Method and system for realizing enterprise collaborative decision-making based on AI algorithm and government-enterprise API

Abstract

The embodiment of the invention discloses a method and a system for realizing enterprise collaborative decision-making based on an AI algorithm and a government-enterprise API (Application Program Interface). The method comprises the following steps: receiving input data interacted by using a natural language; converting the input data into standard format data; determining a service type related to the input data by using an AI model; the service type comprises public service and enterprise application service; the public service and the enterprise application service are both associated with one or more subtasks; determining a thinking chain providing the service; the sub-tasks contained in different thinking chains are different, or the execution sequences of the sub-tasks of different thinking chains are different, or the execution proportions of the sub-tasks contained in the thinking chains are different; planning an optimal thinking chain; the optimal thinking chain corresponds to an optimal execution path; the optimal execution path comprises one or more execution nodes which are arranged in order; and calling the API according to the optimal execution path, so that the called API provides the service required by the user based on the standard format data.

Description

Technical Field

[0001] The present invention relates to the field of service technology, and in particular to a method and system for implementing enterprise collaborative decision-making based on artificial intelligence (AI) algorithms and government-enterprise application programming interfaces (APIs). Background Art

[0002] During collaborative enterprise decision-making, input data is received from various channels and in various formats. This data format can vary widely, lacking a unified standard. This greatly complicates subsequent processing and analysis. For example, data provided by different systems may use different file formats (TXT, CSV, Excel, JSON, etc.), or data encoding methods and field naming conventions may be inconsistent, making it difficult to directly analyze and utilize the data. Furthermore, the specific services to be invoked require manual selection, which places high demands on users or hinders access to appropriate services. Summary of the Invention

[0003] In view of this, the embodiment of the present invention provides a method and system for implementing enterprise collaborative decision-making based on AI algorithms and government-enterprise APIs. The technical solution of the present invention is implemented as follows: The first aspect provides a method for implementing enterprise collaborative decision-making based on AI algorithms and government-enterprise APIs, which is applied to implementing enterprise collaborative decision-making systems based on AI algorithms and government-enterprise APIs. The method includes: receiving input data using natural language interaction; converting the input data into standard format data; using an AI model to determine the service type involved in the input data; the AI ​​model runs in the agent of the enterprise collaborative decision-making system based on AI algorithms and government-enterprise APIs; the service types include public services and enterprise application services; the public services and the enterprise application services are both associated with one or more subtasks; planning an optimal thinking chain; the optimal thinking chain corresponds to an optimal execution path; the optimal execution path includes one or more ordered execution nodes; calling an API according to the optimal execution path, so that the called API provides the user's required services based on the standard format data.

[0004] The second aspect provides an enterprise collaborative decision-making system based on artificial intelligence AI algorithm and government-enterprise application program interface API, characterized in that the system includes: a receiving module for receiving input data using natural language interaction; a conversion module for converting the input data into standard format data; a determination module for using an AI model to determine the service type involved in the input data; the AI ​​model runs in the agent of the enterprise collaborative decision-making system based on AI algorithm and government-enterprise API; the service types include public services and enterprise application services; the public services and the enterprise application services are both associated with one or more subtasks; a planning module for planning the optimal thinking chain; the optimal thinking chain corresponds to the optimal execution path; the optimal execution path includes one or more ordered execution nodes; a calling module for calling the API according to the optimal execution path, so that the called API provides the user's required services based on the standard format data.

[0005] The technical solutions provided by the embodiments of the present disclosure resolve data format inconsistencies by converting input data into a standard format, achieving data standardization and integration. This allows data from different sources to be uniformly processed and analyzed within the same system, improving data usability and value. Using AI models to determine the type of service to be provided and the optimal thought process for providing the service, the optimal execution path can be automatically planned based on the input data and service type. This can reduce unnecessary service calls, improve the efficiency and quality of service calls, and lower enterprise operating costs. Furthermore, the AI ​​model runs in a system agent, rather than in the subsystem providing the service. This allows the integration of AI functionality into various legacy systems through the configuration of agents (agent devices or agent resources), improving system compatibility and upgrade adaptability. Accurately calling APIs based on the execution path and providing the required services based on standard format data improves API call reliability and service quality. This ensures smooth collaborative decision-making within the enterprise and reduces business interruptions caused by API call failures or substandard service provision. For example, by implementing error handling and retry mechanisms for API calls, service stability and availability can be improved, ensuring normal enterprise operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings: Figure 1 A flowchart of a method for implementing enterprise collaborative decision-making based on AI algorithms and government-enterprise APIs provided in an embodiment of the present invention; Figure 2A flowchart illustrating another method for implementing enterprise collaborative decision-making based on AI algorithms and government-enterprise APIs provided in an embodiment of the present invention; Figure 3 A schematic diagram of the structure of an enterprise collaborative decision-making system based on AI algorithms and government-enterprise APIs provided in an embodiment of the present invention; Figure 4 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0007] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0008] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0009] like Figure 1 As shown, the embodiment of the present disclosure provides a method for implementing enterprise collaborative decision-making based on AI algorithms and government-enterprise APIs, which is applied to implement an enterprise collaborative decision-making system based on AI algorithms and government-enterprise APIs, including: S1101: receiving input data using natural language interaction; the input data is unordered; S1102: Convert the input data into standard format data; the standard format data is ordered; illustratively, the disorder of the input data is relative to the standard format data; multiple data in the standard format data are sorted according to a predetermined setting.

[0010] S1103: Determine the service type involved in the input data; the service type includes public services and enterprise application services; the public services and the enterprise application services are both associated with one or more subtasks; S1104: Planning an optimal thinking chain; the optimal thinking chain corresponds to an optimal execution path; the optimal execution path includes one or more orderly arranged execution nodes; the AI ​​model runs in the agent of the enterprise collaborative decision-making system based on the AI ​​algorithm and the government-enterprise API; S1105: Calling an API according to the optimal execution path, so that the called API provides the service required by the user based on the standard format data.

[0011] This method for implementing enterprise collaborative decision-making based on an AI algorithm and a government-enterprise API can be used to implement an enterprise collaborative decision-making system based on an AI algorithm and a government-enterprise API. The system includes one or more electronic devices. For example, this method can be used on one or more electronic devices.

[0012] In the embodiments of the present disclosure, there may be one or more AI models running in the system, and / or there may be one or more AI models running in the system.

[0013] For example, AI models can be divided into: Data parsing and conversion models: These models convert the diverse input data received into a standardized format. Natural language processing (NLP) techniques such as word segmentation, part-of-speech tagging, and syntactic analysis can be used to process text data. For image data, computer vision models such as convolutional neural networks (CNNs) are used for feature extraction and recognition, converting it into processable structured data.

[0014] Service type identification model: This model uses machine learning-based classification algorithms, such as decision trees, support vector machines (SVMs), or multi-layer perceptrons (MLPs) from deep learning, to classify input data and determine whether the service involved is a public service or an enterprise application service. This model is trained with a large amount of labeled data to improve classification accuracy.

[0015] Thought Chain Model: The thought chain model determines a thought chain based on the AI ​​model. This thought chain can be used to determine the subtasks involved in providing a corresponding service. The thought chain model outputs thought chain information, which may include the subtask name or task number. These subtasks are arranged in order and have a sequential relationship in the thought chain.

[0016] Planning Model: Reinforcement learning models, such as Deep Q-Network (DQN) or policy gradient algorithms (such as A2C and A3C), can be used. These models continuously try different chains of thought in a simulated environment and learn the optimal chain of thought based on a reward mechanism to achieve efficient service provision. For example, the AI ​​model determines a feasible chain of thought for providing a service; different feasible chains of thought contain different subtasks, or the execution order of subtasks in different feasible chains of thought differs, or the execution percentage of subtasks in a feasible chain of thought differs; this execution percentage can be a focus, for example, on the number of executions or the execution duration ratio. A feasible chain of thought corresponds to a feasible execution path (or alternative execution path). An execution path includes one or more ordered execution nodes; the AI ​​model runs in the agent that implements the enterprise collaborative decision-making system based on AI algorithms and government-enterprise APIs; the execution path includes one or more ordered execution nodes; and each execution node is used to execute at least one subtask.

