Construction method of radio and television service fault diagnosis agent

Abstract

The invention relates to a method for constructing a radio and television service fault diagnosis agent, which comprises the following steps of: 1, completing the construction of a basic agent project based on an MCP architecture, realizing the construction, deployment and association of three components, namely Host, Client and Server, of the MCP, and calling and packaging an existing diagnosis service API; step 2, realizing Qamp; the retrieval of the A document is enhanced and optimized, so that the response accuracy of an agent component MCP Host is improved; 3, based on the semantic slicing technology, retrieval enhancement optimization of a service fault diagnosis guide manual is achieved, and the decision accuracy of an agent component MCP Host is improved; 4, management, association, prompt word optimization and vectorization of an API document are achieved based on the semantic slicing technology, the description detail degree of a tool under an agent component MCP Server on an MCP Host is improved, and the hit rate and the accuracy rate of the MCP Server diagnosis tool are called; the problem that the manual service process of the current traditional radio and television service fault diagnosis scene is low in efficiency is solved, and the expansion of the service fault diagnosis agent scene is completed by using the natural language.

Description

Technical Field

[0001] This invention relates to technologies such as MCP protocol intelligent agent construction, retrieval enhancement, semantic slicing, pre-trained models, and vector similarity retrieval, and particularly to a method for constructing an intelligent agent for fault diagnosis in broadcasting services. Background Technology

[0002] With the development of large language models, enterprises have already begun to deploy and integrate large language models, build general AI application scenarios such as AI knowledge bases and digital humans. Further development requires consideration of building more complex AI engineering applications. Currently, the manual service process in traditional broadcasting fault diagnosis scenarios suffers from low efficiency and high labor costs (see relevant case descriptions in the specific implementation details). Applications in specialized vertical fields like broadcasting fault diagnosis require solutions that enhance agent decision-making and response effectiveness. Summary of the Invention

[0003] To address the inefficiencies and high costs of traditional manual service processes in broadcasting fault diagnosis scenarios, this invention proposes a method for constructing an intelligent agent for broadcasting fault diagnosis based on the MCP architecture and semantic enhancement, leveraging existing resources such as Q&A, diagnostic operation manuals, and diagnostic microservices. This invention combines retrieval enhancement, semantic slicing, and pre-trained models to provide a solution.

[0004] The technical solution of this invention is as follows:

[0005] A method for constructing a broadcast service fault diagnosis intelligent agent, which uses a supporting broadcast service fault diagnosis system, Q&A documents, and a service fault diagnosis guide to construct a broadcast service fault diagnosis intelligent agent for diagnosing microservices, includes the following steps:

[0006] Step 1: Based on the MCP architecture, complete the construction of the basic intelligent agent project, realize the construction, deployment and association of the three MCP components Host, Client and Server; complete the encapsulation of the MCP Server tool calling method to call the existing diagnostic service API;

[0007] Step 2: Implement enhanced and optimized retrieval of Q&A documents; this is used to improve the accuracy of the intelligent agent component MCP Host in responding to basic questions based on user requests and contextual intent analysis.

[0008] Step 3: Based on semantic slicing technology, enhance and optimize the retrieval of the business fault diagnosis guidance manual to improve the accuracy of the intelligent agent component MCP Host in responding to complex problems and making decisions on calling diagnostic tools after analyzing user requests and contextual intent.

[0009] Step 4: Based on semantic slicing technology, implement the governance, association, prompt word optimization and vectorization of API documents; this is used to improve the level of detail in the description of MCP Host by the tools under the intelligent agent component MCP Server, and to improve the hit rate and accuracy of MCP Host decision-making calling MCP Server diagnostic tools in the field of broadcasting business fault diagnosis by associating with enhanced search content.

[0010] Furthermore, the specific implementation steps of step 2 are as follows:

[0011] Step 2.1: Design rules and templates to perform extraction, cleaning, and transformation operations on Q&A documents; generate JSON objects with unique identifiers (ID), questions, answers, document name source, and creation time parameters.

[0012] Step 2.2: Use the keyBERT method to extract keywords from the question and answer text, setting the length limit to 4-10 characters, and supplement the keyword parameters into the JSON object in Step 2.1;

[0013] Step 2.3: For the uniformly formatted text content contained in the above JSON object, use the BERT model to complete token segmentation, segmentation and position tag embedding, and vector calculation; obtain the vector and store it in the vector database.

