A cloud environment intelligent operation and maintenance method and system based on MCP protocol

The intelligent operation and maintenance method based on the MCP protocol solves the model forgetting problem by generating structured operation instructions and incremental context management, thereby improving the efficiency and security of operation and maintenance tasks and making it suitable for intelligent operation and maintenance in highly sensitive environments.

CN120639769BActive Publication Date: 2026-11-17南京中孚信息技术有限公司
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202510777950.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-06-11
Publication Date
2026-11-17
Estimated Expiration
2045-06-11

AI Technical Summary

Technical Problem

Existing intelligent operation and maintenance methods are prone to model forgetting the initial core task in long tasks, and are inefficient, lacking in security and flexibility in scenarios with severe information redundancy.

Method used

By adopting the MCP protocol, structured operation instructions are generated, user confirmation and incremental context management are introduced, and correlation analysis is performed in conjunction with a large model, which reduces the context length and improves the model accuracy.

Benefits of technology

Reducing context length in long tasks improves model performance in information-redundant scenarios, enhances security and flexibility, and meets operational needs in highly sensitive environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120639769B_ABST
    Figure CN120639769B_ABST
Patent Text Reader

Abstract

This invention relates to the field of large-scale model applications and provides a cloud environment intelligent operation and maintenance method and system based on the MCP protocol. The method includes: responding to user-input natural language task instructions; generating structured operation instructions based on the accessed large-scale model; sending the structured operation instructions to the user terminal for confirmation; if the user terminal modifies historical information or content to be confirmed, marking the differences, generating incremental context, and then sending it to the server for specific operation execution; after the user terminal submits modified information or makes a new request, performing correlation analysis on the large-scale model, retaining only relevant conclusions or summarizing key intermediate results, submitting them to the large-scale model, generating the next step decision, directly until the natural language task instructions are executed, and returning to the user terminal. This invention not only significantly reduces the length of the context but also alleviates the problem of model amnesia in long tasks.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of large-scale model applications, and in particular to a cloud environment intelligent operation and maintenance method and system based on the MCP protocol. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Current intelligent operations and maintenance (O&M) systems use large models to make step-by-step decisions for tasks and execute them based on human feedback. However, this approach has several drawbacks. First, its historical session management model is simplistic. In long task chains, dialogue-based error correction leads to excessively long model contexts, particularly in O&M scenarios requiring extensive environmental information (e.g., to obtain the PID of a specific process, the model first retrieves all operating system processes, resulting in a large amount of useless data, which is then retained in the context). This can easily lead to forgetting the initial core task, resulting in poor performance when using smaller models. This simple interactive architecture is unsuitable for O&M scenarios requiring extensive input and containing a large amount of redundant and invalid information. Second, it only supports custom non-standard protocols for unstructured parsing, leading to a high error rate for small models. Third, related task tools require customized development. This approach, which relies on simple interaction with a large model, is suitable for short tasks and high-information-density scenarios, but it is not applicable to O&M interactions involving long tasks and significant information redundancy (debugging logs, file attribute information, etc.).

[0004] Another type of intelligent operations and maintenance (O&M) involves pre-orchestrating existing functions and combining them with large-scale model analysis to solve problems. This method has two main drawbacks: First, it requires pre-orchestrating fixed tasks, resulting in poor flexibility. It can only execute specific types of tasks based on specific data, typically used for environmental remediation after specific metrics become abnormal, and cannot proactively address new problems. Second, it executes pre-orchestrated tasks as a whole, without a manual secondary confirmation process. The large-scale model may then perform dangerous operations, posing security and compliance risks. For example, to address excessive storage usage, the large-scale model might execute high-risk commands, leading to the accidental deletion of critical databases.

