A method, apparatus, electronic device, and storage medium for root cause analysis of user complaints.

Abstract

This disclosure provides a method, apparatus, electronic device, and storage medium for root cause analysis of user complaints. The method includes: creating an MCP session based on a user complaint work order and initializing the MCP context corresponding to the MCP session; based on the initialized MCP context, calling an NLP-to-SQL agent through the MCP protocol to convert the natural language description of the user complaint into a structured query statement for an O-domain database, executing the structured query statement, and writing the query results into the MCP context in a standard data format; based on the query results in the MCP context, calling a multi-dimensional network analysis agent in parallel to perform specialized diagnosis and writing the specialized diagnosis results into the MCP context; and summarizing the specialized diagnosis results in the MCP context through a root cause reasoning agent to perform comprehensive judgment and root cause localization. Embodiments of this disclosure can significantly improve the objectivity and accuracy of root cause analysis results and greatly enhance the overall collaborative capability of the system.

Description

Technical Field

[0001] This disclosure relates to the field of network operation and maintenance technology, and in particular to a method, apparatus, electronic device and storage medium for root cause analysis of user complaints. Background Technology

[0002] With the large-scale commercialization of next-generation mobile communication technologies such as 5G / 6G, mobile networks have become a core infrastructure supporting the digital transformation of society. Users' reliance on network services has deepened unprecedentedly, and their usage scenarios have comprehensively penetrated high-value business areas such as high-definition video, remote work, online finance, and the industrial internet. Against this backdrop, the Quality of Experience (QoE) is directly related to user satisfaction and the core competitiveness of operators. User complaints, as the most direct feedback channel for service quality, have become a key indicator for measuring the level of intelligent operation and maintenance of operators' networks, given their efficiency and accuracy in handling them.

[0003] Currently, mainstream root cause analysis methods for user complaints heavily rely on manual experience. When a user files a complaint, the ticket flows through the customer service system to relevant technical analysts. Analysts need to communicate with the user to obtain the time, location, and description of the problem. They then manually query various network data in the O-domain data warehouse, ultimately relying on personal experience and static threshold rules to perform correlation analysis on massive amounts of heterogeneous O-domain data to infer the potential root cause of the network problem. This analysis model has the following significant drawbacks:

[0004] (1) Low processing efficiency: There are many manual intervention steps in the analysis process. Manual querying, sorting and correlation analysis of O domain data are required. The overall operation takes a long time and it is difficult to meet the timeliness requirements of users for rapid response to complaints.

[0005] (2) The analysis results are highly subjective: the root cause localization results are highly dependent on the individual experience and professional skills of the analysts. Different people may have different analysis and judgment on the same complaint issue, resulting in a lack of unified standards for root cause conclusions and a tendency to misjudge or omit.

[0006] (3) Poor system coordination: Root cause analysis subtasks such as network performance analysis, coverage assessment, and load diagnosis are completed by different tools or modules. There is a lack of unified context management mechanism and standardized communication protocol between the tools and modules, making it impossible to achieve efficient sharing of data and analysis results and difficult to complete multi-dimensional collaborative analysis.

[0007] (4) Insufficient intelligence and scalability: The existing analysis architecture is difficult to integrate artificial intelligence technologies such as natural language processing (NLP) and machine learning. The automation level of the analysis process is low, and system upgrades and the integration of new analysis functions require a lot of customized development, which is costly and cannot adapt to the rapid changes in network analysis needs. Summary of the Invention

[0008] This disclosure provides a method, apparatus, electronic device, and storage medium for root cause analysis of user complaints, which addresses the problems of low processing efficiency, strong subjectivity of analysis results, poor system coordination, and insufficient intelligence and scalability in existing root cause analysis methods for user complaints.

[0009] Firstly, this disclosure provides a method for root cause analysis of user complaints, the method comprising:

[0010] Create a Model Context Protocol (MCP) session based on the user complaint work order, and initialize the MCP context corresponding to the MCP session;

[0011] Based on the initialized MCP context, the Natural Language Processing to Structured Query Language (NLP-to-SQL) intelligent agent is called through the MCP protocol to convert the natural language description of the user's complaint into a structured query statement for the O domain database, execute the structured query statement, and write the query result into the MCP context in a standard data format.

[0012] Based on the query results in the MCP context, the multi-dimensional network analysis agent is invoked in parallel to perform specialized diagnosis, and the specialized diagnosis results are written into the MCP context.

