer readable instructions being executed by the processor to make the processor execute the steps of the above-mentioned business handling method.

[0013] To solve the above technical problems, the application further provides a storage medium storing computer readable instructions, and the computer readable instructions being executed by one or more processors to make the one or more processors execute the steps of the above-mentioned business handling method.

[0014] The beneficial effects of the embodiments created by the present application are: by automatically extracting business information and retrieving relevant rules in the knowledge graph, combining the thinking chain reasoning of the locally fine-tuned large model to generate the business handling path of conflict judgment and processing strategy, the problem of existing business handling system knowledge update lag, unable to cover complex rules and mutual exclusion relationship is effectively solved. This method realizes the full-process intelligentization of business handling, reduces manual intervention, reduces operating costs, and improves handling efficiency and success rate. At the same time, the business handling path is converted into structured operation instructions and sent to the business system for execution, ensuring the accuracy and timeliness of the operation. In addition, the locally fine-tuned large model enhances the understanding of industry knowledge, and the thinking chain reasoning improves the transparency and explainability of the reasoning process, making the business handling more reliable and controllable. BRIEF DESCRIPTION OF DRAWINGS

[0015] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which: Figure 1 Basic flowchart of a business handling method for one specific embodiment of the present application; Figure 2 Basic structure diagram of a business handling device for one specific embodiment of the present application; Figure 3 Basic structure block diagram of a computer device for one specific embodiment of the present application. DETAILED DESCRIPTION

[0016] The embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, in which the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present application.

[0017] Those skilled in the art of the present technology can understand that, unless specifically stated, the singular forms "a", "an" and "the" used herein also include the plural forms. It should be further understood that the phrase "comprising" used in the specification of the present application means that the features, integers, steps, operations, elements and / or components exist, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0018] Those skilled in the art of the technology in question can appreciate that all terms used herein, including technical and scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs unless otherwise defined. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the specification and relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0019] Those skilled in the art of the technology in question can appreciate that the term "terminal" used herein includes both devices with only a wireless signal receiver and devices with both receiving and transmitting hardware capable of performing bidirectional communication over a bidirectional communication link. Such devices can include cellular or other communication devices with or without a multi-line display, a PCS (Personal Communications Service) that can combine a voice, data processing, facsimile, and / or data communications capabilities, a PDA (Personal Digital Assistant) that can include a radio frequency receiver, a pager, an Internet / Intranet access, a Web browser, a notepad, a calendar, and / or a GPS (Global Positioning System) receiver, a conventional laptop and / or palmtop computer or other device that has and / or includes a radio frequency receiver. The term "terminal" used herein can be portable, transportable, installed in a vehicle (aeronautical, maritime, and / or land), or adapted and / or configured to operate locally and / or in a distributed manner in any other location on Earth and / or in space. The term "terminal" used herein can also be a communication terminal, an Internet terminal, a music / video playing terminal, such as a PDA, a MID (Mobile Internet Device), and / or a mobile phone with music / video playing functions, a smart television, a set-top box, and the like.

[0020] The hardware referred to by the terms "server", "client", "service node", and the like in the present application is essentially an electronic device with the equivalent capability of a personal computer, which is a hardware device with necessary components disclosed by the von Neumann principle, including a central processing unit (including an arithmetic unit and a controller), a memory, an input device, and an output device, and the like. A computer program is stored in the memory, the central processing unit loads the program stored in the external memory into the memory and runs it, executes the instructions in the program, and interacts with the input and output devices, thereby completing a specific function.

[0021] It should be noted that the concept of "server" in the present application can also be extended to the case of a server cluster. According to the principle of network deployment understood by those skilled in the art, the servers should be logically divided, and in physical space, these servers can be independent of each other but can be called through an interface, or can be integrated into a physical computer or a computer cluster. Those skilled in the art should understand this variation and should not be restricted by the implementation of the network deployment of the present application.

[0022] One or more technical features of the present application, unless explicitly specified, can be deployed on a server and accessed by a client remotely calling an online service interface provided by the server, or can be directly deployed and run on a client to implement access.

[0023] The AI model referred to or possibly referred to in the present application, unless explicitly specified, can be deployed on a remote server and remotely called by a client, or can be deployed on a client with sufficient device capability for direct calling. In some embodiments, when it runs on a client, its corresponding intelligence can be obtained through transfer learning to reduce the requirement for client hardware running resources and avoid excessive occupation of client hardware running resources.

[0024] Unless explicitly specified, the various data involved in the present application can be stored remotely on a server or stored locally on a terminal device, as long as it is suitable for being called by the technical solutions of the present application.

[0025] Those skilled in the art should know that the various methods of the present application, although based on the same concept and described to present commonality among them, are independently executable unless otherwise specified. Similarly, for each embodiment disclosed in the present application, it is based on the same inventive concept, and therefore, for the same concept of expression, and although the concept of expression is different, it is only for the convenience of appropriate transformation of the concept. The concept should be understood as equivalent.

[0026] Unless explicitly indicated that the embodiments disclosed in the present application are mutually exclusive, the technical features involved in each embodiment can be combined flexibly to construct new embodiments, as long as such combination does not deviate from the spirit of the present application and can meet the needs of the prior art or solve some deficiencies in the prior art. For this variation, those skilled in the art should know.

[0027] Please refer to Figure 1 , Figure 1 The basic flowchart of the business handling method of the present embodiment is shown in the figure.

[0028] As Figure 1As shown, including: S1100, receiving a user service request, extracting service information containing user identity, target service and associated conditions from the service request; The embodiment is applied to the login scene of user terminals in various fields such as finance, insurance, e-commerce, medical treatment, and communication services. In the embodiment, a service system arranged with an AI model (such as a large language model) is configured to monitor user service requests. First, a user service request is received, and service information containing user identity, target service, and associated conditions is extracted from the service request. Specifically, the user submits a service handling request through various channels (such as a web page, a mobile application, a customer service system, etc.), and the system receives the request through an API interface or a message queue. The request content is transmitted in JSON or XML format, containing user identity information (such as a mobile phone number, an ID number), a target service (such as a package change, a service activation), and associated conditions (such as a contract period, a user status, etc.). The system extracts user identity, target service, and associated conditions from the request through natural language processing (NLP) technology or rule matching algorithm. For example, if the user request is "change my package from A package to B package", the system extracts user identity (such as a mobile phone number), target service (package change), and associated conditions (current package is A package) through named entity recognition (NER). The extracted information will be further standardized, such as mapping natural language description (such as "package change") to system recognizable code (such as "package_change"), to ensure the uniformity and standardization of the information. The system checks the received service information to ensure its validity and integrity. For example, verify whether the user identity is legal, whether the target service is within the system support range, and whether the associated conditions meet the business rules, etc. If the information is incomplete, the system will prompt the user to complete the information through a pop-up window or a short message. In some scenarios, the system may use historical data in the existing business system to supplement the missing information in the request, such as the user's current contract status or product usage record.

[0029] It should be noted that in the voice interaction scenario, the user submits a service request through a voice assistant (such as a customer service robot). The system converts the voice signal into text through voice recognition technology (ASR) and analyzes the service information through a natural language understanding (NLU) module. For example, the user says "I want to cancel my video membership" in voice, and the system converts the voice into text after voice recognition, extracts user identity (account information bound through voice recognition), target service (cancel video membership), and associated conditions (whether the user has subscribed to video membership currently). If the information in the user's voice is incomplete, the system can guide the user to supplement the information through multiple rounds of dialogue, such as asking "Which account's video membership do you want to cancel?" or "Your video membership will expire next week. Do you want to cancel it immediately?".

[0030] It should be noted that in another embodiment of the present embodiment, when the user submits a service request, an image (such as an ID photo) or a document (such as a service application form) can be uploaded. The system extracts the text in the image through optical character recognition (OCR) technology, and analyzes the service information through natural language processing technology. For example, the service application form uploaded by the user contains user identity information (ID number, name), target service (such as account opening application), and associated conditions (such as account type, initial deposit amount). After the system identifies the text through OCR, it uses named entity recognition technology to extract key information and standardize it into a format that the system can recognize. If the uploaded image or document is unclear, the system will prompt the user to re-upload or supplement the information through manual review.