[0005] In addition, existing intelligent operation and maintenance also supplements human knowledge through command completion and large model knowledge question answering, and uses large models as a way of knowledge retrieval. This type of operation and maintenance has a high degree of human intervention, requiring operation and maintenance personnel to have some knowledge in this area. Moreover, it only uses large models to query information, and the overall process still relies on humans to make general decisions and input and execute actual commands, resulting in a low degree of automation. Summary of the Invention

[0006] To address the technical problems mentioned above, this invention provides a cloud environment intelligent operation and maintenance method and system based on the MCP protocol. This invention can not only significantly reduce the context length, but also alleviate the model's forgetting problem in long tasks. It can also perform well in small models with limited context length, alleviating the model forgetting problem caused by a large amount of data input in the previous large model interaction structure (the model forgets the original core task after inputting too much content), and meeting the needs of large environmental data input in operation and maintenance scenarios.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of this invention provides a cloud environment intelligent operation and maintenance method based on the MCP protocol.

[0008] A cloud environment intelligent operation and maintenance method based on the MCP protocol includes: In response to user input of natural language task instructions, structured operation instructions are generated based on the accessed large model; The structured operation instructions are sent to the user for confirmation. If the user modifies the historical information or the content to be confirmed, the differences are marked, an incremental context is generated, and then sent to the server to execute the specific operation. After the user submits modified information or makes a new request, the system connects to the large model to perform correlation analysis, retains only the relevant conclusions or summarizes the key intermediate results, submits them to the large model to generate the next decision, and then directly executes the natural language task instructions before returning to the user.

[0009] Furthermore, in response to the natural language task instructions input by the user, a structured operation instruction is generated based on the accessed large model; the method includes: for large models that support Function Calling, directly calling the large model interface to generate structured operation instructions.

[0010] Furthermore, in response to the natural language task instruction input by the user, a structured operation instruction is generated based on the accessed large model. The method also includes: for large models that do not support Function Calling, a prompt word template library is selected through prompt word engineering. The large model generates a natural language instruction template through the prompt word template library, and combines it with regular expressions to parse key parameters to obtain a structured operation instruction. If the parsing fails, the large model optimizes the prompt word template library based on the failure result and retryes.

[0011] Furthermore, the incremental context is a context that retains only key modification records.

[0012] Furthermore, the client supports direct modification of erroneous steps in historical conversations.

[0013] Furthermore, the client supports comparing the corrected content from the user's end with the original erroneous content, using built-in templates to perform error correction analysis on the modified points, and generating enhanced prompt word correction results.

[0014] Furthermore, the server employs a hierarchical storage strategy to store historical sessions, archiving raw natural language commands, structured commands, execution results, and user correction records according to task chain IDs.

[0015] Furthermore, the server-side supports rapid backtracking of the entire process during auditing.

[0016] A second aspect of the present invention provides a cloud environment intelligent operation and maintenance system based on the MCP protocol.

[0017] A cloud environment intelligent operation and maintenance system based on the MCP protocol includes: The user interface is used to input natural language task instructions; The server-side component is used to convert natural language task instructions into structured operation instructions based on the large model being accessed, and then send them to the user for confirmation. On the user side, it is used to confirm structured operation instructions. When modifying historical information or content to be confirmed, it marks the differences, generates an incremental context, and then sends it to the server to execute the specific operation. The server-side component is used to perform correlation analysis on the large model after the user submits modified information or makes a new request. It retains only the relevant conclusions or summarizes the key intermediate results, submits them to the large model, generates the next decision, and then returns to the user after executing the natural language task instructions.

[0018] A third aspect of the present invention provides a cloud environment intelligent operation and maintenance system based on the MCP protocol. A cloud environment intelligent operation and maintenance system based on the MCP protocol includes: The user interaction module is configured to: respond to user-input natural language task instructions, generate structured operation instructions based on the accessed large model; send the structured operation instructions to the user for confirmation; if the user modifies historical information or content to be confirmed, mark the differences, generate incremental context, and then send it to the server to execute the specific operation; after the user submits modified information or makes a new request, perform correlation analysis with the large model, retain only relevant conclusions or summarize intermediate key results, submit them to the large model to generate the next decision, and directly return to the user after the natural language task instructions have been executed. The model processing module is configured as follows: for large models that support Function Calling, it directly calls the large model interface to generate structured operation instructions; for large models that do not support Function Calling, it selects a prompt word template library through prompt word engineering, and the large model generates natural language instruction templates through the prompt word template library, combines regular expressions to parse key parameters, and obtains structured operation instructions; if parsing fails, the large model optimizes the prompt word template library based on the failure result and retryes.