[0017] During the data conversion phase, the data parsing and conversion model receives input data generated using natural language processing and applies appropriate technologies based on the data type. For example, for a text file containing corporate financial information, an NLP model extracts key information and then organizes it according to a pre-set standard format, providing a unified data foundation for subsequent service calls. During the service type determination phase, the service type identification model analyzes the converted standard-format data and determines the service type based on its characteristics and patterns. For example, if the data contains information related to corporate tax filings, it is classified as a public service; if it is internal customer relationship management data, it is classified as an enterprise application service. During the planning phase, the planning model generates an optimal execution path based on the input data and service type, combined with historical data and environmental information. This execution path associates a series of ordered execution nodes, each corresponding to a subservice (or subtask), ensuring that services are provided in a reasonable order and manner. Input data can be either unordered or ordered; unordered input data is one possible scenario.

[0018] In some embodiments, the thinking chain may provide a thinking link or an execution link that serves the AI ​​model. For example, different thinking chains contain different subtasks, or the execution order of the subtasks of different thinking chains is different, or the execution proportion of the subtasks contained in the thinking chain is different; one thinking chain corresponds to at least one execution path; in the embodiment of the present disclosure, the planning model will select the current optimal execution path that is adapted to the thinking chain. The execution path includes one or more orderly arranged execution nodes; one execution node is used to execute at least one subtask; the AI ​​model runs in the agent that implements the enterprise collaborative decision-making system based on AI algorithms and government-enterprise APIs.

[0019] In some embodiments, determining an execution path based on the thought chain may include: determining one or more alternative paths that are compatible with the thought chain; and selecting the optimal alternative path as the execution path for the service provided. In some embodiments, the cost value of each alternative link is determined based on cost or overhead. The optimal alternative path (i.e., the execution path ultimately selected) may be the alternative path with the lowest cost value.

[0020] In some embodiments, the execution path is selected based on the following functional relationship: ,in, is the cost value of the xth alternative path; is the i-th cost or overhead of the j-th execution node; is the weight of the i-th cost or overhead of the j-th execution node. J is the number of execution nodes included in the x-th alternative path.

[0021] Costs or expenses may include, but are not limited to, at least one of the following: waiting time; execution time; API call cost; task completion quality, which may specifically include task completion degree.

[0022] In other embodiments, , is the cost value of the xth alternative path; is the weight of time cost. is the total time cost (i.e., the required time), The weight of the API call cost. The total cost of API calls. Quality of task completion.

[0023] In an embodiment of the present disclosure, the AI ​​model may use a reinforcement learning model to train and optimize the AI ​​model.

[0024] In an embodiment of the present disclosure, the AI ​​model may use a Monte Carlo tree search algorithm to search and find the optimal chain of thought.

[0025] In the present disclosure, an API (Application Programming Interface) is a set of definitions, protocols, and tools used for interaction and communication between different software components, systems, or applications. Government and enterprise APIs can be divided into public service APIs and government and enterprise service APIs.

[0026] Public service APIs may include but are not limited to the following: Bill Authenticity Verification API: This API allows businesses to use OCR technology to parse the textual information in bill images (such as invoices and checks), extract key data such as bill number and amount, and then call the verification API of tax or financial institutions to compare the information with official databases to implement an automated bill authenticity verification process.

[0027] Smart check and certification API: Enterprises can use this API to call the financial system or tax platform interface, use AI models to semantically understand input invoices, automatically check invoices that meet the deduction conditions, and use RPA technology to simulate manual operations to complete certification submission, reducing manual verification costs.

[0028] Digital invoicing API: The government and enterprise agent connects to the tax department's electronic invoice public service API. After the enterprise enters the invoicing information (such as purchaser information and product details), the system generates compliant digital invoice structured data through the AI ​​model, calls the invoicing API to complete the invoice issuance, and returns the digital format file.

[0029] Digital invoice format file download API: This API allows enterprises to call the tax platform API through the government and enterprise agent, enter invoice code, number, and other information, and the AI ​​model automatically downloads the corresponding digital invoice format file (such as OFD format) after verifying the enterprise's permissions. It supports batch download and categorized storage.

[0030] Enterprise tax declaration API: Enterprises submit financial data such as sales and costs (supporting natural language input or file upload). The system converts it into a standard format through an AI model, plans a thinking chain of "data verification - form filling - submission for review", and calls the tax declaration API to complete the automated declaration of taxes such as value-added tax and corporate income tax.

[0031] Individual tax declaration API: For the individual tax declaration scenario of corporate employees, the system receives data such as wages and special additional deductions provided by HR, generates individual tax declaration forms after parsing through AI models, calls the individual tax declaration API of the tax department, automatically completes batch declarations and tax calculations, and supports abnormal data warnings.

[0032] Enterprise Risk Identification API: The third-party agent integrates the company's internal financial data (such as ERP system credentials) and external policy data (such as new tax regulations). Through time series analysis and association rule mining algorithms, it calls the risk identification API to generate assessment reports on the company's tax compliance risk, cash flow risk, etc., and provides early warning recommendations.

[0033] Intelligent document parsing and OCR recognition API: The first agent receives images or scans of documents such as contracts and invoices, extracts text content through OCR technology, and combines it with the NLP model to parse the document structure (such as the amount and date fields in the invoice), converting it into standard format data for subsequent service calls (such as tax declaration and bill verification).

[0034] For example, the ERP system integration API facilitates data exchange and business collaboration between different modules within an enterprise (such as declaration, invoicing, accounting, and bank statement printing), enabling unified management and optimized allocation of enterprise resources. The CRM customer management API helps companies manage customer information, sales opportunities, and customer service, automating the management and analysis of customer relationships. The OA office automation API supports daily office processes within an enterprise, such as document approval, meeting scheduling, and attendance management, improving office efficiency.

[0035] In the implementation of API services, after determining the service type and thought chain associated with the input data, the system calls the corresponding API based on the sub-services corresponding to each execution node in the thought. For public service APIs, the system first performs identity authentication and permission verification to ensure that the enterprise is qualified to call the service. Then, the standard format data is encapsulated according to the API requirements and the request is sent to the public service server. After receiving the request, the server verifies and processes the data and returns the results to the system.

[0036] For enterprise application service APIs, the system also performs identity authentication and permission checks to ensure that only authorized users and systems can call the API. Then, according to the API interface specifications, standard format data is passed to the enterprise application service application. The application processes the data and returns the results to the system. Throughout this process, the system monitors and provides feedback on the results of API calls to ensure accurate service provision and efficient execution.

[0037] In some embodiments, the information included in the execution path of the thought chain may include but is not limited to at least one of the following: The execution node information may include at least one of the following: Service Identification: Each execution node corresponds to a subservice, and the subservice's unique identifier must be clearly defined. For example, when calling a public service API, each service may have a specific service number or name, such as "Tax Filing Service - VAT Filing" or "Social Security Payment Service - Employee Social Security Payment." This allows the API to accurately identify the service being called.

[0038] Service Type: Specify whether the service is a public service or an enterprise application service. Different types of services may have different invocation methods and permission requirements. For example, a public service may need to comply with government-mandated interface standards and security mechanisms, while an enterprise application service may be closely tied to the company's internal business systems.

[0039] Service address: This is the access address of the service. For network-based services, this is usually a URL or IP address. For example, the API address of an enterprise application service might be https: / / api.example.com / enterprise-service, while the address of a public service might be https: / / gov.example.gov / public-service.

[0040] The node execution order information may include at least one of the following: Predetermined dependencies: Each execution node may have its own predetermined dependencies. This means that the node's services can only be called after the predetermined node's services have successfully executed. For example, in a company's financial reimbursement process, the "Reimbursement Approval Service" might have a predetermined dependency on the "Reimbursement Information Entry Service." The approval service can only be executed after the reimbursement information has been entered and passed preliminary verification.