[0031] It should be noted that in another embodiment of the present embodiment, in a complex scenario, the user can submit a service request through multiple modalities, such as uploading voice, image, and text description at the same time. The system integrates these information through multi-modal fusion technology. For example, the user submits "I want to open an international roaming service" through voice, and uploads a passport photo (for identity verification) and a picture of the target country (for service configuration). The system extracts the target service through voice recognition, extracts the user's identity through OCR recognition of the passport information, and analyzes the picture of the target country (such as identifying the country through the national flag or landmark). The system integrates the information of multiple modalities, extracts the complete service information, and checks to ensure the consistency and accuracy of the information. For example, if the country mentioned in the voice is inconsistent with the country identified in the picture, the system will prompt the user to reconfirm or supplement the information.

[0032] S1200, based on the service request, retrieving service rules, mutual exclusion relationships and constraint conditions related to the service information in a pre-built service knowledge graph; After receiving a user service request and extracting service information including user identity, target service, and associated conditions from the service request, based on the service request, the service rules, mutual exclusion relations, and constraint conditions related to the service information are retrieved in the pre-constructed service knowledge graph. Specifically, the service knowledge graph is constructed by integrating service rules, contract terms, product specifications, and other multi-source data. The nodes in the graph represent business entities (such as products, contracts, user states, etc.), and the edges represent the relationships between entities (such as mutual exclusion relations, dependency relations, etc.). The graph is stored in a graph database (such as Neo4j) to support efficient graph structure queries. The extracted service information (such as user identity, target service, and associated conditions) is converted into semantic vectors through natural language processing techniques. Graph Convolutional Networks (GCN) are used to embed the nodes in the knowledge graph, and the semantic similarity between the service information vectors and the graph node vectors is calculated to quickly locate relevant business rules and constraint conditions. Based on the service information, a subgraph retrieval algorithm is used to find matching subgraph structures in the knowledge graph. The algorithm combines shortest path search and adjacency matrix calculation to ensure that all rules and constraints related to the business are covered. Through path analysis, mutual exclusion relations and dependency relations that may exist in the business handling process are identified, providing comprehensive context information for subsequent reasoning.

[0033] It should be noted that in another embodiment of the present embodiment, in a dynamic business scenario, the knowledge graph needs to be updated in real time to reflect the latest business rules and constraint conditions. The system receives business rule change events (such as new product launch, contract term update, etc.) through a message queue and updates the knowledge graph in real time. When retrieving business rules, the system prefers to use the latest version of the graph data to ensure the timeliness and accuracy of the retrieval results. For example, when a user applies for a newly launched business package, the system can immediately retrieve the relevant rules and constraint conditions of the package without waiting for the regular update of the graph.

[0034] It should be noted that the present embodiment considers the business handling history and preferences of different users, and the retrieval process can be personalized in combination with user portraits. The system analyzes the user's historical handling records, preference settings, and other information to construct a user behavior graph and integrate it with the business knowledge graph. When retrieving business rules, the handling path that meets the user's historical behavior pattern is preferentially recommended. For example, if a user frequently handles a certain type of business, the system will preferentially retrieve rules related to that type of business and adjust the recommendation order according to the user's preferences.

[0035] S1300, input the service rules, mutual exclusion relations, and constraint conditions into the local fine-tuned large model, and generate a business handling path containing conflict judgment and processing strategy through thought chain reasoning; After retrieving the business rules, mutual exclusion relationships, and constraint conditions related to the business information in the pre-built business knowledge graph based on the business request, the business rules, mutual exclusion relationships, and constraint conditions are input into a locally fine-tuned large model to generate a business handling path containing conflict judgment and processing strategy through thought chain reasoning. Specifically, the pre-trained large model is fine-tuned using industry-specific data to reduce computational power consumption and enhance the model's understanding of business domain knowledge. Fine-tuning data includes business handling logs, failed cases, artificial work orders, contract terms, etc., to ensure that the model can accurately handle complex business rules and constraint conditions. During the reasoning process, the model uses the Chain-of-Thought (CoT) mechanism to gradually decompose problems and explicitly output the logical deduction process at each step. For example, the model will first analyze the user's target business and current state, then determine whether there is a conflict (such as mutual exclusion relationship) based on the retrieved rules, and then generate a strategy to resolve the conflict (such as canceling existing business or adjusting contract conditions). Each step of reasoning results is recorded to form a detailed reasoning path. The final generated reasoning result is output in a structured form, including conflict judgment (such as whether there is a mutual exclusion relationship), processing strategy (such as how to resolve the conflict), and specific business handling steps (such as retaining business X, canceling business Y, and adding business Z). This structured output provides a foundation for generating executable structured operation instructions in the future.

[0036] It should be noted that, in order to improve the accuracy and success rate of reasoning, the system can introduce reinforcement learning technology, especially Reinforcement Learning with Human Feedback (RLHF). After each business handling is completed, the system will collect feedback from users or business experts (such as whether the handling is successful, whether the path is reasonable, etc.), and adjust the model's reasoning strategy based on the feedback. For example, if a certain reasoning path leads to failure, the system will reduce the weight of that path; if a certain path is frequently marked as successful, the system will increase its weight. In this way, the model can continuously optimize the reasoning path and improve the success rate of business handling.

[0037] It should be noted that the embodiment also aims at the fact that, in some complex scenarios, only relying on the thinking chain reasoning of the large model may not be enough to cover all business rules and constraint conditions. Therefore, the system can combine the graph reasoning capability of the knowledge graph to construct a hybrid reasoning mechanism. Specifically, the business rules and constraint conditions in the knowledge graph can be converted into logical rules (such as Prolog rules) and jointly analyzed with the thinking chain reasoning result of the large model. For example, graph reasoning can quickly identify direct conflicts between the current state of the user and the target business, while the thinking chain reasoning of the large model can further analyze how to solve these conflicts. In this way, the system can more comprehensively cover business rules, and improve the accuracy and reliability of reasoning.

[0038] S1400, generating structured operation instructions recognizable and executable by a business system according to the business handling path; After inputting the business rules, mutual exclusion relationship and constraint conditions into the locally fine-tuned large model, and generating a business handling path containing conflict judgment and processing strategy through thinking chain reasoning, structured operation instructions recognizable and executable by a business system are generated according to the business handling path. Specifically, the business handling path generated by thinking chain reasoning is usually output in natural language form, such as “reserve business X, cancel business Y, add business Z”. The system parses the reasoning result through natural language processing technology, extracts key operations (such as reservation, cancellation, addition) and target objects (such as business X, business Y, business Z). The parsed result is standardized into a unified intermediate representation form, such as JSON or XML. The system predefines a set of operation instruction templates, covering common business operations (such as opening a business, changing a package, canceling a service, etc.). According to the parsed reasoning result, the system matches the corresponding instruction template, and fills the target object and operation parameters into the template. For example, if the reasoning result is “cancel business Y”, the system will match the “cancel business” template and generate a specific SQL statement or API call instruction. The generated structured instructions are verified by a verification module to ensure that they meet the requirements and specifications of the business system. The verification includes syntax checking, parameter legality verification and business logic consistency verification. If the instructions have errors or potential risks, the system will automatically correct or prompt manual intervention. In addition, the system improves the execution efficiency of the instructions through optimization algorithms (such as merging repeated operations, reducing redundant steps). According to the interface requirements of the target business system, the system converts the generated instructions into the corresponding format (such as RESTful API, SOAP message, SQL statement, etc.). For example, if the target system is the core business system of a bank, the generated instructions may be sent in XML format through SOAP protocol; if the target system is an Internet business platform, the generated instructions may be sent in JSON format through HTTP POST request.

[0039] It should be noted that in order to improve the flexibility and efficiency of instruction generation, the system can use a template engine (such as Velocity or Thymeleaf). The template engine allows dynamic instruction templates to be defined, supporting variable substitution and conditional logic. For example, the template can define "if the user is a new user, generate an opening account instruction; if it is an old user, generate a business change instruction". Through the template engine, the system can quickly generate instructions that meet different business scenarios, while reducing the complexity of hard coding.

[0040] It should be noted that for complex business handling paths (such as operations involving multiple steps or multiple systems), the system of the present embodiment combines a workflow engine (such as Camunda or Activiti) to process instructions collaboratively. The workflow engine decomposes the generated instructions into multiple sub-tasks and executes them in a predefined order. For example, first execute the "cancel business Y" instruction, then execute the "add business Z" instruction. The execution result of each sub-task is fed back to the workflow engine to dynamically adjust the execution logic of the subsequent tasks. In this way, the system can handle complex business processes, ensuring the ordered execution of instructions and the complete closed loop of business.

[0041] S1500, send the operation instruction to the business system, prompting the business system to execute the business corresponding to the operation instruction.