[0019] Compared with the prior art, the beneficial effects of the present invention are: This invention can not only significantly reduce the context length, but also alleviate the model's memory loss problem in long tasks. It can also perform well in small models with limited context length, and alleviate the model memory loss problem caused by a large amount of data input in the previous large model interaction structure (the model forgets the original core task after inputting too much content), and meet the needs of a large amount of environmental data input in operation and maintenance scenarios.

[0020] This invention introduces a manual modification and verification process after the large model is generated, which can improve the security and accuracy of the output results of the large model.

[0021] This invention does not require pre-programming of fixed tasks, and can execute relevant instructions based on different input data, which improves the flexibility of implementation and enables intelligent operation and maintenance after new problems are discovered.

[0022] This invention can summarize the context through a large model, extract the context related to the input command, focus more on intent analysis, reduce the length of irrelevant context, allow the large model to focus on the core task, and improve the accuracy and correctness of the large model. Attached Figure Description

[0023] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0024] Figure 1 This is a flowchart illustrating an intelligent cloud environment operation and maintenance method based on the MCP protocol, as shown in an embodiment of the present invention. Figure 2 This is a logical block diagram of a cloud environment intelligent operation and maintenance method based on the MCP protocol, as shown in an embodiment of the present invention. Figure 3 This is a structural diagram of a cloud environment intelligent operation and maintenance system based on the MCP protocol, as shown in an embodiment of the present invention. Detailed Implementation

[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0026] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0027] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0028] To facilitate understanding of the technical solutions of this invention, some technical terms involved in this invention will be introduced below.

[0029] Figure 1 This is a flowchart illustrating an intelligent cloud environment operation and maintenance method based on the MCP protocol, as shown in an embodiment of the present invention; see reference. Figure 1 The method includes: In response to user input of natural language task instructions, structured operation instructions are generated based on the accessed large model; The structured operation instructions are sent to the user for confirmation. If the user modifies the historical information or the content to be confirmed, the differences are marked, an incremental context is generated, and then sent to the server to execute the specific operation. After the user submits modified information or makes a new request, the system connects to the large model to perform correlation analysis, retains only the relevant conclusions or summarizes the key intermediate results, submits them to the large model to generate the next decision, and then directly executes the natural language task instructions before returning to the user.

[0030] This invention can not only significantly reduce the context length, but also alleviate the model's memory loss problem in long tasks. It can also perform well in small models with limited context length, and alleviate the model memory loss problem caused by a large amount of data input in the previous large model interaction structure (the model forgets the original core task after inputting too much content), and meet the needs of a large amount of environmental data input in operation and maintenance scenarios.

[0031] Figure 2 This is a logical block diagram of a cloud environment intelligent operation and maintenance method based on the MCP protocol, as shown in an embodiment of the present invention; refer to Figure 2 The method includes: Step 1: Users input natural language task commands through a visual interface, such as "Fix database master-slave latency".

[0032] Step 2: The system first determines whether the large language model being accessed supports Function Calling: For supported native models (such as GPT-4), its interface is directly called to generate structured operation instructions; for unsupported open-source models (such as DeepSeekR1), natural language instruction templates are generated through a dynamic prompt word template library, and key parameters are extracted using regular expressions and converted into structured instructions.

[0033] Step 3: The system will present the generated instructions to the user's visual interface, where the user can review and correct them. At this point, the user can continue to communicate with the system using natural language, requesting the system to correct the commands automatically, or edit and correct errors in the historical Q&A.

[0034] Step 4: After user confirmation, the system MCP client encapsulates the instruction into a standard protocol request and routes it to the corresponding MCP server (SSH, K8S, or a custom tool) to perform the specific operation.