[0041] Execution sequence number: Each execution node is assigned an execution sequence number to ensure that the API calls services in the correct order. For example, in an enterprise collaborative decision-making process with multiple sub-services, the "Data Collection Service" is numbered 1, the "Data Analysis Service" is numbered 2, and the "Decision Recommendation Generation Service" is numbered 3. The API will call these services in the order 1 -> 2 -> 3.

[0042] The order of execution node information corresponds to the order of each execution node. For example, if the identifier of node A is in the xth position in the execution path, then the node can be considered to be scheduled in the order x.

[0043] Data flow information may include: Input data format and content: Each execution node must specify the format (e.g., JSON, XML, etc.) and content of its input data. For example, a "tax filing service" might require input of a company's financial data, such as sales, costs, and profits, and require the data to be provided in JSON format, such as { "sales": 100000, "cost": 50000, "profit": 50000}.

[0044] Output data format and content: Similarly, each execution node must specify the format and content of its output data. For example, the output of a "data analysis service" might be a JSON object containing the data analysis results, such as { "analysis_result": "Sales are on an upward trend. It is recommended to increase marketing investment"}.

[0045] Data mapping: When the output data of one node serves as the input data for another, the data mapping relationship between them must be clearly defined. For example, the employee information output by the "Data Collection Service" includes fields such as "Employee Name," "Employee ID," and "Employee Department," while the "Employee Performance Evaluation Service" may only require "Employee ID" and "Employee Department" as input. In this case, the mapping relationship between these two fields must be clearly defined.

[0046] Error codes and error messages: Each execution node needs to define possible error codes and corresponding error messages so that the API can accurately identify and handle errors when calling the service. For example, a failed service call might return error code 1, indicating an internal server error, with an error message such as "An error occurred while processing the request. Please try again later."

[0047] Retry strategies may include: For some temporary errors, such as network timeout, service busy, etc., a retry strategy needs to be defined. For example, when a service call returns error code 2 indicating that the service is unavailable, the API can attempt to retry.

[0048] The execution path information also includes flow relationship information. The flow relationship information may indicate but is not limited to at least one of the following: 1. Sequential flow relationships. In most cases, execution nodes flow sequentially in the order in which they are executed. After the service of the previous node completes, its output data becomes the input data of the next node. The API calls the services of each node in sequence. For example, in a company's order processing process, after the "Order Creation Service" completes, the order information is passed as input to the "Inventory Check Service." After the inventory check is completed, the results are passed to the "Shipping Service," and so on.

[0049] 2. Conditional flow relationships. In some cases, the flow of execution nodes must be judged based on certain conditions. For example, in an enterprise's customer credit assessment process, after completing the "Preliminary Credit Assessment Service," if the customer's credit score is above a certain threshold, the flow will be transferred to the "Advanced Credit Assessment Service." If the credit score is below the threshold, the flow will be transferred to the "Risk Warning Service." The API selects different subsequent nodes to call based on the conditional judgment results.

[0050] 3. Parallel flow relationships. For independent services, parallel flow can be adopted. For example, in a company's market research process, the "Competitor Analysis Service" and the "Market Trend Analysis Service" can be executed in parallel, with independent input data and independent output results. The API calls these parallel nodes' services simultaneously to improve overall efficiency. After all parallel nodes complete their execution, the results are aggregated and passed to the next node for processing.

[0051] 4. Jump flow relationships. For some services, if certain conditions are met, a sub-service or step can be skipped to achieve interactions in different situations within a complete execution path. In some embodiments, the agent corresponds to a proxy device. Exemplarily, the proxy device includes a hardware server device and / or a cloud server. Hardware servers: Typically use high-performance enterprise-class servers with multi-core processors, large-capacity memory, and high-speed storage devices. Examples include Dell PowerEdge series and HP ProLiant series. These servers are capable of processing large amounts of input data, running complex AI models, and performing API call operations. They can be deployed in an enterprise's internal data center or hosted by a cloud service provider, such as Alibaba Cloud and Tencent Cloud. Cloud servers: Use virtual server instances provided by cloud service providers, such as Amazon AWS's EC2 and Microsoft Azure's virtual machines. Cloud servers are elastically scalable and can dynamically adjust computing resources according to the enterprise's business needs, avoiding the high upfront investment and subsequent maintenance costs of hardware servers.

[0052] In some embodiments, proxy devices may also include network devices. For example, routers connect an enterprise's internal network and external networks, enabling data forwarding and routing. Enterprise-grade routers from brands like Cisco and Huawei offer high bandwidth, high reliability, and robust security features, ensuring stable data transmission within the network. Firewalls are deployed within these devices. For example, firewalls are deployed at the enterprise network boundary to filter and monitor data entering and leaving the network, preventing attacks and unauthorized access from external networks. Firewalls can be hardware devices or software, such as those from brands like CheckPoint and Fortinet.

[0053] In some embodiments, proxy devices can be deployed locally or in the cloud. Enterprises deploy servers and network equipment in their own data centers, connecting various departments and devices via an internal network. This deployment approach ensures data security and privacy, and provides the enterprise with complete control over the equipment. However, it requires significant investment in hardware procurement, site rental, power supply, and maintenance personnel. Enterprises deploy servers and related applications on a cloud service provider's platform, connecting to the cloud server via the internet. Cloud deployment offers advantages such as low cost, ease of scalability, and maintenance, eliminating the need for enterprises to worry about hardware updates and maintenance. However, attention must be paid to data security and network bandwidth stability.

[0054] The proxy device performs various operations, including one or more of the following: During data reception, the agent receives input data from various departments within the enterprise or external partners. This data can be in various formats, such as text, tables, and images. The agent performs preliminary checks and verification on the data to ensure its integrity and accuracy.

[0055] Data conversion uses AI algorithms and data processing tools to convert received input data into a standard format. For example, for text data, this involves word segmentation, part-of-speech tagging, and named entity recognition; for image data, this involves image recognition and feature extraction.

[0056] The service type determination operation uses AI models to analyze and classify the converted standard format data, determining whether the input data relates to public services or enterprise application services. For example, by analyzing keywords and semantic information in the data, it can determine whether the data is related to public services such as tax declarations, invoice purchase and sales, social security contributions, customs export tax rebates, and bank receipts, or to enterprise application services such as internal customer relationship management and supply chain management.

[0057] The execution path determination operation uses an AI model to determine the execution path for the service based on the input data and service type. This path consists of one or more sequentially arranged execution nodes, each corresponding to a sub-service. For example, in an enterprise collaborative decision-making process, the data collection service may be called first, followed by the data analysis service, and finally the decision recommendation generation service.

[0058] API call operations involve agents sequentially calling the corresponding APIs based on the determined optimal chain of thought. When calling an API, standard format data is passed as input parameters to the API, and the API returns the processed results. For example, when calling a tax declaration API, the company's financial data is input, and the tax department returns the declaration results.

[0059] By introducing and integrating government and enterprise APIs through proxies, data integration can be addressed. Specifically, this can include inconsistent data formats and standards across different departments within an enterprise and with external partners. Proxies can convert this diverse data into standardized formats, enabling data integration and unified management, facilitating collaborative decision-making within the enterprise. This also addresses service selection and invocation issues. For example, faced with numerous public services and enterprise application services, an enterprise may not know how to select the appropriate service and invocation method. Proxies can automatically determine the service type and execution path based on input data and invoke the corresponding API, streamlining the enterprise's service usage process and improving the efficiency and accuracy of service invocation.

[0060] Security and privacy issues can be addressed by introducing proxies. Deployed between an enterprise's internal and external networks, these proxies act as an intermediate data processing layer, encrypting, filtering, and monitoring data to protect the security and privacy of enterprise data. Furthermore, proxies can authenticate and authorize API calls, preventing unauthorized access and data leakage.