[0042] After generating the structured operation instructions that can be recognized and executed by the business system according to the business handling path, the operation instructions are sent to the business system to prompt the business system to perform the business corresponding to the operation instructions. Specifically, according to the interface specification of the business system, the system converts the structured operation instructions into an adaptive communication protocol and data format. Common protocols include HTTP / HTTPS (for RESTful API), SOAP (for traditional enterprise systems), TCP / IP (for systems with low latency requirements), and the like. For example, if the target system is a bank core system, the instructions can be sent in XML format through the SOAP protocol; if it is an Internet business platform, the instructions can be sent in JSON format through the HTTP POST request. After the business system receives the instructions, it parses and executes the operation. The execution result (such as success, failure, or partial success) is fed back to the intelligent business handling system in real time through a callback interface or a message queue. For example, if the instruction is "open business X", the business system will return a status code (such as 200 for success and 400 for failure) and detailed information (such as the reason for failure) after execution. The intelligent business handling system processes the received execution result. If the operation is successful, the system records the log and notifies the user; if the operation fails, the system triggers a rollback mechanism (such as canceling the partially executed operation) and records the failure reason. Failed cases will enter a feedback loop to optimize the knowledge graph and large model inference path. For example, if an instruction fails due to a change in business rules, the system will update the relevant rules in the knowledge graph and retrain the model to avoid similar problems. In scenarios involving multiple business systems, the system uses a distributed transaction management mechanism (such as two-phase commit, 2PC) to ensure the atomicity and consistency of the operation. For example, if the business handling path involves a bank system and a third-party payment platform, the system will use a transaction coordinator to ensure that the operations of both systems are completed synchronously or rolled back synchronously. To reduce transaction blocking, the Saga mode can also be used to split complex transactions into multiple simple transactions, which are submitted gradually and support compensation operations.

[0043] It should be noted that in high-concurrency or low-latency scenarios, the system uses a message queue (such as Kafka, RabbitMQ) for asynchronous instruction execution. Specifically, the generated structured instructions are sent to the message queue, and the business system acts as a consumer to obtain the instructions from the queue and execute them asynchronously. This mode decouples the instruction generation and execution processes, improving the throughput and stability of the system. For example, in the scenario of changing a communication operator's package, asynchronous execution can avoid the accumulation of user requests and improve user experience.

[0044] It should be noted that in another extended embodiment of the present embodiment, for business scenarios with high security and traceability requirements (such as financial services or supply chain management), the system combines blockchain technology to perform and audit instructions. The generated structured instructions and their execution results are recorded on the blockchain, ensuring data tamper-proofing and transparency. For example, cross-border payment services between banks can record the sending, receiving, and execution status of instructions through the blockchain, providing real-time audit capabilities. At the same time, the blockchain smart contract can automatically verify the compliance of the instructions and automatically execute when the conditions are met, reducing manual intervention.

[0045] In an actual application scenario, in the field of communication operators, when a user applies for a package change through a mobile service hall, this method can quickly extract business information such as user identity, target package, and current contract status, and retrieve relevant rules and constraints in the business knowledge graph, such as contract period, mutually exclusive services, etc. The locally fine-tuned large model generates a business handling path that resolves conflicts through thought chain reasoning, such as canceling mutually exclusive services before changing the package. The system converts this path into structured operation instructions and sends them to the operator core system for execution, completing the package change and providing real-time feedback, improving handling efficiency and user experience, and reducing manual intervention costs.

[0046] In the above-mentioned embodiments, by automatically extracting business information and retrieving relevant rules in the knowledge graph, the locally fine-tuned large model generates a business handling path that resolves conflicts through thought chain reasoning, effectively solving the problem of knowledge update lag in existing business handling systems and the inability to cover complex rules and mutually exclusive relationships. This method realizes the full-process intelligence of business handling, reduces manual intervention, reduces operating costs, and improves handling efficiency and success rate. At the same time, the business handling path is converted into structured operation instructions and sent to the business system for execution, ensuring the accuracy and timeliness of the operation. In addition, the locally fine-tuned large model enhances the understanding of industry knowledge, and the thought chain reasoning improves the transparency and explainability of the reasoning process, making the business handling more reliable and controllable.

[0047] In some embodiments, the S1200 retrieves business rules, mutually exclusive relationships, and constraints related to the business information in a pre-built business knowledge graph, including: S1211, using a graph convolutional neural network to optimize the embedding representation of nodes in the pre-built business knowledge graph, generating node vectors containing semantic information; In the embodiment, when searching for business rules, mutual exclusion relationships and constraint conditions related to the business information in the pre-built business knowledge graph, a graph convolutional neural network is used to optimize the embedding representation of the nodes in the pre-built business knowledge graph to generate node vectors containing semantic information. Specifically, the business knowledge graph is built by integrating multiple sources of data such as business rules, contract terms, product specifications, etc., and contains business entities (such as products, contracts, user states, etc.) and relationships between entities (such as mutual exclusion relationships, dependency relationships, etc.). The graph is stored in a graph database (such as Neo4j) to support efficient graph structure queries. A graph convolutional neural network (GCN) is used to optimize the embedding representation of the nodes in the graph. GCN generates node vectors containing semantic information by aggregating the feature vectors of the node itself and its neighbor nodes. The specific process is as follows: Initialize node feature vectors: assign an initial feature vector to each node to represent the basic attributes of the node (such as name, type, etc.); Graph convolution operation: through multiple layers of graph convolution operations, the feature vectors of the node and its neighbor nodes are aggregated to update the embedding representation of each node; Generate node vectors: after multiple layers of graph convolution operations, the final embedding vector of each node contains its own semantic information and the context information of its neighbor nodes.

[0048] The generated node vectors are stored in a vector database (such as Faiss or Milvus) to support efficient similarity search. Each node vector is associated with the corresponding node ID to facilitate quick positioning of the node in subsequent retrieval processes.

[0049] It should be noted that in the scenario where business rules are frequently updated, a dynamic updating node embedding optimization scheme can be used. The system receives business rule change events (such as new product launch, contract term update, etc.) in real time through a message queue and updates the nodes and relationships in the knowledge graph in real time. After each update, the embedding vectors of the relevant nodes are recalculated and the updated vectors are stored in the vector database. This method ensures that the node vectors can reflect the latest business rules in real time, improving the timeliness and accuracy of the retrieval.

[0050] S1212, based on the node vectors, calculate the semantic similarity between the business request and the graph nodes, and locate the initial associated node set; After the node embedding representation optimization of the pre-constructed business knowledge graph using the graph convolutional neural network is performed, and the node vector containing semantic information is generated, the semantic similarity between the business request and the graph node is calculated based on the node vector, and the initial associated node set is located. Specifically, the extracted business information (such as user identity, target business, associated condition) is converted into a semantic vector through natural language processing technology. Specifically, a pre-trained language model (such as BERT) is used to encode the business information to generate a fixed-dimensional vector representation. For example, if the business request is "change my package from A package to B package", the model will encode it into a high-dimensional vector. The semantic similarity between the business request vector and the node vector in the knowledge graph is calculated. Common similarity calculation methods include cosine similarity, Euclidean distance, etc. According to the calculated semantic similarity, the most relevant nodes to the business request are selected to form the initial associated node set. Usually, a similarity threshold (such as 0.7) is set, and the nodes with a similarity higher than the threshold are included in the initial associated node set. In addition, business rules (such as node type, node importance) can also be combined for further screening to ensure the accuracy and relevance of the initial node set. The nodes in the initial associated node set are sorted according to the similarity, and the node with the highest similarity is selected first. At the same time, the node set can be further optimized through clustering analysis or rule filtering to remove redundant nodes or nodes not directly related to the business target. For example, if multiple nodes represent similar business rules, only the node with the highest similarity is retained.