[0035] Step 5: The server returns standardized results (status codes, logs, output, etc.) to the system core, where they are uniformly recorded by the compliance audit layer. The compliance tag engine marks the risk level (the system has default classification and grading standards, which are obtained by processing with relevant algorithms (regular expressions, similarity, large models, etc.)) and then issues a confirmation strategy (simple confirmation, confirmation after authentication, etc.). The confirmation information and decision information are submitted to the user for confirmation and execution.

[0036] Step 6: When a user submits modified information or context, the historical conversation management module intelligently marks error points, emphasizes the modified content, and generates incremental context. Simultaneously, it retains only the core task context information and submits it to the model to generate the next decision. The historical conversation management module records complete dialogues and modification points. This module is used to enhance user modifications and, based on user questions or modifications, collapses irrelevant historical dialogues to reduce the scope of context.

[0037] This invention enhances management through intelligent conversation context, focuses on core tasks, significantly reduces model context length, and alleviates problems such as model amnesia during long tasks.

[0038] Step 7: Repeat the above dialogue and confirmation steps until the model determines that all user input tasks have been completed, then the model analyzes and outputs the results back to the user.

[0039] The entire process operates through a closed loop: "natural language input → historical conversation processing → model decision-making → manual confirmation → protocol execution → result feedback → compliance audit." This balances automation efficiency with manual control, meeting the operational needs of highly sensitive, heterogeneous environments and long-duration, big data input environments such as government and cloud-native sectors. Furthermore, third-party tool developers can dynamically register new servers (such as existing cloud platform APIs) via the MCP protocol, expanding the automation ecosystem.

[0040] In another implementation, the present invention can use plugins such as Cline to interact with large models, but it can only support models that use function calling, and does not support operations such as editing dialogues or correcting commands. It cannot trace long-term dialogues or operation behaviors, nor does it support intelligent filtering out useless context information.

[0041] In another implementation, the present invention can use a cursor instead of an external MCP tool, but it does not support private deployment and can only interact with the local MCP server, without the ability to access remote servers.

[0042] The above combination Figure 1 The cloud environment intelligent operation and maintenance method based on the MCP protocol provided in the embodiments of the present invention has been described in detail. Next, the cloud environment intelligent operation and maintenance system based on the MCP protocol provided in the embodiments of the present invention will be described in conjunction with the accompanying drawings.

[0043] Figure 3 This is a structural diagram of a cloud environment intelligent operation and maintenance system based on the MCP protocol, as shown in an embodiment of the present invention; refer to Figure 3 The system includes: The user interface is used to input natural language task instructions; The server-side component is used to convert natural language task instructions into structured operation instructions based on the large model being accessed, and then send them to the user for confirmation. On the user side, it is used to confirm structured operation instructions. When modifying historical information or content to be confirmed, it marks the differences, generates an incremental context, and then sends it to the server to execute the specific operation. The server-side component is used to perform correlation analysis on the large model after the user submits modified information or makes a new request. It retains only the relevant conclusions or summarizes the key intermediate results, submits them to the large model, generates the next decision, and then returns to the user after executing the natural language task instructions.

[0044] This invention supports the native MCP protocol, has a unified tool interface, can be directly integrated into the existing standard ecosystem, and is easy to extend.

[0045] This invention adopts a dual-path adaptation architecture and a layered architecture. For large models that do not support function calling, and for large models that do support function calling, different methods are used. It performs compatibility processing for large models that do not support function calling in order to adapt to more models and achieve model independence.

[0046] This invention supports a closed-loop decision-making process based on differentiated risk assessment, enabling human-machine collaboration.

[0047] According to embodiments of the present invention, the cloud environment intelligent operation and maintenance system based on the MCP protocol can correspond to the execution of the methods described in the embodiments of the present invention, and the above and other operations and / or functions of each module of the cloud environment intelligent operation and maintenance system based on the MCP protocol are respectively for implementing Figure 1 For the sake of brevity, the corresponding processes of each method in the code will not be elaborated here.

