Business performance anomaly detection and disposal method based on AI intelligent agent

Abstract

The invention provides a service performance anomaly detection and disposal method based on an AI agent, and belongs to the field of mobile communication. Establishing a professional knowledge base for the business themes; packaging and developing a tool set with independent functions according to business function requirements; the method comprises the following steps: constructing an AI agent by adopting a large model platform, integrating a professional knowledge base and an intelligent tool set, and arranging a business analysis processing flow of the intelligent tool set; and the AI intelligent agent can automatically execute a business analysis processing flow, carry out anomaly detection and analysis processing on business performance, and output a comprehensive analysis conclusion. Automation of the disposal process is achieved through the AI agent, manual analysis is replaced, the working difficulty of complaint analysis of a user is lowered, the usability of a complaint analysis tool is improved, and the efficiency and accuracy of complaint analysis processing are improved.

Description

Technical Field

[0001] The present invention relates to the field of mobile communication technologies, and in particular to a method for detecting and handling business performance anomalies based on an AI agent. Background Art

[0002] Mobile Internet services have become the main form of communication and entertainment for people. Mobile Internet service quality detection and complaint handling are important tasks for major operators. There are many links that affect the service, which are generally divided into user terminals, base station cells, core networks and content source sides. For the overall service quality detection and evaluation of the terminal side, base station cells, core networks and content source sides, the business KPI and KQI indicator models of the dimensions of terminals, cells, core network elements and services are currently mainly established based on DPI data for anomaly detection. Anomaly detection based on DPI data is generally achieved by setting thresholds for indicators, which can be fixed thresholds or dynamic thresholds. For the analysis of complaints perceived by a single user, a tool set such as single user overview information, multi-dimensional indicator analysis, and detailed list query is mainly constructed based on user-level DPI detailed list data. These tool sets are used manually to traverse information of various dimensions for single user perception analysis.

[0003] The shortcomings of the current single-user perception complaint analysis method are:

[0004] The use of current tool sets and analysis process methods rely heavily on expert experience, and often require the accumulation of deep signaling knowledge and big data knowledge to achieve efficient use of tools and locate key information from a huge amount of information; the training of analysts often takes a long time and is costly; the tool set is difficult to use and promote, and it is difficult to play its due role in the front line. Moreover, manual analysis and processing is slow and prone to errors, making it difficult to ensure a certain efficiency and accuracy. Summary of the invention

[0005] In order to solve the problems existing in the prior art, the present invention provides a business performance anomaly detection and handling method based on AI intelligent agent, which realizes the automation of the handling process by calling the intelligent tool set through the AI ​​intelligent agent, replaces manual analysis, reduces the difficulty of user complaint analysis, improves the usability of complaint analysis tools, and improves the efficiency and accuracy of complaint analysis.

[0006] In order to achieve the above technical objectives, the present invention adopts the following technical solutions:

[0007] A method for detecting and handling business performance anomalies based on an AI agent, comprising the following steps:

[0008] S1. Establish a professional knowledge base for business topics;

[0009] S2. Encapsulate and develop intelligent tool sets based on business function requirements;

[0010] S3. Use the big model platform to build an AI agent that integrates the professional knowledge base and intelligent tool set and orchestrates the business analysis and processing flow of the intelligent tool set.

[0011] S4. The AI ​​agent automatically executes the business analysis and processing flow, detects and analyzes anomalies in business performance, and outputs comprehensive analysis conclusions.

[0012] Preferably, the establishment of the professional knowledge base includes a business process knowledge base, a business rule knowledge base, and an expert experience knowledge base.

[0013] Preferably, the intelligent toolset is an API-style toolset.

[0014] Preferably, the encapsulation of the API-style toolset is specifically as follows: first, a capability list of existing API tools is sorted out, and then encapsulated into an API toolset with independent functions according to business requirements.

[0015] Preferably, the development of the API-style toolset is specifically to develop personalized dedicated functional APIs for different business logics in combination with business rules and expert experience and knowledge.

[0016] Preferably, the large model platform adopts the DIFY large model platform.

