Core network complaint processing agent construction method and system based on large language model

Abstract

The invention relates to the technical field of communication networks, in particular to a core network complaint processing agent construction method and system based on a large language model, and the method comprises the steps: carrying out the text cutting and semantic analysis operation of a complaint work order through a core network large model, and recognizing a key complaint scene from the complaint work order; the method has the beneficial effects that text segmentation is performed on complaint work order information through a large language model, a key scene is identified, a complaint analysis process is decomposed into a plurality of subtasks, and the subtasks are allocated to different agents (such as a signing data quick agent and a workflow agent) for cooperative processing. By constructing an orchestratable capability component platform (accumulating 40 + small model capability and 30 + component capability), automatic definition positioning of eight scenes such as white lists and area limitation is realized. Practice verifies that the white list type complaint hit rate reaches 80% or above, the complaint processing efficiency is improved by 500%, and the manual operation cost is remarkably reduced.

Description

Technical Field

[0001] The present invention relates to the field of communication network technology, and in particular to a method and system for constructing a core network complaint processing intelligent agent based on a large language model. Background Art

[0002] In the traditional core network complaint handling process, the following steps are mainly completed manually: (1) Complaint classification: Manually classify work orders into types, such as whitelist issues, contract data issues, etc.; (2) Data query: Manually query user contract data (such as regional restrictions, machine-card separation strategies), online status, DPI (deep packet inspection) data, etc. through multiple systems; (3) Problem location: Traverse possible fault points based on manual experience, such as checking whitelist configuration, routing anomalies, etc.; (4) Conclusion integration: Manually summarize information from each link to form a processing plan.

[0003] The defects of existing technologies are as follows: (1) Low efficiency: a single complaint processing requires manual execution of multiple rounds of queries, which takes up to several hours. In particular, the processing cycle is significantly extended in complex scenarios; (2) Dependence on experience: Problem location is highly dependent on expert experience, which is difficult for newcomers to get started and the processing quality is uneven; (3) Cumbersome process: Manual switching between multiple systems to query data, such as the EOMS system and the core network element management system, results in redundant operation steps; (4) Poor reusability: The analysis process needs to be repeatedly constructed for different complaint scenarios, and existing capabilities cannot be quickly reused. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for constructing a core network complaint processing intelligent agent based on a large language model to solve the problems raised in the above background technology.

[0005] To achieve the above objectives, the present invention provides the following technical solutions: a method for constructing a core network complaint handling agent based on a large language model, comprising:

[0006] Use the core network's large model to perform text segmentation and semantic analysis on complaint tickets to identify key complaint scenarios.

[0007] With the help of complaint location agents, the complex complaint analysis process is broken down into multiple subtasks, and these subtasks are assigned to corresponding quick agents or workflow agents; among them, quick agents at least cover user contract data quick agents and DPI query quick agents, which are used to execute atomic capability queries; workflow agents connect the quick agents in series according to the preset process to achieve scenario-based complaint delimitation.

[0008] Preferably, text segmentation and semantic parsing specifically include: extracting numbers, fault time, business codes, and URL entity information from work orders; determining complaint scenario types based on keyword matching and semantic analysis, and the complaint scenario types include whitelist scenarios and area restriction scenarios.

[0009] Preferably, the subtask decomposition is specifically as follows: for the whitelist scenario, it is decomposed into number PCC function analysis, service coding key rule analysis, and bureau data and equipment configuration consistency comparison subtasks; for each decomposed subtask, the corresponding quick intelligent agent is called to perform the operation, such as querying the area restrictions and machine-card separation strategy through the user contract data quick intelligent agent.

[0010] Preferably, it also includes a capability component sedimentation mechanism, which is as follows: accumulating more than 40 small model capabilities and more than 30 component capabilities through the low-code platform. The small model capabilities include number PCC function activation status analysis and whitelist three-layer IP binding relationship check; solidifying the analysis process of the eight complaint scenarios into a reusable workflow template.

[0011] Preferably, it also includes active and passive analysis modes, specifically: in passive mode, it automatically reads the work order information and performs delimitation operations according to the preset process; in active mode, it supports customized selection of components through the intelligent agent analysis window, and saves commonly used analysis processes; at the same time, in the conclusion integration process, it summarizes the output results of each intelligent agent, and generates fault delimitation conclusions through the rule engine. The output conclusions include "the whitelist seven-layer URL is inconsistent with the bureau data configuration" and "the user's contracted area restriction policy causes access failure."