[0061] By leveraging AI models and automated operational processes, agents can rapidly process large amounts of data, determine optimal execution paths and service invocation methods, and improve the efficiency and quality of collaborative enterprise decision-making. For example, in an emergency, agents can quickly invoke relevant services to provide timely decision-making support.

[0062] During the collaborative decision-making process of an enterprise, input data from different channels and in different formats is received. The formats of this data may vary greatly, as the data provided by different users or enterprises has diverse data formats. In the disclosed embodiment, standard format data is pre-determined in the system. This solves the lack of unified standards, which brings great difficulties to the subsequent processing and analysis of data. For example, data provided by different departments may use different file formats (such as TXT, CSV, Excel, JSON, etc.), or the data encoding method and field naming rules may be inconsistent, resulting in the inability to directly perform effective analysis and utilization.

[0063] During collaborative decision-making, enterprises need to invoke different types of services, including public services and enterprise application services. However, accurately identifying the service type associated with input data and matching it with the appropriate service using the methods provided in the embodiments of this disclosure is a challenging problem. Different services may have different functionalities and interface requirements. Failure to accurately identify the service type may result in invoking the wrong service, thus failing to meet the enterprise's actual needs. To achieve collaborative decision-making, it is necessary to determine the execution path for providing the service. This path consists of a series of ordered execution nodes, each corresponding to a sub-service. Using AI models to plan the optimal execution path based on the input data and service types is key to improving service efficiency and quality. Improper execution path planning can lead to inefficient service invocation and unnecessary time and cost. After determining the execution path, the corresponding API must be called according to the path to provide the required service. However, different APIs may have different invocation methods, parameter requirements, and return formats. Ensuring that the API is correctly invoked and the required service is provided based on standard formatted data is a technical challenge that needs to be addressed. Failure to invoke the API or non-compliant service delivery will impact the entire collaborative decision-making process.

[0064] In short, converting input data into a standard format resolves data format inconsistencies and achieves data standardization and integration. This allows data from different sources to be processed and analyzed uniformly within a single system, improving data usability and value. For example, standardized data can be more easily stored, queried, and statistically analyzed, providing more accurate and comprehensive information support for enterprise decision-making. Technical methods can accurately identify the service type associated with input data and match it with the appropriate service. This ensures that enterprises can access services that meet their specific needs, improving the relevance and effectiveness of services. For example, for input data related to tax filings, the corresponding tax public service API can be accurately identified and invoked, avoiding business failures caused by mismatched service types. Using AI models to determine the execution path for service delivery automatically plans the optimal execution path based on input data and service type. This reduces unnecessary service calls, improves service call efficiency and quality, and reduces enterprise operating costs. For example, by optimizing the execution path, the number of service calls and waiting time can be reduced, accelerating business processing and improving enterprise responsiveness.

[0065] Accurately calling APIs based on execution paths and providing required services based on standardized data formats can improve API call reliability and service quality. This ensures smooth collaborative decision-making within the enterprise and reduces business interruptions caused by API call failures or substandard service delivery. For example, implementing error handling and retry mechanisms for API calls can improve service stability and availability, ensuring normal enterprise operations.

[0066] In some embodiments, the enterprise collaborative decision-making system based on AI algorithms and government-enterprise APIs includes a first agent, a second agent and / or a third agent; the first agent runs a first AI model and is used as an agent for providing public services; the second agent runs a second AI model and is used as an agent for providing enterprise application services; the third agent runs a third AI model and is used to provide decision data based on data within the system and / or data outside the system.

[0067] For example, the first agent could be a government-enterprise agent, responsible for interacting with public service APIs. The second agent could be an enterprise service agent, enabling the core to schedule enterprise software using RPA and a user interface (GUI). The third agent could be a decision agent, enabling the core to generate decision reports using data models and retrieval-augmented generation (RAG) technology.

[0068] The system includes three types of agents, which collaborate sequentially. For example, the first agent primarily handles public service-related tasks. When an enterprise has a public service need, such as tax filing, invoice purchase and sales, social security payments, customs export tax rebates, or bank receipts, the first agent is activated. It receives relevant input data, processes it using the first AI model, converts the data into a standard format, determines the service type and execution path, and then calls the public service API to complete the business operation. The second agent focuses on enterprise application services. After the first agent completes public service processing, if the enterprise's internal business processes also involve enterprise application services, such as customer relationship management or supply chain management, the second agent is triggered. The second agent receives data that may have been passed on by the first agent, or newly generated data within the enterprise, and analyzes and processes it using the second AI model. It also standardizes the data, determines the service type, and plans the execution path. It then calls the enterprise application service API to provide the enterprise application service. The third agent provides decision-making data based on both internal and external data. After the first and second agents complete their business operations, they aggregate the relevant data and send it to the third agent. The third-party agent uses a third-party AI model to comprehensively analyze this data, along with externally acquired data on market trends, policies, and regulations, to provide valuable insights for collaborative decision-making. For example, after a company completes tax filings (handled by the first agent) and customer order management (handled by the second agent), the third-party agent can use this data, along with market demand forecasts, to inform production planning and sales strategies.

[0069] These three types of agents interact with each other. Specifically, data exchange may occur between the first and second agents. For example, when filing tax returns (a task performed by the first agent), an enterprise may need to access internal sales and cost data (enterprise application service data managed by the second agent). The second agent then provides the relevant data to the first agent to complete the public service processing. Both the first and second agents pass the processed data to the third agent. The third agent then conducts in-depth analysis of this data, uncovering its potential value and supporting enterprise decision-making. For example, the first agent's tax filing data and the second agent's sales performance data can help the third agent analyze the enterprise's financial situation and market competitiveness, thereby providing more accurate decision-making recommendations.

[0070] The location relationship between these three types of agents can include, but is not limited to, a combination of local and cloud deployments. Because the first agent involves public services, some key components can be deployed in the enterprise's local data center to ensure data transmission security and compliance. Furthermore, to improve processing efficiency and cope with peak business demands, some non-sensitive computing tasks and data can be stored in the cloud. For example, the first agent's data reception and preliminary processing modules can be deployed locally, while some of the first AI model's computations can be performed on cloud servers. The second agent primarily handles internal enterprise services and is typically deployed in the enterprise's local data center. This allows for better integration with various internal business systems, ensuring fast data transmission and processing. For example, the second agent can directly connect to the enterprise's ERP system, CRM system, and other systems to facilitate the acquisition and processing of data related to enterprise application services. To obtain more comprehensive external data, such as market trends and industry reports, some components of the third agent can be deployed in the cloud. Furthermore, to ensure secure access and analysis of internal enterprise data, data interfaces and some processing modules can also be deployed locally. For example, the third agent's external data collection module can be deployed in the cloud, while analysis of internal enterprise data and execution of decision models can be performed locally.

[0071] In some embodiments, the three types of agents are network-isolated and interact with each other. While the first, second, and third agents can exchange data via the enterprise's internal network, certain network-level isolation is implemented. For example, the first agent interacts with the public service API through a dedicated secure channel, physically or logically isolated from the enterprise's internal network to prevent public service data leakage. Access control and permission management are also implemented between the second and third agents to ensure data security and confidentiality.

[0072] In some embodiments, the first and second agents can perform backup and disaster recovery for data processing and storage. For example, the second agent can back up some public service data processed by the first agent. If the first agent fails, the second agent's backup data can be used for recovery and processing. Conversely, the first agent can also back up some key enterprise application service data of the second agent.

[0073] In some embodiments, the decision data of the third agent is critical, and disaster recovery can be achieved by configuring partial redundant storage in the first and second agents. For example, the decision recommendations generated by the third agent can be backed up in both the first and second agents. If the third agent fails, the backed-up decision data can be retrieved from the first or second agent, ensuring the continuity of enterprise decision-making.