[0017] Preferably, the process of constructing an AI agent in step S3 includes:

[0018] S31, select cloud deployment or local deployment of the DIFY large model platform;

[0019] S32, select the model and create an Agent;

[0020] S33. Upload professional knowledge base and set permissions;

[0021] S34. Integrate and orchestrate the business analysis processing flow of the intelligent tool set.

[0022] Preferably, the step of integrating and arranging the business analysis processing flow of the intelligent tool set in step S34 includes:

[0023] S341. Sort out the business analysis process based on business scenarios;

[0024] S342. Create a workflow on the DIFY large model platform;

[0025] S343. For the business analysis process, create multiple nodes in the workflow, and connect the intelligent tool set in the nodes.

[0026] Preferably, in step S343, multiple nodes are created in the workflow, and the specific steps of connecting the intelligent tool set in the node are:

[0027] S3431, creating a start node, the start node connection function is a smart tool set for reading data;

[0028] S3432. Create an intermediate node, which is the core of the workflow and includes processing logic and conditional branches, and connects the intelligent tool set that is packaged and integrated according to the process sequence;

[0029] S3433. Create an end node, and the end node connection function is an intelligent tool set for outputting result reports.

[0030] Preferably, Agent is defined based on LLM function call or ReAct.

[0031] Compared with the prior art, the beneficial effects of the present invention are:

[0032] 1. The present invention realizes the textualization of business processes, business rules, and expert experience through planning, designing, and optimizing professional knowledge bases, realizes the sharing and dissemination of expert knowledge, and provides accurate and comprehensive knowledge support for autonomous query of AI agents.

[0033] 2. The present invention encapsulates existing tool capabilities into independent functional APIs, and develops personalized dedicated APIs based on business characteristics, providing a rich but independent API tool resource library. API tool sets can be freely combined according to different business needs, and API tools can be flexibly added, deleted, and replaced according to subsequent business adjustments, which has strong flexibility and reusability. Then, by integrating the tool set into the AI ​​agent, the traditional tools that can only be used by humans can be turned into intelligent tool sets for use by AI agents.

[0034] 3. The present invention organizes and arranges the business analysis and processing flow executed by AI intelligent bodies to realize machine work replacing manual work and automate the processing flow, thereby solving the pain points of deep reliance on expert analysis, long training cycle for professionals, and difficulty in realizing the effectiveness of tools, reducing the difficulty of anomaly detection and analysis processing, and improving the efficiency and accuracy of anomaly detection and analysis processing.

[0035] 4. The present invention summarizes the analysis results of each functional API, outputs a comprehensive analysis conclusion, and finally presents a comprehensive analysis conclusion with strong readability and comprehensive interpretation to the user. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flow chart of a method for detecting and handling business performance anomalies based on an AI agent according to an embodiment of the present invention;

[0037] Figure 2 is a flowchart of a business analysis process of integrating and arranging an intelligent tool set according to an embodiment of the present invention;

[0038] Figure 3 It is a flowchart of the AI ​​agent of an embodiment of the present invention automatically executing business analysis when processing complaint analysis business. DETAILED DESCRIPTION

[0039] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0040] Example 1

[0041] like Figure 1 The present invention provides a method for detecting and handling business performance anomalies based on an AI agent, comprising the following steps:

[0042] S1. Establish a professional knowledge base for business topics;

[0043] S2. Encapsulate and develop intelligent tool sets based on business function requirements;

[0044] S3. Use the big model platform to build an AI agent that integrates the professional knowledge base and intelligent tool set and orchestrates the business analysis and processing flow of the intelligent tool set.

[0045] S4. The AI ​​agent automatically executes the business analysis and processing flow, detects and analyzes anomalies in business performance, and outputs comprehensive analysis conclusions.

[0046] The big model platform specifically refers to an application development platform, which has the following main features: it can support the access of development models in multiple languages; it provides a visual interface and tools to simplify the complex code development process, such as completing workflow orchestration and defining prompt words through a visual interface, without the need to write complex code; it has powerful data management and processing capabilities, such as automatic text preprocessing and knowledge retrieval; it has rapid deployment and iteration, and can be quickly deployed to a production environment.