[0014] Furthermore, the specific implementation steps of step 3 are as follows:

[0015] Step 3.1: Based on the semantics of "diagnostic problem", "diagnostic steps", and "functional points", perform document slicing, association, and structuring; form a JSON object of diagnostic guidance content;

[0016] Step 3.2: For the uniformly formatted text content contained in the JSON data object in Step 3.1, use the BERT model to complete token segmentation, conversion, and position tag embedding, and vector calculation; obtain the vector and store it in the vector database.

[0017] Furthermore, the specific implementation steps of step 4 are as follows:

[0018] Step 4.1: Slice the relevant semantic content of the API document based on the semantic "function points"; generate a JSON object containing the API content with "function point description", "API path", "tool name", "request data type" and "request parameters";

[0019] Step 4.2: Based on the JSON object in Step 4.1, generate the tool description content required by the tool discovery method list_tools() in the MCP Server; the generated content includes name, description, and inputSchema.

[0020] Furthermore, the repair reporting process after construction is completed is as follows:

[0021] Step 01: The agent receives the user's repair request and uses vector similarity to retrieve content from the vector database; the retrieved content is then combined with the context and handed over to the large model service for response decision and return; among which, the response decision is: responding based on the content matched in the vector database or providing a diagnostic tool invocation scheme that the agent believes is matched;

[0022] Step 02: The user refines their input based on the response or confirms the diagnostic tool.

[0023] Step 03: If the user refines the input, continue using Step 01 to advance to the next interaction;

[0024] Step 04: If the user confirms the diagnostic tool call plan, the agent will complete the diagnostic service API call through the MCP Server. The call result will be added to the context content and reused to advance the next interaction step 01.

[0025] Preferably, the diagnostic microservice is RESTful and the system is a B / S architecture; when the diagnostic microservice is deployed, it provides standard API documentation in accordance with the Swegger specification; the API documentation is a structured description, including HTTP methods, request data types, response data types, request examples, request parameters, response status, and response parameters.

[0026] The beneficial effects of this invention are as follows:

[0027] 1) Developers only need to focus on building the underlying diagnostic microservice API. Business fault diagnosis technical support personnel can build business fault diagnosis intelligent agent scenarios with low code by writing standardized documentation and tooltips.

[0028] 2) Service support personnel with different technical backgrounds and service capabilities can use natural language to expand the intelligent agent scenarios for business fault diagnosis, based on the requirements for writing Q&A documents and operation guide documents. This improves the timeliness of scenarios, the accuracy of training, and the novelty of user experience. Attached Figure Description

[0029] Figure 1 A schematic diagram of the manual response workflow for fault diagnosis in traditional broadcasting services;

[0030] Figure 2This is a schematic diagram of the architecture of the intelligent agent for diagnosing faults in broadcasting services according to the present invention. Detailed Implementation

[0031] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0032] Definitions:

[0033] OLT: optical line terminal.

[0034] PON port: Passive Optical Network Port, the fiber optic interface connected to the OLT.

[0035] ONU: Optical Network Unit. In the optical network scenario described in this article, it is described as a user terminal.

[0036] PPPoE: Point-to-Point Protocol over Ethernet, is a point-to-point protocol over Ethernet. By encapsulating the PPP protocol within the Ethernet framework, it enables functions such as authentication and encryption, and is widely used in broadband access (such as ADSL and fiber optic).

[0037] MCP: Model Context Protocol. It is an open-source protocol introduced by Anthropic, designed to enable the integration of large language models with external data sources and tools, and to establish a secure, bidirectional connection between large models and data sources.

[0038] REST stands for Representational State Transfer, a software architectural style. Among the three mainstream web service implementation schemes, REST-based web services are significantly simpler compared to the complex SOAP and XML-RPC, leading to an increasing number of web services adopting REST-style design and implementation.

[0039] B / S architecture: Browser / Server mode.

[0040] Semantic Chunking: Semantic slicing.

[0041] BERT model: BidirectionalEncoder Representations from Transformers, is a pre-trained language model based on the Transformer architecture.

[0042] JSON: JavaScript Object Notation. It is an open standard file and data exchange format designed based on a subset of ECMAScript. It is easy for humans to read and write, and also easy for machines to parse and generate.