[0074] In some embodiments, the entire system can employ a distributed architecture, with the first, second, and third agents deployed in different physical locations or cloud regions. If a natural disaster, network failure, or other issues occur in the region where one agent is located, the other agents can continue to operate and recover using backup data, ensuring overall system availability. For example, if the first agent is deployed in both a local data center and the cloud, if the local data center fails, the first agent in the cloud can continue to handle public service operations.

[0075] In some embodiments, the S1101 may include: the receiving of input data using natural language interaction, including: the first agent receiving multimodal data provided by the terminal device; the converting of the input data into standard format data, including: the first agent converting the multimodal data into standard format data; the determining of the optimal thinking chain for providing services, including: determining the optimal thinking chain for providing services based on at least one of user operation history, similar enterprise data and policy changes.

[0076] In some embodiments, the first agent is responsible for acquiring multimodal data from terminal devices. These devices include various office computers and mobile devices within the enterprise, as well as devices used by external partners for interaction. Multimodal data comes in a variety of forms, including text data such as corporate tax filing instructions and social security application documents; tabular data such as financial statements and employee information sheets; and image data such as scanned copies of business licenses and images of tax invoices.

[0077] To ensure efficient data transmission, FirstAgent utilizes advanced communication technologies, such as HTTP / HTTPS protocols, to ensure stable and secure data transmission, ensuring accurate data delivery even in complex network environments. Furthermore, it supports a variety of data transmission interfaces to accommodate the data output formats of various terminal devices, seamlessly integrating with both common USB interfaces and network-based wireless transmission.

[0078] After receiving the multimodal data, the first agent leverages powerful data processing tools and intelligent algorithms to convert the data format. For text data, natural language processing (NLP) techniques are employed. Word segmentation algorithms break down long texts into meaningful lexical units, while part-of-speech tagging and syntactic analysis further refine the text structure. Ultimately, the data is reorganized into the format required by the public service API.

[0079] When processing tabular data, the table header and data structure are first identified. Using a data mapping algorithm, the column names and data values ​​in the table are precisely matched and converted to the target format, ensuring that the data conforms to standards in both structure and content. For image data, optical character recognition (OCR) technology is used to extract text information. This is then combined with an image semantic understanding algorithm to convert the key information in the image into structured data, and then into a standard format. This entire conversion process is implemented using a carefully designed conversion engine that is highly scalable and can flexibly adapt to the ever-changing format requirements of public service APIs.

[0080] The First Agent's AI model plays a key role in determining the optimal execution path. The model integrates data from multiple sources. User operation history records an enterprise's past behavior with public services, including steps, frequency of use, problems encountered, and solutions. This data helps uncover user habits and preferences. Data from similar companies provides common patterns and best practices within the industry, enabling the model to leverage the successful experiences of other companies in similar scenarios. Policy change data tracks real-time adjustments to public service regulations, such as updates to tax and social security policies, to ensure that the execution path complies with the latest policy requirements.

[0081] In some implementations, the first model uses deep learning algorithms, such as recurrent neural networks (RNNs) and their variants, long short-term memory networks (LSTMs), and gated recurrent units (GRUs), to conduct in-depth analysis and learning of this data. The RNN family of algorithms can effectively process historical user operation data with time series characteristics and capture long-term dependencies in the data. Through continuous training, the model constructs a complex decision-making model that comprehensively considers multiple factors such as service response time, cost, and success rate, calculates the score of each potential execution path, and ultimately selects the path with the highest score as the optimal execution path. In the tax declaration scenario, the model will plan the optimal operational process from data submission to review feedback based on the latest tax policies, the company's past declaration habits, and the declaration experience of peers, ensuring efficient, accurate, and compliant declarations.

[0082] The First Agent is equipped with an advanced multimodal data receiving unit capable of establishing stable communication connections with various terminal devices. This unit supports multiple communication protocols and data transmission interfaces, ensuring the reception of multimodal data such as text, tables, and images, and performs preliminary verification and preprocessing on the data to ensure its integrity and accuracy.

[0083] The First Agent consists of a multimodal data parsing submodule and a format conversion submodule. The multimodal data parsing submodule uses technologies such as natural language processing and optical character recognition to extract key information from multimodal data. The format conversion submodule, based on pre-set standard formatting rules, uses algorithms such as data mapping and structural reorganization to convert the parsed data into a standard format recognizable by public service APIs, achieving data standardization. The First Agent integrates a First AI model that integrates user operation history, data from similar companies, and policy change data. This model uses deep learning algorithms to conduct in-depth mining and analysis of multi-source data to construct a multi-dimensional decision-making model. By calculating the comprehensive scores of different execution paths, it determines the optimal execution path for providing public services, ensuring efficient, accurate, and compliant service invocation. Improving data processing efficiency: This module enables rapid ingestion and efficient conversion of multimodal data, reducing data processing time and improving the overall efficiency of government-enterprise service interactions. Optimizing execution paths: Using intelligent algorithms, the optimal execution path is determined, reducing service invocation costs, increasing service success rates, and ensuring the smooth operation of enterprise businesses. Adapting to policy changes: This module promptly tracks policy developments and adjusts execution paths to ensure that enterprises remain compliant with the latest policy requirements and avoid business risks associated with policy changes.

[0084] In some embodiments, the enterprise collaborative decision-making system based on AI algorithms and government-enterprise APIs includes one or more subsystems that provide enterprise application services. Receiving input data interactively using natural language includes: a second agent receiving request data sent by a first subsystem; identifying the request data and determining user intent; determining, based on the user intent, a second subsystem associated with the request data; invoking the API of the second subsystem to provide the enterprise application service of the second subsystem; and the first and second subsystems providing different subsystems for enterprise application services. In this manner, subsystems can be divided to provide enterprise application services and public services.

[0085] In some embodiments, certain services may involve both enterprise application services and public services. In this case, the system will, based on user intent, include both execution nodes for providing enterprise application services and execution nodes for providing public services in a single thought chain, with the execution nodes for enterprise application services and public services arranged in an orderly fashion, thus achieving a one-stop service for both phases. In some embodiments, if a service involves both enterprise application services and public services, the scheduling node (or coordination node) of the collaborative decision-making system will interact with the first and second agents to coordinate processing between them. This thought chain involving both enterprise application services and public services can be generated by the first and second agents independently and then combined by the collaborative decision-making system.

[0086] In some embodiments, the use of an AI model to determine the service type involved in the input data includes: performing natural language understanding on the input data to extract the semantics of natural sentences; determining the service type based on the semantics and performing intent recognition; determining the constraints for providing services based on the service type and the recognized intent; planning the optimal thinking chain includes: planning a feasible thinking chain based on the service type and the constraints; and selecting the optimal thinking chain from the feasible thinking chains.

[0087] For example, a feasible thought chain is a thought chain that can provide the service required by the user, and an optimal thought chain is a thought chain that is most optimized to provide the service required by the user.

[0088] In some embodiments, upon determining that the enterprise collaborative decision-making system based on AI algorithms and government-enterprise APIs has established a business scenario that meets the service needs of the requesting enterprise, and upon determining that the requesting enterprise has not signed a contract for the corresponding business scenario, a push message is sent to the enterprise application service page. Sending a push message before a contract is signed can help users independently discover missing services and provide intelligent service setup.

[0089] In some embodiments, the method further includes: the third agent analyzing multi-source data using one or more of a variety of data analysis algorithms, such as time series analysis, association rule mining, and cluster analysis, to obtain decision data; wherein the multi-source data includes at least internal and external system data. This approach can achieve decision optimization based on multiple data sources and multiple data analysis methods.

[0090] In some embodiments, the method further includes: optimizing the planning model of the optimal thinking chain according to the service upgrade, user feedback and / or service provision effect evaluation of the enterprise collaborative decision-making system based on the AI ​​algorithm and the government-enterprise API.

[0091] In some embodiments, an enterprise collaborative decision-making method is implemented based on AI algorithms and government-enterprise APIs. The specific steps include: when a user inputs a business description through natural language, the agent will first identify the intent of the user's input and generate a thought chain; then it is decomposed into executable steps, and each decomposed step will correspond to a different API. The big model will automatically generate the corresponding API request input parameters and call the corresponding API step by step according to the decomposition steps. The result of each API call will be handed over to the big model for parsing and output in the corresponding format. Finally, the big model will perform unified reflection and verification on the execution results and give the final result.