[0047] According to a specific implementation scheme of the present invention, the establishment of a professional knowledge base can be multi-dimensional, and may include a business process knowledge base, a business rule knowledge base, and an expert experience knowledge base. The corresponding professional knowledge base can be planned, designed, and optimized. The knowledge base is provided for autonomous query by AI agents. For example, a specific embodiment is a complaint analysis topic in the field of communications, and a corresponding professional knowledge base is planned, designed, and optimized. Three modules are set up: a business process knowledge base: for the subdivision of Internet complaint scenarios, the analysis process of scenarios where Internet access is not possible and Internet access is slow is sorted out to guide the analysis steps of complaint cases. The business rule knowledge base sorts out the existing error code delimitation rules and poor quality judgment rules. The expert experience knowledge base sorts out the scene rapid delimitation rules based on expert experience, and summarizes the conclusions.

[0048] According to a specific embodiment of the present invention, the intelligent toolset may be an API-style toolset.

[0049] According to a specific implementation scheme of the present invention, the encapsulation of the API tool set is specifically as follows: first, the capability list of the existing API tools is sorted out, and then the existing API tools are encapsulated into an API tool set with independent functions according to business requirements.

[0050] According to a specific embodiment of the present invention, the development of API-based toolsets is specifically to develop personalized dedicated functional APIs for different business logics in combination with business rules and expert experience knowledge, and finally to form a freely combinable intelligent toolset.

[0051] According to a specific embodiment of the present invention, the large model platform can adopt the DIFY large model platform.

[0052] According to a specific embodiment of the present invention, the process of constructing an AI agent in step S3 includes:

[0053] S31. Choose cloud deployment or local deployment of DIFY large model platform;

[0054] S32, select the model and create an Agent;

[0055] S33. Upload professional knowledge base and set permissions;

[0056] S34. Integrate the intelligent tool set and orchestrate the business analysis processing flow of the intelligent tool set.

[0057] Among them, in step S3, an AI agent is constructed using a large model platform, and the following process of integration, execution, result feedback, and debugging of the professional knowledge base on the DIFY large model platform is carried out to enable the AI ​​agent to integrate the professional knowledge base:

[0058] 1. In the DIDY platform, access the large model Qwen2-72B-Instruct (parameter quantity: 72B, supported interface: OPENAI-like interface, maximum number of supported context tokens: 32K).

[0059] 2. Configure a chat assistant or agent: Create a chat assistant or agent and configure the model it uses. Plug in a professional knowledge base to the chat assistant or agent so that the knowledge in the professional knowledge base can be called when performing tasks.

[0060] 3. Write prompt words: Prompt words are the key to guiding the big model to perform specific tasks. When writing prompt words, it is necessary to clearly indicate in which cases the big model should call the professional knowledge base and give the specific method of calling the professional knowledge base.

[0061] 4. Execution process: When a user makes a request through a chat assistant or agent, the big model will determine whether the professional knowledge base needs to be called based on the prompt word. If the professional knowledge base needs to be called, the big model will perform the corresponding search operation and retrieve the relevant process knowledge from the professional knowledge base. The retrieved process knowledge will be used to guide subsequent operations or answer user questions.

[0062] 5. Result feedback: After executing the process, the DIDY big model platform will return the results to the user. The results may include direct answers, relevant text snippets, or other forms of information.

[0063] 6. Optimization and debugging: recall test, recall test of the retrieval effect of the professional knowledge base to ensure the accuracy and relevance of the retrieval results. Debugging and adjustment are used to adjust the prompt words, indexing technology or professional knowledge base content according to the test results to optimize the process execution effect. Regularly monitor the use and effect evaluation of the professional knowledge base to ensure that the professional knowledge base can continue to provide strong support for process execution.

[0064] According to a specific embodiment of the present invention, the business analysis process of the orchestration intelligent tool set is as follows:

[0065] According to a specific embodiment of the present invention, the text format of the uploaded professional knowledge base includes TXT, HTML, Markdown, PDF, XLSX and other formats.