[0051] S1213, taking the initial associated node set as the starting point, performing subgraph retrieval based on shortest path and adjacency matrix analysis to extract subgraph structure containing complete rule chain and mutual exclusion relationship; After calculating the semantic similarity between the service request and the graph nodes based on the node vectors and locating the initial set of associated nodes, the system performs subgraph retrieval based on shortest path and adjacency matrix analysis, starting from the initial set of associated nodes. This step aims to extract subgraph structures that contain complete rule chains and mutual exclusion relationships. To achieve this, the system uses a graph traversal algorithm such as Breadth-First Search (BFS) to expand the node set from the initial set of associated nodes. The BFS algorithm starts from the initial node and iteratively explores its neighbor nodes until it reaches a pre-defined depth limit (e.g., depth 3). This step aims to cover the business rules, mutual exclusion relationships, and constraints related to the initial node. The system then calculates the shortest paths between the initial node and other nodes in the graph using Dijkstra's algorithm or A* algorithm. These paths represent direct or indirect relationships between business rules and constraints. The calculation of shortest paths helps identify the most critical rule chains and mutual exclusion relationships in the business handling process. The system constructs an adjacency matrix for the expanded node set, representing the connection relationships and weights between nodes. The weights can represent the strength or importance of relationships between nodes. Through matrix operations such as matrix multiplication, the system analyzes the reachability and relationship strength between nodes to ensure that the extracted subgraph structures are complete and consistent with business logic. Based on the results of shortest path and adjacency matrix analysis, the system extracts subgraph structures that contain complete rule chains and mutual exclusion relationships. The subgraph extraction process ensures that all relevant nodes and edges are included. The extracted subgraph is checked by a verification module to ensure that it meets business rules and constraints. For example, the verification module checks whether the subgraph covers all key mutual exclusion relationships and dependency relationships. The extracted subgraph is optimized to remove redundant nodes and edges, ensuring that the subgraph structure is concise and efficient. The optimized subgraph is output in the form of a graph structure or a list, providing a basis for subsequent steps of retrieving business rules, mutual exclusion relationships, and constraints. Through the above steps, the system can extract subgraph structures highly related to the business request from the knowledge graph, ensuring that the generated business handling path is complete and consistent with business logic.

[0052] S1214、From the subgraph structure, retrieve business rules, mutual exclusion relationships, and constraints related to the business information.

[0053] After performing subgraph retrieval based on shortest path and adjacency matrix analysis starting from the initial set of associated nodes, the business rules, mutual exclusion relationships, and constraint conditions related to the business information are retrieved from the subgraph structure. Specifically, the extracted subgraph structure is parsed to obtain node and edge information. Nodes represent business entities (such as products, contracts, user states, etc.), and edges represent relationships between entities (such as mutual exclusion relationships, dependency relationships, compatibility relationships, etc.). During the parsing process, the system extracts the attributes of each node (such as name, type, state) and the attributes of each edge (such as relationship type, weight). Using a rule matching algorithm, the business information (such as user identity, target business, associated conditions) is matched with the nodes and edges in the subgraph. The matching algorithm is based on business rule templates (such as "Product A and Product B are mutually exclusive") and conditional logic (such as "the user's current state is within the contract period"). Through matching, the system filters out the nodes and edges directly related to the business information. By traversing the edge relationships in the subgraph, rules related to mutual exclusion relationships and constraint conditions are extracted. For example, if there is an edge in the subgraph representing "Product A and Product B are mutually exclusive", the system will record it as a mutual exclusion relationship. Similarly, if there is an edge representing "the handling of Product C requires the cancellation of Product D", the system will record it as a constraint condition. The extracted business rules, mutual exclusion relationships, and constraint conditions are combined into a rule chain. The rule chain represents the order and conditions that need to be followed in the business handling process. For example, the rule chain may be "cancel Product D first, then handle Product C". The generated rule chain is checked by a verification module to ensure that it meets the business logic and rule requirements. The retrieved business rules, mutual exclusion relationships, and constraint conditions are output in a structured format, such as JSON or XML. The output results include the type of rules (such as mutual exclusion relationships, dependency relationships), the entities involved (such as Product A, Product B), and the conditions of the rules (such as user state). At the same time, the system improves the simplicity and readability of the output results through optimization algorithms (such as removing redundant rules, merging similar rules).

[0054] This embodiment optimizes node embedding through graph convolutional neural networks, improving the accuracy of semantic similarity calculation in business knowledge graphs, and thus more accurately locating rules and conditions related to business requests. Its benefits include: first, enhancing the accuracy of retrieval to ensure that the generated business handling path meets the actual business rules; second, improving the system's understanding of complex business scenarios and reducing handling errors caused by rule omissions or misjudgments; third, through effective semantic analysis and subgraph retrieval, speeding up the response speed of business handling and improving user experience. In addition, this method can also adapt to changes in business rules, update the knowledge graph and model parameters to maintain the long-term effectiveness and adaptability of the system.

[0055] In some embodiments, the S1300 generates a business handling path containing conflict judgment and processing strategy through thought chain reasoning, including: S1311, based on the retrieved constraint conditions, gradually reason and explicitly output business conflict detection logic and potential risk warning; In this embodiment, when generating a business handling path containing conflict judgment and processing strategy through thought chain reasoning, based on the retrieved constraint conditions, gradually reason and explicitly output business conflict detection logic and potential risk warning. Specifically, the system analyzes the retrieved constraint conditions and extracts key information (such as business rules, mutual exclusion relationships, dependency relationships, etc.). According to the nature and influence range of the constraint conditions, they are classified into business conflict category (such as mutually exclusive business), compliance category (such as regulatory restrictions), resource limitation category (such as insufficient inventory), etc. For example, if the constraint condition is "the user cannot handle business A and business B at the same time", it will be classified as a business conflict category. Initialize the thought chain reasoning process, set the initial state (such as the user's current business state, target business state) and reasoning target (such as detecting conflicts, generating handling path). The reasoning process uses the step-by-step problem decomposition method, and each step of reasoning records the detailed logical derivation process. For example, the initial state is "the user has handled business A", and the target state is "the user wants to handle business B", then the reasoning process starts from detecting the relationship between business A and business B. According to the classified constraint conditions, the business conflict is detected step by step. For each constraint condition, the reasoning model will judge whether it conflicts with the business request and record the reasoning logic. For example, if it is detected that business A and business B are mutually exclusive, the reasoning process will record "business A and business B are mutually exclusive, the user has handled business A, so directly handling business B will cause a conflict". Each step of logic in the reasoning process is explicitly output in natural language form to ensure the transparency and interpretability of the reasoning process. While detecting business conflicts, the system will assess potential risks and generate warning information. Risk assessment is based on historical data and business rules, considering the impact of conflicts on user state, contract terms, and business processes. For example, if canceling business A will cause the user to lose certain rights, the system will warn "canceling business A will cause the user to lose right X, please confirm whether to continue". The warning information includes risk type (such as loss of rights, compliance risk), impact range (such as user state, contract state) and suggested measures (such as alternative solutions). The detected business conflicts and potential risk information are integrated into structured output, including conflict type, conflict reason, risk assessment result and suggested measures. The output result is presented in natural language form to ensure that users and business personnel can clearly understand. For example, the output result may be "detect business conflict: business A and business B are mutually exclusive, the user has handled business A. Potential risk: canceling business A will cause the user to lose right X. Suggested measures: handle business C first, then decide whether to handle business B according to the state of business C".

[0056] It should be noted that the embodiment combines user portrait for personalized analysis when detecting business conflict. The user portrait includes information such as user's historical behavior, preference settings, business usage habits, etc. The system adjusts the priority and strategy of conflict detection according to the user portrait. For example, if the user often handles a certain type of business, the system will prioritize the detection of conflicts related to this type of business and generate more detailed reasoning paths and risk warnings. In addition, the system can recommend personalized solutions based on the user portrait, such as "based on your usage habits, we recommend that you handle business C first, and then upgrade business A to avoid conflicts and achieve a better experience".

[0057] S1312, for the identified conflict, simulate multiple processing strategies and evaluate their cascading effects on user state and contract terms; Further, after inferring and explicitly outputting the business conflict detection logic and potential risk warning based on the retrieved constraints, the embodiment simulates multiple handling strategies for the identified conflicts and evaluates their cascading effects on user status and contract terms. Specifically, the system first analyzes the identified conflicts in detail, determining the type of conflict (such as mutual exclusion, dependency, resource limitation, etc.) and the scope of impact (such as business status, user rights, contract terms, etc.). Based on the analysis results, multiple possible handling strategies are generated. For example, if the conflict is "business A and business B are mutually exclusive," possible handling strategies include canceling business A and proceeding with business B, adjusting the configuration of business A to be compatible with business B, and recommending an alternative business C. Each handling strategy is simulated and executed through thought chain reasoning to predict its cascading effects on user status and contract terms. During the simulation process, the system considers factors such as the user's current state, business rules, and contract conditions. For example, when simulating the cancellation of business A, the reasoning process records that "canceling business A will result in the user losing right X, but can meet the conditions for proceeding with business B." The impact of each handling strategy is evaluated, including its impact on user status (such as changes in rights, changes in business status), contract terms (such as changes in fees, adjustments to contract duration), and business processes (such as step addition or reduction, time delay). Impact evaluation can be quantified as a score or weight, such as "strategy 1 results in a right loss score of -2, and strategy 2 results in a fee increase score of -1." Based on business rules and user profiles, the system performs a risk-benefit trade-off analysis for each handling strategy. For example, some strategies may be able to solve the conflict, but may result in user rights loss or fee increase; while other strategies may require additional steps but provide better user experience. The system recommends the optimal strategy through a rule engine or machine learning model based on historical data and business goals. Based on the evaluation results, the handling strategy is optimized and specific recommendations are generated. For example, if the evaluation results of strategy 1 and strategy 2 are similar, but strategy 2 has less impact on user rights, the system will preferentially recommend strategy 2. The recommendations include specific business operation steps, expected results, and risk warnings, presented in natural language to ensure that users and business personnel can clearly understand.