[0092] For example, in the declaration scenario, the input text is: When the user enters my tax identity type as a tax officer, please log in to the Beijing District Electronic Taxation Bureau, the tax app mobile phone number 136xxxx5678 and the personal user password 123456, please help me file the small-scale VAT declaration for this period. The large model will analyze this sentence and call the following public service API interfaces in sequence: After calling the API interface through MCP, the JSON returned by the interface is given to the large model for processing, and reorganized to generate text, tables, pictures and other information that users can intuitively see.

[0093] The system provided by the embodiments of the present disclosure has one or more of the following features.

[0094] Business collaboration: Internal business systems (ERP, OA, CRM) operate independently, with inconsistent data formats and interfaces. When business processes involve multiple systems, manual switching between them and data transfer are required, leading to delays and inconsistencies in information flow. For example, when processing customer orders, after obtaining order information from the CRM system, the relevant data must be manually entered into the ERP system for inventory query and production scheduling. The OA system then initiates the approval process, and the business data ultimately enters the financial system for financial accounting and tax declaration. Finally, the data is submitted to the National Electronic Taxation Bureau for declaration and payment. This is inefficient and prone to data errors. Decision Support: The statistical analysis capabilities of traditional enterprise application service software are mostly based on simple data aggregation and report generation, lacking in-depth data mining and predictive capabilities. Some enterprises utilize rule-based AI systems: These systems develop a series of rules for specific business scenarios to achieve automated processing. In invoice review scenarios, rules such as invoice format and amount range are set, and programs determine whether invoices are compliant. However, these systems lack flexibility. When business rules change or new business scenarios emerge, a large number of rules need to be rewritten, resulting in high maintenance costs. Some enterprises attempt to use machine learning algorithms for tasks such as customer classification and sales forecasting. Clustering algorithms are used to classify customers, but due to issues such as uneven data quality and incomplete feature extraction, the model's accuracy and generalization capabilities are limited, making it unable to provide accurate and effective decision support for enterprises.

[0095] In some cases, manual processing of unstructured data is extremely inefficient and unable to meet the needs of enterprises for rapid processing of massive amounts of data. Manual operations are prone to fatigue and negligence, resulting in high data entry error rates, impacting the accuracy of subsequent business analysis and decision-making. Data is relatively siloed and difficult to unify, making it difficult to extract deep, critical information and limiting the value of data mining. Data silos are severe across systems, preventing real-time data sharing and interaction, increasing internal communication costs and operational risks. Inventory data in the Enterprise Resource Planning (ERP) system cannot be synchronized promptly with the Customer Relationship Management (CRM) system, resulting in inaccurate inventory information for sales staff and hindering the processing of customer orders. Use proxies to call APIs across various systems or connect to various systems through GUIs and RPA. Proxies serve as tools for all enterprise application service software, allowing them to be called by agents. The proxy call model does not conflict with the traditional single-system user login model, allowing for long-term coexistence.

[0096] Manually switching between multiple systems is not only inefficient but also prone to missed steps or incorrect data entry, impacting the consistency and accuracy of business processes. Traditional statistical analysis methods are unable to comprehensively analyze heterogeneous data from multiple sources, making it difficult to identify potential connections and trends between data, and thus failing to provide a comprehensive and accurate basis for business decision-making. Decision-making methods that rely on experience and intuition lack scientific and forward-looking perspectives. In a rapidly changing market environment, companies are prone to missing development opportunities or facing the risk of decision-making errors.

[0097] In view of this, this system provides large model retrieval enhancement and Figure 2 The system provided by the embodiment of the present disclosure is shown as follows: Phase 1: Data collection and preprocessing: Multi-source data acquisition: Comprehensively collect internal enterprise data, including structured data such as transaction records, financial statements, and customer information from business databases, as well as unstructured data such as contracts, project reports, and meeting minutes from document management systems. Simultaneously, external data such as industry news, policy and regulatory documents, and market research reports are collected to broaden the data source dimensions. Cleaning and denoising: Use data cleaning techniques to remove duplicate, invalid, and erroneous data. For unstructured text data, use regular expressions and text classification algorithms to remove irrelevant noise. Annotation and feature engineering: Data is annotated, such as contract data with contract type and key clauses. Feature selection and transformation are performed on structured data, such as normalizing numerical features and encoding categorical features. For unstructured text data, the word embedding model (BERT-Embedding) is used to convert text into vector representations and extract text features.

[0098] Phase II: Retrieval system construction: Vector Space Model Construction: We use a pre-trained language model (RoBERTa) based on the Transformer architecture to encode the data and convert the text data into a high-dimensional vector representation. We use metrics such as cosine similarity and Euclidean distance to calculate the similarity between vectors and construct a vector space model. Index structure design: To improve search efficiency, we designed an efficient inverted index. We established an index relationship between keywords in text data and their corresponding document identifiers (IDs), enabling rapid retrieval of documents containing relevant keywords. Furthermore, we incorporated data structures such as Bloom filters to reduce unnecessary search scope and improve search speed. Retrieval algorithm optimization: We introduced a deep learning-based retrieval algorithm and a retrieval model based on a convolutional neural network (CNN). This model automatically learns text feature representations to improve retrieval accuracy. By training a ranking model, we rerank retrieval results, placing more relevant documents at the top. Phase 3: Retrieval Enhancement Training: Fusion of Retrieval Information: During the fine-tuning of the large model, relevant information retrieved is fused with the original input text. The top N most relevant documents retrieved are concatenated with the user's question and used as input for the large model. This allows the large model to draw upon more context when generating responses, improving the accuracy and professionalism of the responses. It is important to note that the input to the large model fusion of retrieval information can include model context and retrieval-augmented generation (RAG) information.

[0099] Training strategy optimization: Utilizing a multi-task learning training strategy, we simultaneously optimize multiple tasks related to enterprise application services (text classification, question-answering, and text generation) during fine-tuning. We share model parameters to improve the model's generalization and adaptability to diverse tasks. Furthermore, we utilize transfer learning techniques to migrate model parameters pre-trained on large-scale general datasets to the enterprise application service domain, accelerating model convergence and reducing training data requirements. In addition, the thought chain planning can be as follows: First, task parsing: Natural language understanding: Using natural language processing technology, we perform lexical analysis, syntactic analysis, and semantic understanding of user-entered task requests. We use part-of-speech tagging (POS tagging) and named entity recognition (NER) to extract key entities and concepts in the task. Dependency parsing is used to determine the grammatical relationships between words and understand the structure and semantics of sentences. Task classification and intent identification: By training a classification model, we classify task requests into different business types (sales forecasting, project management, customer service). We identify user intent and determine whether they want to obtain information, perform an action, or conduct decision analysis. For the task request "Analyze this quarter's product sales," we identify the task type as sales data analysis and the intent as obtaining information. Constraint extraction: We extract various constraints (time limits, resource limits, quality requirements) from the task request. Thinking chain generation can include: Knowledge Graph Construction: Build a knowledge graph for enterprise application services, integrating internal business knowledge, process knowledge, and external industry knowledge. Nodes in the knowledge graph represent entities, and edges represent relationships between entities. This knowledge graph provides rich knowledge support for thought chain generation. Heuristic search algorithm: Based on the task analysis results, a heuristic search algorithm is used to search for possible thought chains in the knowledge graph. Taking the sales forecasting task as an example, based on historical sales data, market trends, and other information, combined with the business rules and causal relationships in the knowledge graph, a thought chain is generated that includes subtasks such as market research, data analysis, model selection, and forecast result evaluation. Generate diverse thought chains: To enhance decision diversity and flexibility, multiple thought chains are generated by adjusting search algorithm parameters and heuristic functions. These chains may differ in subtask selection, execution order, or focus, meeting diverse business needs and scenarios. For example, the process of mind linking is described as follows: 1. Natural language processing and intent recognition: A large-scale Transformer-based language model (Qwen3-32B) is used for intent recognition and semantic parsing. Model parameters are fine-tuned to adapt to specialized tax terminology (e.g., "declaration list," "small-scale VAT declaration," and "tax payment" in the declaration business domain). Attention mechanism weights are adjusted to enhance recognition of key entities (e.g., mobile phone number, password, and region). The semantic similarity between input text and predefined tax task templates (e.g., "Login → Authentication → Declaration → Submit") is calculated to assess whether the input contains necessary entities (e.g., mobile phone number, password, region, and tax ID number).