[0012] A system for constructing a core network complaint handling intelligent agent based on a large language model, comprising:

[0013] The text semantic analysis module is used to perform text segmentation and semantic analysis on complaint tickets using the core network large model to identify key complaint scenarios.

[0014] The task decomposition and assignment module is used to use the complaint location agent to decompose the complex complaint analysis process into multiple subtasks and assign them to the corresponding shortcut agents or workflow agents;

[0015] A quick agent set, including at least a user contract data quick agent and a DPI query quick agent, used to execute atomic capability queries;

[0016] The workflow agent module connects various quick agents in series based on preset processes to achieve scenario-based complaint demarcation.

[0017] Preferably, the text semantic analysis module comprises: an entity information extraction unit configured to extract number, fault time, service code, URL entity information in the work order; and a scene type determination unit configured to determine the complaint scene type based on keyword matching and semantic analysis, such as a whitelist scene and a regional restriction scene.

[0018] Preferably, the task decomposition and distribution module comprises: a subtask decomposition unit configured to decompose the complaint analysis process into a number PCC function analysis, a service code key rule analysis, and a local data and device configuration consistency comparison subtask for the whitelist scene; and an agent calling unit configured to call a corresponding shortcut agent for each subtask, such as querying a regional restriction and a card separation strategy through a user subscription data shortcut agent.

[0019] Preferably, the method further comprises a capability component sedimentation module, which comprises: a capability accumulation unit configured to accumulate 40+ small model capabilities and 30+ component capabilities through a low-code platform, wherein the small model capabilities comprise a number PCC function opening state analysis and a whitelist three-layer IP binding relationship check; and a template solidification unit configured to solidify the analysis processes of the eight complaint scenes into reusable workflow templates.

[0020] Preferably, the method further comprises an analysis mode selection module and a conclusion integration module; the analysis mode selection module comprises: a passive analysis unit configured to automatically read work order information and perform delimitation according to a preset process in a passive mode; and an active analysis unit configured to support self-defined selection of components through an agent analysis window and save commonly used analysis processes in an active mode; and the conclusion integration module is configured to integrate the output results of the agents, generate a fault delimitation conclusion through a rule engine, and output the conclusion, which comprises a whitelist seven-layer URL and local data configuration inconsistency and a user subscription regional restriction strategy leading to access failure.

[0021] Compared with the prior art, the method has the following beneficial effects:

[0022] The core network complaint processing agent construction method and system based on a large language model can cut text of a complaint work order information through a large language model, recognize a key scene, decompose a complaint analysis process into multiple subtasks, and assign the subtasks to different agents (such as a subscription data shortcut agent and a workflow agent) for collaborative processing. By constructing a programmable capability component platform (accumulating 40+ small model capabilities and 30+ component capabilities), automatic delimitation and positioning of eight scenes such as a whitelist and a regional restriction can be achieved. Practice has proved that the method can make the hit rate of whitelist complaints reach more than 80%, improve the complaint processing efficiency by 500%, and significantly reduce the labor cost. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The method flowchart of the application. DETAILED DESCRIPTION

[0024] In order to clearly and completely describe the objectives and technical solutions of the present invention and make the advantages more clearly understood, the embodiments of the present invention are further described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are part of the embodiments of the present invention, not all of them, and are only used to explain the embodiments of the present invention, not to limit the embodiments of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0025] In a first embodiment, the present invention provides a technical solution: a method for constructing a core network complaint handling agent based on a large language model, comprising the following steps:

[0026] (1) Intelligent agent architecture design

[0027] Core network large model: Built based on deep learning, it has text understanding and policy analysis capabilities and is used to interpret the semantics of complaint tickets.

[0028] Complaint Location Agent: After receiving a work order, it extracts key information (such as "IoT card cannot access the Internet" and "whitelist") through text segmentation, identifies the complaint scenario, and decomposes it into subtasks (such as "query whitelist configuration" and "analyze machine-card separation strategy").

[0029] Quick Agent Cluster: This includes user contract data quick agents and DPI query quick agents, encapsulating atomic capabilities (such as "querying regional restriction status" and "obtaining user access IP / URL") and supporting fast API calls.

[0030] Workflow agent: preset processing procedures for specific scenarios (such as whitelist complaints), connecting various quick agents in series to achieve automated task orchestration.