[0043] Traditional cable TV service fault diagnosis scenarios are typically triggered by user reports. After receiving a user report, service support personnel complete the problem localization, diagnosis, and response according to the service process. Among these, common problems are usually identified and responded to by customer service personnel or frontline installation and maintenance personnel based on Q&A documents.

[0044] Examples of common Q&A document content:

[0045] Problem reference: Unable to access the internet (fiber optic network)

[0046] Answer Reference:

[0047] 1) Confirm whether the power indicator light of the user terminal ONU is constantly on, and guide the user to check the power-on status of the user terminal.

[0048] 2) Confirm whether the ONU optical signal light on the user terminal is constantly on, and guide the user to confirm the fiber optic connection status of the user terminal.

[0049] 3) Confirm whether the ONU diagnostic light on the user terminal is constantly on, and guide the user to enter the user terminal management interface to check the registration status.

[0050] 4) If the power supply and light indicator lights are constantly on, and the user terminal management interface registration is normal or the user cannot cooperate in checking, then a work order needs to be transferred to the back-end technical support for remote diagnosis.

[0051] Acceptance Rules:

[0052] 1) The user is an optical network user.

[0053] 2) The user status is normal (no pause or shutdown).

[0054] In point 4) above, remote diagnostics typically require the intervention of back-end technical support personnel. Back-end technical support personnel should complete further diagnostics and provide feedback according to the procedures and methods recorded in the business fault diagnosis guide.

[0055] Examples of common business fault diagnosis guide content:

[0056] Users who have subscribed to broadband or interactive home services in an optical network environment may experience internet access failures and need to report them for repair. The diagnostic steps are as follows:

[0057] 1) Open the cable TV service fault diagnosis system, access the user query function, and query using the user ID number or device serial number. The query results include basic user information, status information, device information, and package subscription information. This verifies whether the user status is normal and whether the device type is ONU.

[0058] 2) The system will automatically display the user's network type and PPPoE dial-up account information based on the user's basic information. If the user's address resource has an optical network identifier, an entry point for EPON or XGPON network status will be provided. If the user has subscribed to broadband products, an entry point for PPPoE dial-up diagnostics will be provided. This allows verification of the user's address resource identifier and PPPoE account information.

[0059] 3) By accessing the EPON / XGPON network status function, the system can automatically generate topology information from ONU to PON port to OLT based on the user's ONU device information. This can verify whether the user terminal has been correctly connected to the network and managed.

[0060] 4) Clicking on the ONU will initiate a real-time ONU query. The diagnostics will provide feedback on the PON port connection number, online status, whether it is misaligned, and transmit / receive power information. This can verify whether the ONU status is normal.

[0061] 5) Clicking the PON port will initiate an ONU whitelist query. The query will provide feedback on the whitelist configuration at the ONU activation location. This verifies whether the activation authorization location is normal.

[0062] 6) Clicking on the OLT will initiate an OLT information query. The query will provide basic OLT management information and service VLAN configuration details. This can verify whether the service channel configuration is normal.

[0063] 7) Return to the user's basic information page and access the PPPoE dialing diagnostic function. The system will automatically complete a simulated dialing based on the user's PPPoE account information and provide feedback on the dialing result. This verifies whether the account is properly activated.

[0064] 8) Provide diagnostic conclusions based on the above process and submit a work order.

[0065] The cable TV service fault diagnosis system described in the aforementioned business fault diagnosis guide is a B / S architecture system built by Oriental Cable based on multiple diagnostic microservices (REST style). When deployed, the diagnostic microservices provide standard API documentation in accordance with the Swegger specification. The API documentation includes structured descriptions of HTTP methods, request data types, response data types, request examples, request parameters, response status, and response parameters.

[0066] In summary, traditional fault diagnosis scenarios in the broadcasting industry rely on manual service processes assisted by Q&A documents, business fault diagnosis guides, and broadcasting business fault diagnosis systems (integrated diagnostic microservices) (see reference). Figure 1 ).

[0067] Referring to the traditional manual service process for fault diagnosis in broadcasting services, this invention provides a method for professionally responding to and optimizing tools for intelligent agents built from standard MCP projects based on existing documents and system resources. For example... Figure 2 As shown, the method construction process is as follows:

[0068] Step 1: Based on the MCP architecture, complete the construction of the basic intelligent agent project, realizing the construction, deployment, and association of the three core components of MCP: Host, Client, and Server. Complete the encapsulation of the MCP Server tool's calling methods for calling existing diagnostic service APIs.