[0100] 2. Thinking chain generation and task decomposition: The thought chain production algorithm uses the hierarchical task network (HTN) planning algorithm combined with Monte Carlo tree search (MCTS) to generate multiple feasible thought chains, then performs path evaluation, calculates the semantic distance between the current path and the target task through the knowledge graph, and finally selects the optimal thought chain.

[0101] A thought chain can correspond to multiple paths, and the optimal path is selected using the following formula. Path evaluation formula f(path) = w1 × time cost(path) + w2 × API call cost(path) + w3 × task completion(path) Example thinking chain 1: 1. Call the "Get Verification Code" API → Parameter: phone=136xxxx5678 2. Call the "Login to the Electronic Tax Bureau" API → Parameters: phone, password, verification_code 3. Call the "Query Tax Types to be Declared" API → Parameters: user_id, tax_period 4. Call the "Fill in VAT Return" API → Parameters: user_id, tax_type, amount 5. Call the "Submit Declaration" API → Parameters: form_id, signature Example thinking chain 2: 1. Call the "Verify Tax Agent Identity" API → Parameters: phone, password, role=Tax Agent 2. Call the "Get Enterprise Tax ID" API → Parameters: user_id, region=Beijing 3. Call the "Login to Electronic Tax Bureau" API → Parameters: tax_id, password 4. Call the "Check Declaration Status" API → Parameters: tax_id, tax_period 5. Call the "Batch Fill in Declaration Form" API → Parameters: tax_id, tax_type_list, data 6. Call the "Submit and Pay" API → Parameters: form_id, payment_method It can be seen that the thinking chain information corresponding to the thinking chain includes the API information called and the parameters of calling the corresponding API. Multiple APIs are arranged in sequence.

[0102] 3. API Call and Parameter Generation: The large model generates API call parameters through Retrieval-Augmented Generation (RAG). It retrieves API templates matching the current subtask from the knowledge base (e.g., "Login API" requires phone, password, and areaCode parameters). It then maps the user input entities (e.g., "13612345678" → phone) to the templates, and uses constraints to filter invalid parameters (e.g., password length verification). The parameter confidence formula is: confidence(param) = semantic match(param, user input) × domain constraint satisfaction(param). Phase 4: Execution Monitoring and Dynamic Adjustment: Reinforcement learning (RL) is used to dynamically adjust the thought chain execution path and optimize strategies based on historical API call results. If an API call fails (such as a login prompt requiring a verification code), an alternative path is triggered (such as switching to SMS verification code login during login), and subsequent API calls continue normally.

[0103] Phase 5: Reflection, verification and result optimization: The large model checks the logic and integrity of the final result through a self-verification mechanism.

[0104] Final Evaluation and Selection: Build an Evaluation Metric System: A comprehensive evaluation metric system is established, encompassing multiple aspects of task completion, including feasibility, efficiency, cost, and accuracy. For each evaluation metric, a corresponding quantitative calculation method is developed. Feasibility is assessed by determining whether each subtask in the thought chain is achievable within the company's existing resources and technical capabilities. Efficiency is measured by estimating the time required to complete the thought chain. Cost considers resource consumption, including human, material, and financial resources. Evaluation Model Training: A large amount of historical task data, including task requests, thought chain execution processes, and results, is collected to train the evaluation model. The model uses machine learning algorithms such as support vector machines (SVMs) or deep learning algorithms (convolutional neural networks) to accurately assess the performance of different thought chains by learning patterns and patterns from historical data. Select the Optimal Thought Chain: Based on the evaluation model's output, the thought chain with the highest score is selected as the implementation plan. During the selection process, evaluation metrics can be weighted to emphasize certain metrics based on actual business needs. In urgent tasks, efficiency metrics are prioritized; in resource-limited scenarios, cost metrics are prioritized. 6. Execution and feedback: Task execution monitoring: Execute tasks according to the selected thinking chain, and monitor the task progress and status in real time during the execution process. By embedding a monitoring module in the business system, data is collected during the task execution process, such as the start time, end time, execution results, and other information of each subtask. Using visualization technology, the task execution progress is displayed to users and managers in the form of a chart, so that problems can be discovered and adjusted in a timely manner. Dynamic adjustment: During the task execution process, if the actual situation is found to be inconsistent with expectations, such as a subtask encounters difficulties and cannot be completed on time, or the external environment changes, the thinking chain is dynamically adjusted according to the preset adjustment strategy. Adjustment strategies can include replanning unexecuted subtasks, adjusting the execution order of subtasks, changing the execution method, etc.

[0105] Feedback and Learning: After the task is completed, the task execution results and related data are fed back to the system. The data during the task execution is analyzed, lessons learned are summarized, and the knowledge graph and evaluation model are updated.

[0106] System Architecture Design: The User Interface Layer provides diverse interaction methods, supporting multimodal input such as text, voice, and images to meet the needs of diverse users. Natural language interaction technology allows users to express task requests in natural language, allowing the system to understand and respond in real time. Furthermore, a visual interface design presents system processing results to users in an intuitive and easy-to-understand manner, such as charts, reports, and text summaries. The Business Logic Layer is the core processing layer of the system, comprising a large model processing module, a retrieval system module, and a thought chain planning module. The large model processing module is responsible for preliminary processing and semantic understanding of user input. It combines relevant information provided by the retrieval system and uses a fine-tuned large model to generate preliminary answers or suggestions. The retrieval system module rapidly retrieves relevant data from the knowledge base based on user input and the requirements of the large model processing module. The thought chain planning module parses, plans, and optimizes complex tasks, generating the optimal execution path and transmitting the planning results to the large model processing module and related business systems for execution. The Data Storage Layer builds an enterprise application service knowledge base to store various internal enterprise data, business rules, industry knowledge, and intermediate and historical data generated during system operation. The system uses a combination of a distributed database (HBase) and a graph database (Neo4j) to store structured and unstructured data, as well as knowledge graph data, to meet the storage and query requirements of different data types. Furthermore, data warehouse technology (Snowflake) is used to integrate and analyze data, providing data support for the system. Interaction process between multiple nodes in the system: User Input: Users enter task requests through the user interface layer. Requests can be natural language text, voice commands, or image information. Request Processing: The user interface layer passes task requests to the large model processing module in the business logic layer. The large model processing module first preprocesses the request, performing operations such as word segmentation and stop word removal, and then sends the preprocessed request to the retrieval system module. Retrieve relevant information: The retrieval system module searches the knowledge base based on the request content. Using the vector space model and index structure, it quickly finds relevant data documents and returns them to the large model processing module. Thought Chain Planning: When the task is complex, the large model processing module passes the request and retrieved relevant information to the thought chain planning module. The thought chain planning module parses the task, generates thought chains, evaluates and selects the optimal thought chain, and returns the planning results to the large model processing module. Generate results: The large model processing module combines the search information and thought chain planning results, and uses the fine-tuned large model to generate the final answer or suggestion. Result display: The large model processing module passes the generated results to the user interface layer, and the user interface layer presents the results to the user in an appropriate manner.

[0107] Reflection and verification: To ensure the correctness of the execution of the thinking chain, a closed-loop process of path evaluation → dynamic adjustment → reflection and verification will be carried out after each thinking chain execution.