[0031] (2) Complaint handling process

[0032] Work order parsing: The large model performs word segmentation and entity recognition (such as numbers, fault times, and business codes) on the work order text to generate structured data;

[0033] Scenario Identification: Based on keyword matching (such as "whitelist" and "inaccessible") and semantic analysis, the complaint scenario (such as whitelist scenario and terminal disconnection scenario) is determined.

[0034] Task decomposition: Break down the process into subtasks based on the scenario. For example, a whitelist scenario can be broken down into "querying PCC function status" and "comparing bureau data with device configuration."

[0035] Agent scheduling: The complaint location agent is assigned to the corresponding quick agent according to the subtask type. For example, the user contract data query task calls the contract data quick agent and queries the PGW / UPF / PCF network element policy through the API;

[0036] Process orchestration: The workflow agent connects subtasks in series according to preset logic. For example, in a whitelist scenario, the "business coding key rule analysis" is first called, and then the "seven-layer URL consistency check" is performed.

[0037] Conclusion integration: Summarize the output results of each intelligent agent and generate fault demarcation conclusions (such as "whitelisted third-layer IP is not bound" and "DNSSniffer function is not enabled") through the rule engine.

[0038] (3) Capability component precipitation

[0039] Low-code accumulation: The low-code platform encapsulates 40+ small model capabilities (such as "number PCC function analysis" and "IP analysis") and 30+ component capabilities (such as "failure code analysis" and "routing anomaly detection");

[0040] Scenario-based reuse: The analysis processes of eight scenarios (whitelist, regional restrictions, slice configuration, etc.) are solidified into templates to support rapid matching and calling of new work orders.

[0041] (4) Active and passive analysis

[0042] Passive analysis: Automatically reads key information from work orders and performs delimitation according to preset processes. For example, if "device-card separation binding" is detected, it automatically triggers a suggestion to unbind.

[0043] Active analysis: Provides an intelligent analysis window, supports manual selection of components (such as "5G core network performance indicator query" and "user traffic over-set analysis"), customizes the analysis process and saves it as a commonly used template.

[0044] Example 2, based on Example 1, proposes a system for constructing a core network complaint handling agent based on a large language model, including:

[0045] The text semantic parsing module is used to perform text segmentation and semantic parsing on complaint tickets through the core network large model to identify key complaint scenarios. It includes: an entity information extraction unit, which is used to extract the number, fault time, service code, and URL entity information in the ticket; a scenario type determination unit, which determines the complaint scenario type based on keyword matching and semantic analysis, such as whitelist scenarios and area restriction scenarios.

[0046] The task decomposition and allocation module is used to use the complaint location agent to decompose the complex complaint analysis process into multiple sub-tasks, and assign them to the corresponding quick agents or workflow agents; it includes: a sub-task decomposition unit, which decomposes the complaint analysis process into number PCC function analysis, business coding key rule analysis, and bureau data and equipment configuration consistency comparison sub-tasks for whitelist scenarios; an agent calling unit, which calls the corresponding quick agent for each sub-task, such as querying area restrictions and machine-card separation strategies through the user contract data quick agent.

[0047] A quick agent set, including at least a user contract data quick agent and a DPI query quick agent, used to execute atomic capability queries;

[0048] The workflow agent module connects various quick agents in series based on preset processes to achieve scenario-based complaint demarcation.

[0049] It also includes a capability component sedimentation module, which includes: a capability accumulation unit, which accumulates 40+ small model capabilities and 30+ component capabilities through a low-code platform. The small model capabilities include number PCC function activation status analysis and whitelist three-layer IP binding relationship check; a template solidification unit, which solidifies the analysis process of the eight complaint scenarios into a reusable workflow template.

[0050] It also includes an analysis mode selection module and a conclusion integration module; the analysis mode selection module includes: a passive analysis unit, which automatically reads the work order information in passive mode and performs delimitation according to the preset process; an active analysis unit, which supports customized selection of components through the intelligent agent analysis window in active mode and saves commonly used analysis processes; the conclusion integration module is used to summarize the output results of each intelligent agent and generate fault delimitation conclusions through the rule engine. The output conclusions include "the whitelist seven-layer URL is inconsistent with the bureau data configuration" and "the user's contracted area restriction policy causes access failure."