[0058] S1313、Integrate the evaluation results to generate an interpretable business handling path containing a specific business operation sequence.

[0059] Further, the embodiment simulates multiple handling strategies for the identified conflicts, evaluates their cascading effects on user status and contract terms, integrates the evaluation results, and generates an interpretable business handling path containing specific business operation sequences. Specifically, the system collects the generated reasoning results, including business conflict detection logic, potential risk warnings, multiple handling strategies, and their impact evaluations. These results are organized into structured data to ensure information integrity and consistency. According to the reasoning results and handling strategy evaluation, specific business operation sequences are generated. The operation sequences are arranged in logical order to ensure that each operation can effectively solve the conflict and promote business handling. For example, the operation sequence may be "cancel business A, then handle business B, and finally open business C." Detailed explanations and descriptions are generated for the generated business handling path, including the reason for each operation, expected effect, and potential impact. The explanation content is presented in natural language form to ensure that users and business personnel can understand. For example, the explanation may include "canceling business A is because it is mutually exclusive with business B, handling business B can meet your needs, and opening business C is to provide a better service experience." The generated business handling path is optimized to remove redundant steps, combine similar operations, and ensure the path is concise and efficient. Through simulation execution or rule verification, the feasibility and compliance of the path are ensured. For example, verify whether the path covers all key conflicts and whether it meets business rules and contract terms. The optimized business handling path is output in a user-friendly manner, such as through a web page, mobile application, or customer service system, to users. The output content includes operation steps, explanations, expected results, and risk warnings. At the same time, interactive guidance is provided to help users complete business handling step by step.

[0060] The embodiment generates an interpretable business handling path by simulating multiple handling strategies and integrating reasoning results, which has significant beneficial effects. It can effectively identify business conflicts, provide early warning of potential risks, and provide key support for decision-making. At the same time, by evaluating the impact of different strategies, the system can recommend the optimal business handling scheme, improving the success rate of handling and user experience. In addition, the generated interpretable path is easy for users and business personnel to understand, enhancing the transparency and acceptability of business handling, and improving the efficiency and quality of overall business handling.

[0061] In some embodiments, S1400 generates structured operation instructions that can be recognized and executed by a business system according to the business handling path, including: S1411, parse the business handling path and convert each atomic operation step in it into corresponding system call logic; In this embodiment, when generating structured operation instructions that can be recognized and executed by the business system according to the service handling path, the system also parses the service handling path and converts each atomic operation step therein into corresponding system call logic. Specifically, the system parses the generated service handling path sentence by sentence and extracts the key information of each atomic operation step. The parsing process identifies the operation type (such as "cancel", "add", "change", etc.) and the target object (such as "service A", "package B", etc.) through natural language processing techniques (such as dependency syntax analysis). For example, "cancel service A" in the path is parsed as operation type "cancel" and target object "service A". A mapping relationship between operation steps and system call logic is established. The system predefines a set of mapping rules to map common business operations (such as "cancel service", "add service", "change package", etc.) to corresponding system call logic (such as API interface name, SQL operation type, etc.). For example, the call logic of "cancel service A" may correspond to calling the "cancel_service" interface of the business system and passing in the identifier of service A. According to the parsing result, the parameter information (such as service ID, user ID, operation time, etc.) of each operation step is extracted and filled into the corresponding system call logic. Parameter extraction is achieved through regular expressions or named entity recognition techniques. For example, if the path contains "cancel service A for user 12345", the user ID is extracted as "12345" and the service ID is extracted as "A", which are filled into the parameters of the "cancel_service" interface. The generated system call logic is verified to ensure that it meets the requirements and specifications of the business system. Verification includes interface existence check, parameter legality verification, permission check, etc. If the call logic has errors or potential risks, the system will automatically correct or prompt manual intervention. In addition, the system improves the execution efficiency of the call logic through optimization algorithms (such as merging repeated calls, reducing redundant parameters). The system call logic that passes the verification is converted into a unified intermediate representation form (such as JSON or XML) for further processing and generation of specific structured instructions in subsequent steps. The intermediate representation form ensures the compatibility and scalability of the call logic between different systems and modules.

[0062] S1412, mapping the system call logic into specific structured instructions that can be directly executed by the business system, the structured instructions including API call commands, SQL statements or workflow tasks; Further, after parsing the service handling path and converting each atomic operation step into corresponding system call logic, the system call logic is mapped to concrete structured instructions that can be directly executed by the service system, including API call commands, SQL statements, or workflow tasks. Specifically, the system predefines a set of instruction templates covering common business operations (e.g., opening a service, changing a package, canceling a service, etc.). Each template contains a fixed syntax structure and placeholders for filling in specific parameter values. According to the generated system call logic, the corresponding parameter values (e.g., service ID, user ID, etc.) are extracted and filled into the placeholders in the instruction templates. For business steps involving database operations, the system maps the system call logic to SQL statements. For complex business processes, the system maps the system call logic to workflow tasks. For example, if the logic is "submit a service change application and wait for approval", a workflow task is generated, including task type (e.g., approval task), task parameters (e.g., applicant, change content, etc.), and task flow rules (e.g., execute the next step after approval). According to the interface specifications of the business system, the generated instructions are converted to the appropriate format and protocol. For example, API call instructions may need to be converted to JSON format and sent through HTTP POST, SQL statements may need to be encapsulated in transaction blocks for execution, and workflow tasks may need to be triggered through a message queue. The generated structured instructions are checked to ensure they meet the syntax and semantic requirements of the business system. The check includes syntax correctness check, parameter legality verification, business logic consistency verification, etc. At the same time, optimization algorithms (such as merging repeated instructions, reducing redundant operations) are used to improve the execution efficiency of the instructions.

[0063] S1413, according to the business logic, the structured instructions are sorted and dependency checked to form an ordered operation instruction sequence.

[0064] Further, after mapping the system call logic into specific structured instructions that can be directly executed by the business system, including API call commands, SQL statements, or workflow tasks, the structured instructions are sorted and dependency checked according to the business logic to form an ordered operation instruction sequence. Specifically, the system parses the generated structured instructions one by one, extracts the operation object, operation type, and related business attributes of each instruction. By analyzing the dependency relationship between instructions (such as data dependency, execution order dependency, etc.), an instruction dependency graph is constructed. For example, if instruction A needs to update the user state and instruction B needs to query the state, instruction A must be executed before instruction B. According to the instruction dependency graph, a topological sorting algorithm is used to determine the execution order of the instructions. Topological sorting ensures that all instructions are ordered according to the dependency relationship, avoiding situations where dependencies are not met during execution. For instructions without dependency relationships, they can be arranged in parallel or series according to business priority or system load. In the sorted instruction sequence, the dependency relationship between instructions is checked one by one to see if it meets the requirements. Check if there is a circular dependency or unresolved dependency conflict. For example, if instruction A depends on instruction B, and instruction B depends on instruction A, the business logic needs to be adjusted or the instructions need to be split to solve the circular dependency problem. For complex operations involving multiple business systems, a distributed transaction management mechanism (such as two-phase commit, 2PC) or Saga mode is used to ensure the atomicity and consistency of the operation. Transaction control instructions are added to the instruction sequence to ensure that if a certain link fails, the executed operations can be rolled back. For example, in a bank transfer scenario, if the deduction is successful but the deposit fails, the rollback mechanism is triggered to cancel the deduction operation. According to the system resource status and business logic, the instruction sequence is optimized and scheduled. For example, instructions without dependency relationships are executed in parallel to improve execution efficiency, or the priority of instruction execution is dynamically adjusted according to system load. At the same time, the instruction sequence is merged or split to reduce unnecessary network communication and system overhead.

[0065] The present embodiment realizes the automation and efficient execution of business handling by analyzing the business handling path and generating structured operation instructions. The beneficial effects include: 1. The business handling path is converted into system recognizable instructions to ensure that the business system can directly execute and improve the handling efficiency; 2. The instructions are sorted and dependency checked to form an ordered operation sequence, ensuring the accuracy and reliability of business handling; 3. Support for various structured instruction forms such as API calls, SQL statements, and workflow tasks makes the system adaptable and expandable, meeting the needs of different business scenarios.