[0108] like Figure 3 As shown, the embodiment of the present disclosure implements an enterprise collaborative decision-making system based on artificial intelligence (AI) algorithms and government-enterprise application programming interfaces (APIs), the system comprising: Receiving module 3101, for receiving input data using natural language interaction; The conversion module 3102 is used to convert the input data into standard format data; the standard format data is ordered; A determination module 3103 is configured to use an AI model to determine a service type related to the input data; the service type includes a public service and an enterprise application service; and the public service and the enterprise application service are both associated with one or more subtasks. Planning module 3104, for planning an optimal chain of thought; the optimal chain of thought corresponds to an optimal execution path; the optimal execution path includes one or more ordered execution nodes; the AI ​​model runs in the agent implementing the enterprise collaborative decision-making system based on the AI ​​algorithm and the government-enterprise API; The calling module 3105 is configured to call the API according to the optimal execution path, so that the called API provides the service required by the user based on the standard format data.

[0109] In some embodiments, the system includes a first agent, a second agent and / or a third agent; the first agent runs a first AI model and is used as an agent for providing public services; the second agent runs a second AI model and is used as an agent for providing enterprise application services; the third agent runs a third AI model and is used to provide decision data based on data within the system and / or data outside the system.

[0110] Combine Figure 4 As shown, an embodiment of the present application provides an electronic device, which can be a component device of a device management system based on task priority evaluation, including a processor 10 and a memory 11. Optionally, the device may also include a communication interface 12 and a bus 9. Among them, the processor 10, the communication interface 12, and the memory 11 can communicate with each other through the bus 9. The communication interface 12 can be used for information transmission. The processor 10 can call the logic instructions in the memory 11 to execute the above-mentioned embodiment of the method for realizing enterprise collaborative decision-making based on AI algorithm and government-enterprise API.

[0111] An embodiment of the present application provides a computer program product, which includes a computer program stored on a storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes the above-mentioned method for implementing enterprise collaborative decision-making based on AI algorithms and government-enterprise APIs.

[0112] The aforementioned computer-readable storage medium may be a transient computer-readable storage medium or a non-transitory computer-readable storage medium.

[0113] The technical solutions of the embodiments of the present application can be embodied in the form of a software product, which is stored in a storage medium and includes one or more instructions for causing a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in the embodiments of the present application. The aforementioned storage medium can be a non-transitory storage medium, including: a USB flash drive, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk, an optical disk, and other media that can store program code, or a transient storage medium.

[0114] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0115] The embodiments or examples disclosed in this application are not exhaustive, but are merely illustrations of some embodiments or examples, and are not intended to be specific limitations on the scope of protection of this disclosure. In the absence of contradiction, each step in a certain embodiment or example can be implemented as an independent example, and the steps can be arbitrarily combined. For example, a solution after removing some steps in a certain embodiment or example can also be implemented as an independent example, and the order of the steps in a certain embodiment or example can be arbitrarily exchanged. In addition, the optional methods or optional examples in a certain embodiment or example can be arbitrarily combined; in addition, the various embodiments or examples can be arbitrarily combined. For example, some or all of the steps in different embodiments or examples can be arbitrarily combined, and a certain embodiment or example can be arbitrarily combined with the optional methods or optional examples of other embodiments or examples.

[0116] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0117] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0118] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0119] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the existing technology, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc. Various media that can store program code.

[0120] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for implementing enterprise collaborative decision-making based on artificial intelligence (AI) algorithms and government-enterprise application programming interfaces (APIs), which is applied to implementing enterprise collaborative decision-making systems based on AI algorithms and government-enterprise APIs, and is characterized by: The method comprises: Receiving input data using natural language interaction; the input data is unordered; Converting the input data into standard format data; the standard format data is ordered; Using an AI model to determine the service type involved in the input data; the service type includes public services and enterprise application services; the public services and the enterprise application services are both associated with one or more subtasks; the AI ​​model runs in the agent that implements the enterprise collaborative decision-making system based on the AI ​​algorithm and the government-enterprise API; Planning an optimal thinking chain; the optimal thinking chain corresponds to an optimal execution path; the optimal execution path includes one or more execution nodes arranged in an orderly manner; The API is called according to the optimal execution path, so that the called API provides the service required by the user based on the standard format data.

2. The method according to claim 1, characterized in that The enterprise collaborative decision-making system based on AI algorithm and government-enterprise API includes a first agent, a second agent and / or a third agent; the first agent runs a first AI model and is used as an agent for providing public services; the second agent runs a second AI model and is used as an agent for providing enterprise application services; the third agent runs a third AI model and is used to provide decision data based on data within the system and / or data outside the system.

3. The method according to claim 2, characterized in that The receiving of natural language input data includes: the first agent receiving multimodal data provided by a terminal device; the converting of the input data into standard format data includes: the first agent converting the multimodal data into standard format data; the determining of the optimal thinking chain for providing services includes: determining the optimal thinking chain for providing services based on at least one of user operation history, similar enterprise data, and policy changes.

4. The method according to claim 2, characterized in that The enterprise collaborative decision-making system based on AI algorithms and government-enterprise APIs includes one or more subsystems that provide enterprise application services; the receiving of input data using natural language interaction includes: a second agent receiving request data sent by the first subsystem; identifying the request data and determining the user's intention; determining the second subsystem associated with the request data based on the user's intention; calling the API of the second subsystem to provide the enterprise application service of the second subsystem; the first subsystem and the second subsystem are different subsystems provided for enterprise application services.

5. The method according to any one of claims 1 to 4, characterized in that Determining the service type involved in the input data using the AI ​​model includes: performing natural language understanding on the input data to extract the semantics of natural sentences; determining the service type and performing intent recognition based on the semantics; and determining constraints for providing the service based on the service type and the recognized intent. The planning of the optimal thinking chain includes: planning feasible thinking chains according to the service type and the constraint conditions; and selecting the optimal thinking chain from the feasible thinking chains.

6. The method according to claim 5, characterized in that The method further comprises: When it is determined that the enterprise collaborative decision-making system based on AI algorithms and government-enterprise APIs has built a business scenario that meets the service needs of the requesting enterprise, and when it is determined that the requesting enterprise has not signed a contract for the corresponding business scenario, a push message is sent to the enterprise application service page.

7. The method according to any one of claims 2 to 4, characterized in that The method further comprises: The third agent uses one or more of a variety of data analysis algorithms such as time series analysis, association rule mining, and cluster analysis to analyze multi-source data to obtain decision data; wherein, the multi-source data includes at least data within the system and data outside the system.

8. The method according to any one of claims 2 to 4, characterized in that The method further comprises: According to the service upgrade, user feedback and / or service provision effect evaluation of the enterprise collaborative decision-making system based on AI algorithm and government-enterprise API, the planning model of the optimal thinking chain is optimized.

9. An enterprise collaborative decision-making system based on artificial intelligence (AI) algorithms and government-enterprise application programming interfaces (APIs), characterized by: The system comprises: A receiving module, configured to receive natural language input data; the input data is unordered; A conversion module, configured to convert the input data into standard format data; the standard format data is ordered; a determination module configured to use an AI model to determine the service type associated with the input data; the service type comprising a public service and an enterprise application service; the public service and the enterprise application service are each associated with one or more subtasks; and the AI ​​model is run in an agent implementing an enterprise collaborative decision-making system based on an AI algorithm and a government-enterprise API; A planning module, configured to plan an optimal thinking chain; the optimal thinking chain corresponds to an optimal execution path; the optimal execution path includes one or more execution nodes arranged in an orderly manner; The calling module is used to call the API according to the optimal execution path, so that the called API provides the service required by the user based on the standard format data.

10. The system according to claim 9, characterized in that The system includes a first agent, a second agent and / or a third agent; the first agent runs a first AI model and is used as an agent for providing public services; the second agent runs a second AI model and is used as an agent for providing enterprise application services; the third agent runs a third AI model and is used to provide decision data based on data within the system and / or data outside the system.

Images