[0051] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a core network complaint handling agent based on a large language model, characterized by: include: Use the core network's large model to perform text segmentation and semantic analysis on complaint tickets to identify key complaint scenarios. With the help of complaint location agents, the complex complaint analysis process is broken down into multiple subtasks, and these subtasks are assigned to corresponding quick agents or workflow agents; among them, quick agents at least cover user contract data quick agents and DPI query quick agents, which are used to execute atomic capability queries; workflow agents connect the quick agents in series according to the preset process to achieve scenario-based complaint delimitation.

2. The method for constructing a core network complaint handling agent based on a large language model according to claim 1, characterized in that: Text segmentation and semantic analysis specifically include: extracting numbers, fault time, business codes, and URL entity information from work orders; determining complaint scenario types based on keyword matching and semantic analysis, which include whitelist scenarios and regional restriction scenarios.

3. The method for constructing a core network complaint handling agent based on a large language model according to claim 2, characterized in that: The subtask decomposition is specific as follows: for the whitelist scenario, it is decomposed into number PCC function analysis, service coding key rule analysis, and bureau data and equipment configuration consistency comparison subtasks; for each decomposed subtask, the corresponding quick intelligent agent is called to perform the operation, such as querying the area restrictions and machine-card separation strategy through the user contract data quick intelligent agent.

4. The method for constructing a core network complaint handling agent based on a large language model according to claim 3, characterized in that: It also includes a capability component sedimentation mechanism, as follows: accumulating more than 40 small model capabilities and more than 30 component capabilities through the low-code platform. The small model capabilities include number PCC function activation status analysis and whitelist three-layer IP binding relationship check; solidifying the analysis process of the eight complaint scenarios into a reusable workflow template.

5. The method for constructing a core network complaint handling agent based on a large language model according to claim 4, characterized in that: It also includes active and passive analysis modes, specifically: in passive mode, it automatically reads the work order information and performs delimitation operations according to the preset process; in active mode, it supports customized selection of components through the intelligent agent analysis window and saves commonly used analysis processes; at the same time, in the conclusion integration process, it summarizes the output results of each intelligent agent and generates fault delimitation conclusions through the rule engine. The output conclusions include "the whitelist seven-layer URL is inconsistent with the bureau data configuration" and "the user's contracted area restriction policy causes access failure." 6. A system for constructing a core network complaint handling agent based on a large language model as claimed in claim 1, characterized in that: include: The text semantic analysis module is used to perform text segmentation and semantic analysis on complaint tickets using the core network large model to identify key complaint scenarios. The task decomposition and assignment module is used to use the complaint location agent to decompose the complex complaint analysis process into multiple subtasks and assign them to the corresponding shortcut agents or workflow agents; A quick agent set, including at least a user contract data quick agent and a DPI query quick agent, used to execute atomic capability queries; The workflow agent module connects various quick agents in series based on preset processes to achieve scenario-based complaint demarcation.

7. A system according to claim 6, characterized in that: The text semantic parsing module includes: an entity information extraction unit, which is used to extract the number, fault time, business code, and URL entity information in the work order; a scenario type determination unit, which determines the complaint scenario type based on keyword matching and semantic analysis, such as whitelist scenario and area restriction scenario.

8. A system according to claim 7, characterized in that: The task decomposition and allocation module includes: a subtask decomposition unit, which decomposes the complaint analysis process into number PCC function analysis, business coding key rule analysis, and bureau data and equipment configuration consistency comparison subtasks for the whitelist scenario; an intelligent agent calling unit, which calls the corresponding quick intelligent agent for each subtask, such as querying regional restrictions and machine-card separation strategies through the user contract data quick intelligent agent.

9. A system according to claim 8, characterized in that: It also includes a capability component sedimentation module, which includes: a capability accumulation unit, which accumulates 40+ small model capabilities and 30+ component capabilities through a low-code platform. The small model capabilities include number PCC function activation status analysis and whitelist three-layer IP binding relationship check; a template solidification unit, which solidifies the analysis process of the eight complaint scenarios into a reusable workflow template.

10. A system according to claim 9, characterized in that: It also includes an analysis mode selection module and a conclusion integration module; The analysis mode selection module includes: a passive analysis unit, which automatically reads work order information and performs delimitation according to the preset process in passive mode; an active analysis unit, which supports customized selection of components through the intelligent agent analysis window in active mode and saves commonly used analysis processes; a conclusion integration module is used to summarize the output results of each intelligent agent and generate fault delimitation conclusions through the rule engine. The output conclusions include "the whitelist seven-layer URL is inconsistent with the bureau data configuration" and "the user's contracted area restriction policy leads to access failure."

Images