[0066] In some embodiments, the business handling path includes an alternative handling path, and the S1300 further includes: S1321, detecting business conflicts between the user target business and the existing business; In this embodiment, when generating the service handling path, the service conflict between the user's target service and the existing service is also detected. Specifically, first, the service handling request submitted by the user is analyzed to extract the key features of the target service (such as service type, service content, contract period, etc.). At the same time, the information of the existing service currently handled by the user is obtained from the service knowledge graph, including the service type, state, and related contract clauses. The features of the target service and the existing service are matched with the predefined service conflict rules. The conflict rules are stored in the service knowledge graph and cover mutual exclusion relationship, dependency relationship, resource limitation, etc. For example, the rule can define that "service A and service B are mutually exclusive and cannot be handled at the same time". The conflict detection algorithm is used to compare the features of the target service and the existing service to determine whether there is a conflict. The algorithm traverses the related nodes and edges in the service knowledge graph to find the direct or indirect conflict relationship between the target service and the existing service. For example, if there is a mutual exclusion edge between the target service and the existing service in the knowledge graph, it is determined that there is a conflict. The detected conflict is classified to identify the conflict type (such as mutual exclusion relationship, resource conflict, contract limitation, etc.). For example, if the target service requires a specific network resource and the existing service has occupied the resource, it is identified as a resource conflict. The detected conflict information is recorded, including the conflicting service pair, conflict type, impact range, etc. The conflict result is output in a structured format to provide a basis for extracting the mutually exclusive service identifier and contract clauses for the subsequent steps.

[0067] S1322, when it is detected that the user's target service and the existing service have a mutual exclusion relationship, the existing service identifier causing the mutual exclusion and the corresponding contract clause are extracted; Further, after detecting the service conflict between the user target service and the existing service, when it is detected that there is a mutual exclusion relationship between the user target service and the existing service, the existing service identifier causing the mutual exclusion and the corresponding contract clause are extracted. Specifically, in the service knowledge graph, the existing service node with a mutual exclusion relationship with the target service is located through a graph traversal algorithm (such as depth-first search or breadth-first search). The mutual exclusion relationship is usually represented in the form of an edge in the knowledge graph, the type of the edge is “mutual exclusion relationship”, and two business nodes are associated. The business identifier (such as business ID, business name, etc.) is extracted from the located existing service node. The business identifier is a feature that uniquely identifies the business in the business system, and is used for the generation and execution of subsequent operation instructions. For example, if the existing service is “business A”, its business ID “123” is extracted. In the knowledge graph, the contract clause node associated with the mutual exclusion business is queried. The contract clause node usually contains clause content, clause type (such as mutual exclusion clause, fee clause, etc.), effective time, etc. By traversing the outgoing edge and the incoming edge of the mutual exclusion business node, the contract clause node associated with it is found. The clause content and related attributes are extracted from the contract clause node. The extracted content includes clause text, clause type, effective time, expiration time, etc. For example, the extracted clause content may be “the user cannot handle business A and business B at the same time, and needs to cancel business A before handling business B”. The extracted existing service identifier and contract clause content are arranged into a structured data format (such as JSON or XML). The integrity and accuracy of the extraction result are verified to ensure that the identifier and the clause content are consistent with the information in the knowledge graph. For example, it is verified whether the extracted business ID is correct and whether the clause content is complete. The extracted mutual exclusion business identifier and contract clause content are output for use in generating the alternative handling path in the subsequent step. The output result can include business identifier, clause content, clause type, effective time, etc. to ensure that the subsequent steps can accurately understand and apply these data.

[0068] S1323, according to the existing service identifier and the corresponding contract clause, confirming the mutual exclusion business represented by the mutual exclusion relationship, and generating an alternative handling path for canceling the mutual exclusion business and handling the target service.

[0069] Further, when detecting that the user's target service has a mutual exclusion relationship with an existing service, after extracting the existing service identifier and corresponding contract terms that cause the mutual exclusion, the mutual exclusion service represented by the mutual exclusion relationship is confirmed according to the existing service identifier and corresponding contract terms, and a replacement handling path for canceling the mutual exclusion service and handling the target service is generated. Specifically, after detecting the mutual exclusion relationship and extracting the relevant service identifier and contract terms, the system first analyzes the contract terms to determine the prerequisites and operation process for canceling the mutual exclusion service. Then, using the business rule engine, combined with the user's current state and the requirements of the target service, a series of operation steps are generated to form a replacement path. These steps include but are not limited to initiating a request to cancel the mutual exclusion service, waiting for confirmation of the cancellation completion, processing any dependent services or data updates related to the cancellation, and submitting a request to handle the target service after all prerequisites are met. The entire process is implemented through the automatic operation of the corresponding interfaces of the business system, while recording the execution status of each step to facilitate rollback or adjustment of strategies in case of problems. The system also dynamically updates the knowledge graph and business handling model according to the execution results to optimize the processing flow of similar requests in the future, ensuring efficient and accurate resolution of business conflicts and meeting user needs.

[0070] In this embodiment, by detecting business conflicts and generating a replacement handling path, the mutual exclusion problem between the user's target service and existing services is effectively solved. The beneficial effects include: first, automatic identification of mutual exclusion relationships and extraction of relevant services and contract terms provide specific conflict reasons and solutions for users; second, generation of a replacement path that includes canceling the mutual exclusion service and handling the target service improves the flexibility and success rate of business handling; third, reduces manual intervention and handling time, and improves user experience and system efficiency.

[0071] In some embodiments, after S1323 confirms the mutual exclusion service represented by the mutual exclusion relationship according to the existing service identifier and corresponding contract terms, and generates a replacement handling path for canceling the mutual exclusion service and handling the target service, it further includes: S1331: Calculate the cost information of the replacement handling path; In this embodiment, after confirming the mutually exclusive services represented by the mutually exclusive relationship according to the existing service identifier and the corresponding contract terms, and generating the alternative handling path for canceling the mutually exclusive services and handling the target service, the system calculates the cost information of the alternative handling path. Specifically, the system extracts cost-related information from the business knowledge graph and the alternative handling path, including direct costs (such as business change fees, new business opening fees), indirect costs (such as potential penalties for canceling existing services), time costs (such as the time required for business handling), and potential benefits (such as new business discounts or subsidies). These information are stored in the corresponding nodes of the knowledge graph or the database of the business system. A cost calculation model is constructed to quantify different types of costs and benefits into a unified measurement unit (such as monetary amount). The model calculates the sum of each cost according to business rules and financial formulas. Real-time data (such as current exchange rates, promotional activities, system load, etc.) are combined to adjust the cost calculation. For example, if there is a limited-time discount for a new service, the system will dynamically adjust the calculation of potential benefits to ensure the timeliness and accuracy of the cost information. Sensitivity analysis is performed on the cost calculation results to evaluate the impact of different factors (such as cost changes, exchange rate fluctuations) on the total cost. By simulating cost changes in different scenarios, the system provides users with the handling path with the lowest cost or the highest cost-performance ratio. The calculated cost information is organized into a structured data format (such as JSON or XML), including detailed cost breakdown, total cost amount, and detailed explanation of cost composition. The output results will be used in subsequent steps to associate with the alternative handling path and finally included in the structured operation instructions.

[0072] S1332: Establish the association between the alternative handling path and the cost information to include both the alternative handling path and the cost information in the finally generated structured operation instructions.

[0073] In this embodiment, after calculating the cost information of the alternative handling path, the association between the alternative handling path and the cost information is established to integrate them into the finally generated structured operation instruction. Specifically, the alternative handling path exists in the form of a sequence of operation steps, while the cost information records the detailed expenses and total cost in the form of structured data (such as a JSON object or an XML fragment). By attaching the cost information to the data structure of the alternative path, a direct association between the two is established. For example, in the JSON format, the cost information can be attached as a subfield of the alternative path object, ensuring that the correspondence between the path and the cost is clear and easy to access. When generating the structured operation instruction, the integrated alternative handling path and cost information are included. According to the execution requirements of the business system, the operation steps of the path and the cost information are converted into the corresponding instruction format. For example, in API calls, the operation steps can be converted into a series of service requests, and the cost information is included in the request header or payload as metadata; for workflow systems, the path and cost information can be combined into a workflow instance, where the operation steps are task nodes and the cost information is a process variable for reference in the approval or audit link. According to the interface specifications and data requirements of the business system, the instructions containing the path and cost information are format-adapted. For example, if the business system is based on RESTful API, the instructions can be encapsulated as HTTP requests, with the path as a JSON array in the request body and the cost information embedded as an object property; for systems based on message queues, the instructions can be serialized as XML or JSON strings and published to the specified queue. The finally output structured operation instruction contains not only the operation details required to execute the alternative path, but also the cost information to support business decision-making and subsequent financial processing.