[0066] According to a specific embodiment of the present invention, Agent may be defined based on LLM function call or ReAct.

[0067] Example 2

[0068] Different from Example 1, this example further describes in detail the process of packaging and developing the intelligent tool set.

[0069] In response to the complaint analysis business needs in the communications field, based on the combination of business scenarios and analysis methods, from the perspective of manual practice complaint analysis, we select the basic and indispensable tools in the complaint analysis process, and design personalized special tools by sorting out the capability list. It is also possible to add special scenario analysis tools through scenario mining. The following table is an example of a tool set for a specific embodiment of the present invention, in which the centralized analysis API is a personalized special function API newly developed in combination with business rules and expert experience library knowledge, which can be used to replace manual analysis work; the user speed limit API is a scenario analysis tool that is iteratively newly developed through scenario mining. The intelligent toolset provides a rich tool resource library that can be freely combined for subsequent AI agents to call on demand. The specific contents are shown in the following table.

[0070]

[0071]

[0072] Example 3

[0073] Different from Example 1 and Example 2, this embodiment further describes in detail the construction of an AI agent, the integration of a professional knowledge base and an intelligent tool set, and the orchestration of the business analysis processing flow of the intelligent tool set.

[0074] Using the DIFY large model platform, combined with the agent API form tool set and professional knowledge base, build an AI agent. The specific implementation steps are as follows:

[0075] 1. Deploy the DIFY platform on dedicated container cloud resources.

[0076] 2. Select the model: On the DIFY platform, select a suitable LLM model as the basis of the AI ​​agent. The DIFY platform supports hundreds of proprietary / open source LLMs and dozens of inference providers and self-hosted solutions, such as GPT, Mistral, Llama3, etc.

[0077] 3. Create Agent: Define Agent based on LLM function call or ReAct, and add pre-built or custom tools to Agent. DIFY provides more than 50 built-in tools for AI Agent, such as Google Search, DELL·E, Stable Diffusion, and WolframAlpha.

[0078] 4. Prompt Arrangement: Use Prompt IDE to create prompts and set the behavior and interaction methods of the agent when answering questions. You can set the opening remarks, prompt words, response style, etc. to improve the sense of immersion and user experience of the conversation.

[0079] 5. Upload professional knowledge base: On the DIFY platform, upload the organized knowledge resources (support TXT, HTML, Markdown, PDF, XLSX and other formats) to the platform through the upload function. DIFY will segment and clean the documents to improve the accuracy of the query.

[0080] 6. Set the professional knowledge base permissions: Set the visibility permissions, index mode, embedding model, and retrieval settings of the professional knowledge base as needed.

[0081] 7. Integrate professional knowledge base: In the setting of the agent, integrate the previously built professional knowledge base. In this way, when the agent encounters relevant problems, it can automatically retrieve and return accurate answers from the professional knowledge base.

[0082] 8. The specific steps for integrating and orchestrating the business analysis process of the intelligent tool set include:

[0083] 1) Sort out specific business analysis processes based on business scenarios,

[0084] 2) Create workflows on the DIFY large model platform;

[0085] 3) For the business analysis process, create multiple nodes in the workflow and connect the intelligent tool sets in the nodes.

[0086] The specific implementation process can be found in Figure 2 , wherein the intelligent toolset adopts an API-style toolset, which is composed of one or more API tools. First, a workflow is created on the DIFY large model platform based on the business analysis processing flow, and multiple nodes are added in sequence to the workflow, including a start node, an intermediate node, and an end node. For the business analysis process, first create a start node, and connect the API toolset that reads the detailed data of the user query period; then create an intermediate node, which is the core of the workflow, including processing logic and conditional branches. It can encapsulate an API toolset that integrates independent functions according to the process sequence, and the intermediate node connects the API toolset; finally, create an end node, and connect the relevant API toolset that outputs the result report. After completing the business process orchestration, debug and publish the workflow, and optimize the workflow according to actual usage.