[0069] Step 2: Implement enhanced and optimized Q&A document retrieval. This improves the accuracy of the MCP Host component in answering basic questions based on user requests and contextual intent analysis. The specific implementation steps are as follows:

[0070] Step 2.1: Design rules and templates to perform extraction, cleaning, and transformation operations on Q&A documents. This should generate JSON objects with a unique identifier (ID), question, answer, document name source, and creation time.

[0071] Step 2.2: Use the keyBERT method to extract keywords from the question and answer text, setting the length limit to 4-10 characters, and add the keyword parameters to the JSON object in Step 2.1.

[0072] Step 2.3: For the uniformly formatted text content contained in the above JSON object, use the BERT model to complete token segmentation, conversion, and position tag embedding, as well as vector calculation. The resulting vector is then stored in a vector database.

[0073] Step 3: Optimize and enhance the retrieval of the business fault diagnosis guidance manual based on semantic slicing technology (such as the BERT-CRF sequence labeling model). This improves the accuracy of the intelligent agent component MCP Host's responses to complex problems and its decision-making regarding diagnostic tool invocations based on user requests and contextual intent analysis. The specific implementation steps are as follows:

[0074] Step 3.1: Slice, associate, and structure the document based on the semantics of "diagnostic problem," "diagnostic steps," and "functional points." This results in (for example): a JSON object containing diagnostic guidance content: {"diagnostic problem": "optical broadband internet access problem," "diagnostic steps": [{"functional point description": "user query"}, {"functional point description": "network status diagnosis"}, {"functional point description": "ONU real-time query"}, ...]}

[0075] Step 3.2: For the uniformly formatted text content contained in the JSON data object in Step 3.1, use the BERT model to complete token segmentation, conversion, position tag embedding, and vector calculation. The resulting vector is then stored in a vector database.

[0076] Step 4: Implement API document governance, association, prompt word optimization, and vectorization based on semantic slicing technology. This is used to improve the detail of the description of the MCP Host by tools under the intelligent agent component MCP Server, and to improve the hit rate and accuracy of MCP Host decision-making in calling MCP Server diagnostic tools in professional vertical fields by associating with enhanced search content. The specific implementation steps are as follows:

[0077] Step 4.1: Slice the relevant semantic content of the API document based on semantic "function points". Generate a JSON object with (for example): {"Function point description":"ONU real-time query", "API path":" / api / snmp / ftth / onu / query", "Tool name":"snmp_ftth_onu_query", "Request data type":"application / x-www-form-urlencoded", "Request parameters":"mac",......}

[0078] Step 4.2: Based on the JSON object in Step 4.1, generate the tool description content required by the tool discovery method list_tools() in the MCP Server. The generated content must include name (tool name), description (feature description), and inputSchema (method, input parameters, output parameters, etc.).

[0079] In summary, the present invention describes a method for enabling intelligent agents built using the MCP solution to achieve accurate question answering of basic business fault knowledge bases and accurate workflow diagnosis and response for complex business faults in a professional vertical field (broadcasting) based on existing resources.

[0080] After optimization using the methods described above, the intelligent agent for fault diagnosis in broadcasting services should handle the following user repair request process:

[0081] Step 1: The agent receives a user's repair request and uses vector similarity to retrieve content from the vector database. The retrieved information, combined with context, is then submitted to a large model service for a response decision (either responding based on content matching the vector database or providing a diagnostic tool invocation scheme deemed suitable by the agent) and a return message. (The specialized vertical domain enhancement of the agent's response decision-making in this invention is reflected in this step.)

[0082] Step 2: The user refines the input based on the response or confirms the diagnostic tool.

[0083] Step 3: If the user refines the input, continue with Step 1 to advance to the next interaction.

[0084] Step 4: If the user confirms the diagnostic tool invocation plan, the agent completes the diagnostic service API call through the MCP Server. The call result will be added to the context content and reused to proceed to the next interaction step in Step 1. (The standardization of agent tool invocation and the enhanced result parsing capabilities of this invention are reflected in this step.)

[0085] The above-described embodiments are merely one implementation of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention should be determined by the appended claims.