[0074] In this embodiment, by calculating the cost information of the alternative handling path and incorporating it into the structured operation instruction, comprehensive decision support is provided for users and business systems. The benefits include: first, users can clearly understand the possible expenses and cost composition when handling business, making more intelligent choices; second, the association of cost information with the handling path facilitates resource scheduling and cost control during the execution of the business system; third, the transparency and explainability of business handling are enhanced, improving user experience and satisfaction.

[0075] In some embodiments, after the 1500 sends the operation instruction to the business system to prompt the business system to execute the corresponding business of the operation instruction, it further includes: S1511, monitoring the execution of the operation instruction; In this embodiment, after sending the operation instruction to the business system and prompting the business system to execute the operation instruction corresponding to the business, the execution of the operation instruction is also monitored. Specifically, after sending the operation instruction to the business system, the system automatically creates a monitoring task, records the unique identifier of the operation instruction, the target business system, the expected execution time, and other key information. The monitoring task is associated with the operation instruction in a one-to-one manner, ensuring that the execution status of each instruction can be accurately tracked. The system monitors the execution status of the instruction in two ways: one is to periodically poll the execution feedback interface of the business system to obtain the current status of the instruction (such as success, failure, execution, etc.); the second is to listen to the execution events (such as instruction completion events, error throwing events, etc.) feedback by the business system through the message queue or event bus. The polling frequency can be dynamically adjusted according to business requirements and system load, usually once every 30 seconds to 5 minutes. The obtained execution status information is analyzed to extract key status codes and description information. The status code usually follows a unified specification (such as HTTP status code or business system custom code), for example, "200" indicates success, "500" indicates server error, and "409" indicates business conflict. The system records the status information in the execution log of the operation instruction, which is convenient for subsequent auditing and problem troubleshooting. According to the change of the execution status, the system real-time feedbacks the instruction execution progress to the user or the business operator. For example, through the front-end page push notification, sending emails or short messages, etc. to inform the user of the "instruction accepted", "execution in progress" or "execution failed" status. For the failure status, a brief failure reason is provided (such as "the business system is busy, please try again later"). When the execution of the instruction fails, the system automatically triggers the error handling process for error analysis and classification processing. If the execution is successful, the final execution result is recorded, the monitoring task is closed, and the related resources are released.

[0076] S1512, when the operation instruction execution fails, the error code returned by the execution failure is analyzed to determine whether it is a temporary system exception or a permanent business rule conflict; Further, after monitoring the execution of the operation instruction, when the operation instruction execution fails, the error code returned by the execution failure is parsed to determine whether it is a temporary system exception or a permanent business rule conflict. Specifically, the system receives the error code and description information returned by the business system and parses them by querying a predefined error code mapping table. The mapping table divides error codes into two categories: temporary exceptions (such as system busy, network timeout, resource unavailable, etc.) and permanent conflicts (such as business rule restrictions, mutual exclusion relationships, contract clause constraints, etc.). For example, error code "503" indicates system busy (temporary exception), while error code "409" indicates business conflict (permanent conflict). According to the parsing result, the system determines the exception type through judgment logic. For temporary exceptions, record the error type as "temporary" and prepare for retry; for permanent conflicts, record the error type as "permanent" and prepare to push to manual processing. The judgment logic can be implemented through simple conditional judgment or rule engine, supporting dynamic updating and extension to adapt to new error codes and business scenarios. Record the error code, error description, exception type, and judgment basis in the detailed execution log. Log information is not only used for problem troubleshooting and subsequent auditing, but also serves as model training data to optimize the accuracy of exception classification. For example, record "error code: 503, description: system busy, determined as temporary exception, retry count: 3". According to the exception type, the system sends a notification to the user or administrator. For temporary exceptions, notify the user "system busy, retrying, please wait"; for permanent conflicts, notify the user "business handling failed, reason: [specific description], please contact customer service later or try other solutions". At the same time, administrators can view detailed error logs through the background system for further analysis and processing.

[0077] S1513, if it is a temporary exception, a retry instruction is generated and added to the delay task queue to wait for retriggering; Further, if it is a temporary exception, a retry instruction is generated and added to the delay task queue to wait for triggering again. Specifically, the system predefines a retry strategy, including the maximum number of retries, retry interval time, and retry algorithm (such as fixed interval, exponential backoff, etc.). For example, the maximum number of retries is configured to be 3 times, the initial retry interval is 30 seconds, and the retry interval is increased by an exponential factor each time (such as 30 seconds, 1 minute, 2 minutes). The retry strategy can be dynamically adjusted according to business needs and system load. The retry instruction is generated according to the original operation instruction and the retry strategy. The retry instruction contains the complete information of the original operation instruction, the retry count counter, the retry interval time, etc. The system assigns a unique task ID to the retry instruction to facilitate tracking and management. The retry instruction is added to the delay task queue, and the task triggering time is set according to the configured retry interval time. The delay task queue is usually based on a distributed message queue or a timing task scheduling system (such as the delay plugin of RabbitMQ, the delay queue function of Apache Kafka). The system ensures that tasks are executed in order and supports high concurrency and high availability. The status of the retry instruction in the delay task queue is monitored, and the execution time and result of each retry are recorded. If the retry is successful, the task status is updated to "success" and the related resources are released; if the retry fails, the retry count counter is incremented, and it is determined whether to continue retrying or to be permanently failed according to the retry strategy. When the number of retries reaches the maximum limit and is still unsuccessful, the system marks the task as "permanent failure" and triggers step 74 to push the case context and reasoning chain to the artificial processing queue. At the same time, the user and administrator are notified of the specific reasons for the retry failure and the recommended operation.

[0078] S1514, if it is a permanent conflict or retry failure, terminate the automated process and push the case context and reasoning chain to the artificial processing queue.

[0079] Further, if it is a permanent conflict or retry fails, terminate the automation process and push the case context and reasoning chain to the manual processing queue. Specifically, when it is determined that there is a permanent business rule conflict or retry fails, terminate the automation process and push the case context and reasoning chain to the manual processing queue. Collect and organize the relevant context information of the case, including user request, business handling path, executed operation instructions, failed error code and description, key steps and logic in the reasoning chain, etc. These information are recorded in a structured format (such as JSON or XML) to ensure that the manual processing can fully understand the case background and failure reason. The organized case context and reasoning chain information are pushed to the manual processing queue. The manual processing queue is usually based on a ticket system or task scheduling platform (such as JIRA, ServiceNow), which supports manual intervention and processing. The system assigns a unique ticket ID to each pushed case for easy tracking and management. Create a processing task in the manual processing queue and assign it to the appropriate processing personnel (such as customer service personnel, business experts, system administrators, etc.) according to the priority and type of the case. When creating a task, it is accompanied by detailed case information and suggested operation steps to help processing personnel quickly locate the problem and take measures. Send a notification to the user to inform him that the business handling has been transferred to manual processing and provide the ticket ID for the user to query the processing progress. The notification content includes the estimated processing time, the reason for manual processing (such as "business rule restrictions" or "system multiple retry failures") and the next operation that the user can take (such as contacting customer service, providing supplementary information, etc.). After the manual processing is completed, the system receives the processing result feedback and updates the case status to "solved" or "closed". If it involves business rule update or system configuration adjustment, trigger the knowledge graph and model update process to optimize the system's processing ability for similar problems in the future.

[0080] In this embodiment, by monitoring the operation instruction execution and handling the failure situation, the reliability and user experience of business handling are significantly improved. Its beneficial effects include: real-time monitoring of instruction execution status, timely discovery and differentiation of temporary anomalies and permanent conflicts, targeted retry or manual intervention measures, and improved business handling success rate. At the same time, detailed recording and pushing of case context and reasoning chain facilitate manual quick problem locating, shorten processing time, enhance business handling process transparency and traceability, and ensure efficient and stable business handling.

[0081] For details, please refer to Figure 2 , Figure 2 The figure is a schematic diagram of the basic structure of the business handling device of the embodiment.