[0048] As another implementation, the cloud environment intelligent operation and maintenance system based on the MCP protocol described in this invention further includes: a user interaction module, a model processing module, a distributed MCP service module, and a compliance audit module.

[0049] Distributed MCP Service Module: This module uses the MCP (Model Context Protocol) to define tool interfaces, abstracting remote SSH commands, Kubernetes APIs, and other operations into standardized services. It supports dynamic tool registration via configuration files without requiring intrusive code modifications. This design is similar to the "plug-and-play" concept in the industrial sector, enabling third-party tool developers to quickly integrate into the ecosystem and promoting the modular development of intelligent operation and maintenance tools.

[0050] Model processing module: For models that support Function Calling (such as GPT-4), it directly calls the native interface; for models that do not support it (some open-source LLMs, such as DeepSeekR1), it introduces a semantic parsing adaptation layer. Through prompt word engineering, it dynamically selects appropriate prompt word templates, allowing the model to generate call instructions that conform to the format, and combines regular expressions to parse the returned results, achieving compatibility with various large models. When parsing fails, the model processing layer will have the model re-analyze the failure results and optimize the prompt word template, and then retry.

[0051] Historical Session Management Module: To meet the needs of operation and maintenance scenarios, in addition to supporting traditional continuous contexts, this system has the following key technical and system design points: Context editing and compression: Users can directly modify erroneous steps (such as incorrect commands and error messages) in historical conversations. The system automatically marks the differences and generates enhanced incremental context (only key modification records are retained, rather than the full context of the modification comparison), which can greatly avoid the accumulation of redundant information.

[0052] Error point reinforcement learning: The system compares the user's corrected content with the original error content, uses the system's built-in templates to perform error correction analysis on the modified points and generates reinforcement prompt words to correct the results, making it easier for the large model to grasp the corrected content and the focus of subsequent analysis.

[0053] Intelligent context filtering: By connecting to a large model for correlation analysis, it automatically filters irrelevant data in intermediate execution results (such as debugging information and irrelevant file information in logs), retains only relevant conclusions or summarizes key intermediate results (such as retaining key status codes, exception summaries, and information required to complete core tasks), and automatically collapses irrelevant intermediate dialogue data on the interface to ensure that the context focuses on the core task and alleviates the forgetting phenomenon of large models.

[0054] The process of relevance analysis includes: first, using semantic similarity analysis (based on vocabulary set statistics, word embedding, etc.) to analyze new question content; if existing content is similar, it is marked as retained; then, using a large model to perform intent analysis on new dialogue content or modification points, folding intermediate data or dialogues unrelated to intent. For example, if a user wants to query the logs of server A, then data from queries on server B and other data from server A in historical dialogues will be marked as useless data. This part is mainly achieved through entity labeling; intermediate data obtained from the user's historical dialogues will be tagged with multiple labels, such as server A data, log data, statistical data, application A data, etc.

[0055] The above solution not only significantly reduces the context length but also alleviates the model's memory loss problem in long tasks. It can also perform well in small models with limited context length, mitigating the model memory loss problem caused by a large amount of data input in the previous large model interaction structure (the model forgets the original core task after inputting too much content), and meeting the needs of a large amount of environmental data input in operation and maintenance scenarios.

[0056] User Interaction Module: In terms of the interaction mechanism, users select tools and input task objectives through a visual interface. After the large model generates the tool's call chain, manual confirmation is required for execution. The results are returned to the model for analysis and iteration, and all operation processes are automatically recorded for backtracking. In terms of scenario value, in highly sensitive fields such as finance and data security, this mechanism retains the final decision-making power for key operations while leveraging AI to improve efficiency, complying with the risk control requirements of the "Administrative Measures for Generative Artificial Intelligence Services." In terms of the audit trail mechanism, user confirmation logs are recorded, and historical sessions adopt a hierarchical storage strategy. Original natural language commands, structured commands, execution results, and user correction records are all archived by task chain ID, supporting rapid backtracking of the entire process during audits, meeting industry regulatory requirements.