[0087] Example 4

[0088] Different from the previous embodiments 1 to 3, this embodiment further explains in detail how the AI ​​agent automatically executes the arranged business analysis processing flow and outputs the comprehensive analysis conclusions.

[0089] Specific implementation examples Figure 3 When processing complaint analysis services, after the query end enters the complaint number, query time period, and selects the business scenario, the AI ​​agent automatically executes the pre-arranged business process. First, call the user detailed bill query API to query the user's basic Internet information, and determine whether there is a 4G or 5G bill based on the information obtained. If so, call the corresponding function API in the arranged order. Calling each API will output the corresponding conclusion. Finally, call the report generation API to summarize the conclusions and then output the comprehensive analysis conclusion.

[0090] The rules summarized by experts can be used to classify and explain the causes of a single functional API output set one by one, and define the TOP N causes by setting delimited priority rules and weight rules. The AI ​​agent further explains the output conclusions of the functional API called in the business process according to the rules summarized by experts, and finally outputs a comprehensive analysis conclusion and processing suggestions, ultimately presenting users with a highly readable and comprehensive comprehensive analysis conclusion.

[0091] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for detecting and handling business performance anomalies based on AI agents, characterized in that: The steps include: S1. Establish a professional knowledge base for business topics; S2. Encapsulate and develop intelligent tool sets based on business function requirements; S3. Use the big model platform to build an AI agent, which integrates the professional knowledge base and the intelligent tool set, and choreographs the business analysis and processing flow of the intelligent tool set; S4. The AI ​​agent automatically executes the business analysis and processing flow, performs abnormal detection and analysis on business performance, and outputs comprehensive analysis conclusions.

2. According to claim 1, the method for detecting and handling abnormal business performance based on AI agent is characterized in that: The professional knowledge base includes a business process knowledge base, a business rule knowledge base, and an expert experience knowledge base.

3. According to claim 1, the method for detecting and handling abnormal business performance based on AI agent is characterized in that: The intelligent toolset is an API-style toolset.

4. According to claim 3, the method for detecting and handling abnormal business performance based on AI agent is characterized in that: The encapsulation of the API tool set is specifically as follows: firstly, the capability list of the existing API tools is sorted out, and then the existing API tools are encapsulated into an API tool set with independent functions according to business requirements.

5. According to claim 3, the method for detecting and handling abnormal business performance based on AI agent is characterized in that: The API tool set development is specifically to develop personalized special function APIs for different business logics in combination with business rules and expert experience knowledge.

6. According to claim 1, the method for detecting and handling abnormal business performance based on AI agent is characterized in that: The large model platform adopts the DIFY large model platform.

7. The method for detecting and handling abnormal business performance based on AI agent according to claim 6 is characterized in that: The process of building an AI agent in step S3 includes: S31, select cloud deployment or local deployment of the DIFY large model platform; S32, select the model and create an Agent; S33, uploading the professional knowledge base and setting permissions; S34. Integrate and orchestrate the business analysis processing flow of the intelligent tool set.

8. The method for detecting and handling abnormal business performance based on AI agent according to claim 7 is characterized in that: The steps of integrating and arranging the business analysis processing flow of the intelligent tool set in step S34 include: S341. Sort out the business analysis process based on business scenarios; S342, creating a workflow on the DIFY large model platform; S343. For the business analysis process, create multiple nodes in the workflow, and connect the intelligent tool set in the nodes.

9. The method for detecting and handling abnormal business performance based on AI agent according to claim 8 is characterized in that: In step S343, multiple nodes are created in the workflow, and the specific steps of connecting the intelligent tool set in the nodes are: S3431, creating a start node, wherein the start node connection function is an intelligent tool set for reading data; S3432, creating an intermediate node, which is the core of the workflow, including processing logic and conditional branches, and connecting the intelligent tool set packaged and integrated in the process sequence; S3433. Create an end node, wherein the end node connection function is an intelligent tool set for outputting a result report.

10. The method for detecting and handling abnormal business performance based on AI agent according to claim 7, characterized in that: Define Agent based on LLM function call or ReAct.

Images