[0082] As Figure 2As shown, a business handling device includes: a business request module 1100 configured to receive a user business request and extract business information including user identity, target business and associated conditions from the business request; a graph retrieval module 1200 configured to retrieve business rules, mutual exclusion relationships and constraint conditions related to the business information in a pre-constructed business knowledge graph based on the business request; a business path module 1300 configured to input the business rules, mutual exclusion relationships and constraint conditions into a locally fine-tuned large model and generate a business handling path including conflict judgment and processing strategy through thought chain reasoning; a business instruction module 1400 configured to generate structured operation instructions that can be recognized and executed by a business system according to the business handling path; and an instruction execution module 1500 configured to send the operation instructions to the business system to prompt the business system to execute the business corresponding to the operation instructions.

[0083] The above-mentioned business handling device effectively solves the problems of knowledge update lag, inability to cover complex rules and mutual exclusion relationships in existing business handling systems by automatically extracting business information and retrieving related rules in a knowledge graph, generating a business handling path with conflict judgment and processing strategy through thought chain reasoning of a locally fine-tuned large model. This method realizes the full-process intelligentization of business handling, reduces manual intervention, reduces operating costs, and improves handling efficiency and success rate. At the same time, the business handling path is converted into structured operation instructions and sent to the business system for execution, ensuring the accuracy and timeliness of the operation. In addition, the locally fine-tuned large model enhances the understanding of industry knowledge, and the thought chain reasoning improves the transparency and explainability of the reasoning process, making the business handling more reliable and controllable.

[0084] To solve the above technical problems, the embodiments of the present application also provide a computer device. For details, please refer to Figure 3 , Figure 3 The basic structure block diagram of the computer device of the present embodiment is shown.

[0085] As shown in Figure 3 , the internal structure diagram of the computer device. The computer device includes a processor, a non-volatile storage medium, a memory and a network interface connected by a system bus. Among them, the non-volatile storage medium of the computer device stores an operating system, a database and computer readable instructions, and the database can store a control information sequence. The processor of the computer device is used to provide computing and control capability to support the operation of the entire computer device. The memory of the computer device can store computer readable instructions, which when executed by the processor, can cause the processor to execute a business handling method. The network interface of the computer device is used to connect and communicate with the terminal. Those skilled in the art can understand, Figure 3The structure shown in FIG. 1 is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. A specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0086] The processor in the embodiment is configured to execute Figure 2 The functions of the business request module 1100, the graph retrieval module 1200, the business path module 1300, the business instruction module 1400, and the instruction execution module 1500 are achieved by automatically extracting business information and retrieving relevant rules in the knowledge graph, combining the thinking chain reasoning of the locally fine-tuned large model to generate conflict judgment and processing strategy business handling path, effectively solving the problems of knowledge update lag and inability to cover complex rules and mutual exclusion in the existing business handling system. The method realizes the intelligentization of the whole process of business handling, reduces manual intervention, reduces operating costs, and improves handling efficiency and success rate. At the same time, the business handling path is converted into structured operation instructions and sent to the business system for execution, ensuring the accuracy and timeliness of the operation. In addition, the locally fine-tuned large model enhances the understanding of industry knowledge, and the thinking chain reasoning improves the transparency and interpretability of the reasoning process, making the business handling more reliable and controllable.

[0087] The present application also provides a storage medium storing computer readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the business handling method described in any of the embodiments.

[0088] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a computer readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments. The storage medium can be a non-volatile storage medium such as a magnetic disc, an optical disc, a read-only memory (ROM), or a random access memory (RAM).

[0089] A person of ordinary skill in the art can understand that the steps, measures, and schemes in the various operations, methods, and processes discussed in the present application can be alternated, changed, combined, or deleted. Further, other steps, measures, and schemes in the various operations, methods, and processes discussed in the present application can also be alternated, changed, rearranged, decomposed, combined, or deleted. Further, the steps, measures, and schemes in the prior art with the various operations, methods, and processes disclosed in the present application can also be alternated, changed, rearranged, decomposed, combined, or deleted.

[0090] The above merely provides part of the embodiments of the present application, and it should be pointed out that, for those skilled in the art, some improvements and refinements can be made without departing from the principles of the present application, and these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. A business processing method, characterized in that, include: Receive user service requests and extract service information containing user identity, target service, and associated conditions from the service requests; Based on the business request, retrieve the business rules, mutual exclusion relationships and constraints related to the business information from the pre-built business knowledge graph; The business rules, mutual exclusion relationships, and constraints are input into the large model that is fine-tuned locally, and a business processing path containing conflict judgment and handling strategies is generated through thought chain reasoning. Based on the business processing path, generate structured operation instructions that can be recognized and executed by the business system; The operation instruction is sent to the business system, causing the business system to execute the business corresponding to the operation instruction.

2. The business processing method according to claim 1, characterized in that, The step of retrieving business rules, mutual exclusion relationships, and constraints related to the business information from the pre-constructed business knowledge graph includes: A graph convolutional neural network is used to optimize the embedding representation of nodes in a pre-constructed business knowledge graph, generating node vectors containing semantic information. Based on the node vectors, the semantic similarity between the service request and the graph nodes is calculated, and the initial set of associated nodes is located. Starting from the initial set of associated nodes, a subgraph retrieval based on shortest path and adjacency matrix analysis is performed to extract the subgraph structure containing complete rule chains and mutual exclusion relationships; Retrieve business rules, mutual exclusion relationships, and constraints related to the business information from the subgraph structure.

3. The business processing method according to claim 1, characterized in that, The process of generating a business handling path that includes conflict judgment and resolution strategies through thought chain reasoning includes: Based on the retrieved constraints, the logic for detecting business conflicts and warnings of potential risks are inferred step by step and explicitly output. For the identified conflicts, multiple handling strategies are simulated, and their cascading impact on user status and contract terms is evaluated. By integrating the evaluation results, an interpretable business processing path containing a specific sequence of business operations is generated.

4. The business processing method according to claim 1, characterized in that, The step of generating structured operation instructions that can be recognized and executed by the business system based on the business processing path includes: Analyze the business processing path and convert each atomic operation step into a corresponding system call logic; The system call logic is mapped to specific, structured instructions that can be directly executed by the business system. These structured instructions include API call commands, SQL statements, or workflow tasks. The structured instructions are sorted and their dependencies are checked according to the business logic to form an ordered sequence of operation instructions.

5. The business processing method according to claim 1, characterized in that, The business processing path includes alternative processing paths, and the method for generating the alternative processing paths includes: Detect business conflicts between the user's target business and existing business; When a mutual exclusion relationship is detected between a user's target business and an existing business, the identifier of the existing business that caused the mutual exclusion and the corresponding contract terms are extracted. Based on the existing business identifier and the corresponding contract terms, the mutually exclusive business represented by the mutual exclusion relationship is identified, and an alternative processing path for canceling the mutually exclusive business and processing the target business is generated.

6. The business processing method according to claim 5, characterized in that, After confirming the mutually exclusive services represented by the mutual exclusion relationship based on the existing service identifier and the corresponding contract terms, and generating an alternative processing path for canceling the mutually exclusive services and processing the target service, the method further includes: Calculate the cost information of the alternative processing route; Establish the association between the alternative processing path and cost information, so as to incorporate the alternative processing path and cost information into the final generated structured operation instruction.

7. The business processing method according to claim 1, characterized in that, After sending the operation instruction to the business system, causing the business system to execute the business corresponding to the operation instruction, the method further includes: Monitor the execution of the operation instructions; When the operation instruction fails to execute, the error code returned by the failure is parsed to determine whether it is a temporary system exception or a permanent business rule conflict. If the exception is temporary, a retry instruction is generated and added to the delayed task queue to wait for it to be triggered again; If the conflict is permanent or the retry fails, the automated process is terminated and the case context and reasoning chain are pushed to the human processing queue.

8. A business processing device, characterized in that, include: The business request module is used to receive user business requests and extract business information containing user identity, target business and associated conditions from the business requests. The graph retrieval module is used to retrieve business rules, mutual exclusion relationships, and constraints related to the business information from a pre-built business knowledge graph based on the business request. The business path module is used to input the business rules, mutual exclusion relationships and constraints into the large model for local fine-tuning, and generate a business processing path that includes conflict judgment and handling strategies through thought chain reasoning. The business instruction module is used to generate structured operation instructions that can be recognized and executed by the business system based on the business processing path. The instruction execution module is used to send the operation instruction to the business system, causing the business system to execute the business corresponding to the operation instruction.

9. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores computer-readable instructions that, when executed by the processor, cause the processor to perform the steps of the business processing method as described in any one of claims 1 to 7.

10. A storage medium storing computer-readable instructions, characterized in that, When the computer-readable instructions are executed by one or more processors, the one or more processors perform the steps of the business processing method as described in any one of claims 1 to 7.

Images