[0057] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A cloud environment intelligent operation and maintenance method based on the MCP protocol, characterized in that, include: In response to user input of natural language task instructions, structured operation instructions are generated based on the accessed large model; The method includes: for large models that support Function Calling, directly calling the large model interface to generate structured operation instructions; for large models that do not support Function Calling, through prompt word engineering, selecting a prompt word template library, the large model generates natural language instruction templates through the prompt word template library, and combines regular expressions to parse key parameters to obtain structured operation instructions; if parsing fails, the large model optimizes the prompt word template library based on the failure result and retryes. The structured operation instructions are sent to the user for confirmation. If the user modifies the historical information or the content to be confirmed, the differences are marked, an incremental context is generated, and then sent to the server to execute the specific operation. After the user submits modified information or makes a new request, the system connects to the large model to perform correlation analysis, retains only the relevant conclusions or summarizes the key intermediate results, submits them to the large model to generate the next decision, and then directly executes the natural language task instructions before returning to the user.

2. The cloud environment intelligent operation and maintenance method based on the MCP protocol according to claim 1, characterized in that, The enhanced incremental context is a context that retains only key modification records.

3. The cloud environment intelligent operation and maintenance method based on the MCP protocol according to claim 1, characterized in that, The client application supports direct modification of erroneous steps in historical conversations.

4. The cloud environment intelligent operation and maintenance method based on the MCP protocol according to claim 1, characterized in that, The client supports comparing the corrected content from the user's end with the original erroneous content, using built-in templates to perform error correction analysis on the modified points, and generating enhanced prompt words to correct the results.

5. The cloud environment intelligent operation and maintenance method based on the MCP protocol according to claim 1, characterized in that, The server uses a hierarchical storage strategy to store historical sessions, archiving the original natural language commands, structured commands, execution results, and user correction records according to the task chain ID.

6. The cloud environment intelligent operation and maintenance method based on the MCP protocol according to claim 1, characterized in that, When the server supports auditing, it can quickly trace back the entire process.

7. A cloud environment intelligent operation and maintenance system based on the MCP protocol, characterized in that, The cloud environment intelligent operation and maintenance method based on the MCP protocol as described in any one of claims 1-6 includes: The user interface is used to input natural language task instructions; The server-side component is used to convert natural language task instructions into structured operation instructions based on the large model being accessed, and then send them to the user for confirmation. On the user side, it is used to confirm structured operation instructions. When modifying historical information or content to be confirmed, it marks the differences, generates an incremental context, and then sends it to the server to execute the specific operation. The server-side component is used to perform correlation analysis on the large model after the user submits modified information or makes a new request. It retains only the relevant conclusions or summarizes the key intermediate results, submits them to the large model, generates the next decision, and then returns to the user after executing the natural language task instructions.

8. A cloud environment intelligent operation and maintenance system based on the MCP protocol, characterized in that, The cloud environment intelligent operation and maintenance method based on the MCP protocol as described in any one of claims 1-6 includes: The user interaction module is configured to: respond to user-input natural language task instructions, generate structured operation instructions based on the accessed large model; send the structured operation instructions to the user for confirmation; if the user modifies historical information or content to be confirmed, mark the differences, generate incremental context, and then send it to the server to execute the specific operation; after the user submits modified information or makes a new request, perform correlation analysis with the large model, retain only relevant conclusions or summarize intermediate key results, submit them to the large model to generate the next decision, and directly return to the user after the natural language task instructions have been executed. The model processing module is configured as follows: for large models that support Function Calling, it directly calls the large model interface to generate structured operation instructions; for large models that do not support Function Calling, it selects a prompt word template library through prompt word engineering, and the large model generates natural language instruction templates through the prompt word template library, combines regular expressions to parse key parameters, and obtains structured operation instructions; if parsing fails, the large model optimizes the prompt word template library based on the failure result and retryes.

Citation Information

Patent Citations

  • Method and system for realizing network configuration processing of intelligent equipment in mode of directly interacting with user

    CN111343022A

  • Question generation system based on knowledge graph and interrogative word driving

    CN114168749A