Claims

1. A method for constructing an intelligent agent for fault diagnosis in broadcasting services, characterized in that, The construction of a broadcast service fault diagnosis intelligent agent for diagnosing microservices using the supporting broadcast service fault diagnosis system, Q&A documents, and service fault diagnosis guide includes the following steps: Step 1: Based on the MCP architecture, complete the construction of the basic intelligent agent project, realize the construction, deployment and association of the three MCP components Host, Client and Server; complete the encapsulation of the MCP Server tool calling method to call the existing diagnostic service API; Step 2: Implement enhanced and optimized retrieval of Q&A documents; this is used to improve the accuracy of the intelligent agent component MCP Host in responding to basic questions based on user requests and contextual intent analysis. Step 3: Based on semantic slicing technology, enhance and optimize the retrieval of the business fault diagnosis guidance manual to improve the accuracy of the intelligent agent component MCP Host in responding to complex problems and making decisions on calling diagnostic tools after analyzing user requests and contextual intent. Step 4: Based on semantic slicing technology, implement the governance, association, prompt word optimization and vectorization of API documents; this is used to improve the level of detail in the description of MCP Host by the tools under the intelligent agent component MCP Server, and to improve the hit rate and accuracy of MCP Host decision-making calling MCP Server diagnostic tools in the field of broadcasting business fault diagnosis by associating with enhanced search content.

2. The method for constructing a smart agent for fault diagnosis in broadcasting services according to claim 1, characterized in that, The specific implementation steps for step 2 are as follows: Step 2.1: Design rules and templates to perform extraction, cleaning, and transformation operations on Q&A documents; generate JSON objects with unique identifiers (ID), questions, answers, document name source, and creation time parameters. Step 2.2: Use the keyBERT method to extract keywords from the question and answer text, setting the length limit to 4-10 characters, and supplement the keyword parameters into the JSON object in Step 2.1; Step 2.3: For the uniformly formatted text content contained in the above JSON object, use the BERT model to complete token segmentation, segmentation and position tag embedding, and vector calculation; obtain the vector and store it in the vector database.

3. The method for constructing a smart agent for fault diagnosis in broadcasting services according to claim 1, characterized in that, The specific implementation steps for step 3 are as follows: Step 3.1: Based on the semantics of "diagnostic problem", "diagnostic steps", and "functional points", perform document slicing, association, and structuring; form a JSON object of diagnostic guidance content; Step 3.2: For the uniformly formatted text content contained in the JSON data object in Step 3.1, use the BERT model to complete token segmentation, conversion, and position tag embedding, and vector calculation; obtain the vector and store it in the vector database.

4. The method for constructing a smart agent for fault diagnosis in broadcasting services according to claim 1, characterized in that, The specific implementation steps for step 4 are as follows: Step 4.1: Slice the relevant semantic content of the API document based on the semantic "function points"; generate a JSON object containing the API content with "function point description", "API path", "tool name", "request data type" and "request parameters"; Step 4.2: Based on the JSON object in Step 4.1, generate the tool description content required by the tool discovery method list_tools() in the MCP Server; the generated content includes name, description, and inputSchema.

5. The method for constructing an intelligent agent for fault diagnosis in broadcasting services according to claim 1, characterized in that, The repair reporting process after construction is completed is as follows: Step 01: The agent receives the user's repair request and uses vector similarity to retrieve content from the vector database; the retrieved content is then combined with the context and handed over to the large model service for response decision and return; among which, the response decision is: responding based on the content matched in the vector database or providing a diagnostic tool invocation scheme that the agent believes is matched; Step 02: The user refines their input based on the response or confirms the diagnostic tool. Step 03: If the user refines the input, continue using Step 01 to advance to the next interaction; Step 04: If the user confirms the diagnostic tool call plan, the agent will complete the diagnostic service API call through the MCP Server. The call result will be added to the context content and reused to advance the next interaction step 01.

6. The method for constructing a smart agent for fault diagnosis in broadcasting services according to claim 1, characterized in that, The diagnostic microservice is RESTful, and the system is based on a B / S architecture. When the diagnostic microservice is deployed, it provides standard API documentation in accordance with the Swegger specification. API documentation is a structured description that includes HTTP methods, request data types, response data types, request examples, request parameters, response status, and response parameters.

Images