[0013] The root cause reasoning agent summarizes the specific diagnostic results in the MCP context, performs comprehensive judgment and root cause localization, and writes the root cause localization conclusion into the MCP context.

[0014] A structured analysis report is generated based on the root cause localization conclusions, and the MCP session is terminated.

[0015] Furthermore, before creating a Model Context Protocol (MCP) session based on a user complaint ticket, the method further includes:

[0016] We use Kafka data collection tools to receive user complaint tickets from customer service systems, user self-service terminals, or third-party channels.

[0017] Furthermore, the initialized MCP context includes the user's original complaint text and complaint channel metadata. Based on the initialized MCP context, the Natural Language Processing to Structured Query Language (NLP-to-SQL) intelligent agent is invoked through the MCP protocol to convert the natural language description of the user complaint into a structured query statement for the O-domain database. The structured query statement is executed, and the query results are written to the MCP context in a standard data format. Specifically, this includes:

[0018] Using the MCP protocol, the original user complaint text and complaint channel metadata in the MCP context are input into the NLP-to-SQL agent. The NLP-to-SQL agent is used to parse the semantics of the user complaint to obtain key field information, and a structured query statement for the O domain database is generated based on the key field information. After executing the structured query statement, the query result is written into the MCP context in a standard data format.

[0019] Furthermore, the step of using the NLP-to-SQL agent to parse the semantics of user complaints to obtain key field information, and generating a structured query statement for the O-domain database based on the key field information, specifically includes:

[0020] The NLP-to-SQL agent is used to perform the following sub-tasks to obtain key field information including user identifier, time window, geographic location range, business type and symptom label: extracting user identifier based on complaint channel metadata, extracting the time window and geographic location range of user complaints from the original user complaint text, classifying the original user complaint text to identify the corresponding business type, and extracting fault symptom labels from the original user complaint text.

[0021] The NLP-to-SQL agent is used to encapsulate the user identifier, time window, geographic location range, business type, and symptom tags in a structured manner, and then writes them into the MCP context.

[0022] Based on the user identifier, time window, geographic location range, business type, and symptom tags encapsulated in the MCP context, a structured query statement for the O-domain database is generated.

[0023] Furthermore, the step of invoking a multi-dimensional network analysis agent in parallel to perform specialized diagnosis based on the query results in the MCP context, and writing the specialized diagnosis results into the MCP context, specifically includes:

[0024] Based on the query results in the MCP context, at least two network analysis agents from the following dimensions are invoked in parallel to perform corresponding specialized diagnoses, and the specialized diagnoses are written into the MCP context:

[0025] The weak coverage analysis agent is invoked to calculate the statistical values ​​of the reference signal received power (RSRP) and the reference signal received quality (RSRQ) at the user's spatiotemporal point, and to determine whether the statistical values ​​meet the weak coverage determination threshold.

[0026] Call the high-load analysis agent to determine whether the user's cell is under high load.

[0027] The uplink interference analysis agent is invoked to query the average uplink interference level on the PRB bandwidth of each physical resource block in the user's cell, and to determine whether the average uplink interference level exceeds a preset threshold.

[0028] Furthermore, the step of summarizing the specific diagnostic results of each item in the MCP context through the root cause reasoning agent and performing comprehensive judgment and root cause localization specifically includes:

[0029] Multiple specialized analysis agents are invoked in parallel via the MCP protocol to perform specialized task analysis, and the results of the specialized task analysis are written into the MCP context. Each specialized analysis agent performs specialized task analysis based on the data in the MCP context.

[0030] The root cause reasoning agent reads the specific diagnostic results and the specific task analysis results in the MCP context, combines them with B-domain order data to perform multi-dimensional evidence fusion analysis, performs comprehensive judgment, and obtains the root cause judgment result and the corresponding confidence level.

[0031] If the confidence level is greater than or equal to a preset threshold, the root cause determination result is determined as the final root cause localization conclusion; otherwise, the root cause determination result and related analysis data are pushed to a human agent for review.

[0032] Furthermore, the specialized analytical agent includes:

[0033] The current complaint semantic analysis agent is used to analyze user sentiment and determine the suddenness of complaint issues.

[0034] A historical duplicate complaint analysis agent is used to determine whether a user has filed duplicate complaints and whether the locations of the duplicate complaints are the same; and,

[0035] A network problem root cause analysis agent is used to determine the fault status of the base station associated with the location grid of the complaint and generate handling strategy suggestions.

[0036] Secondly, this disclosure provides a user complaint root cause analysis device, the device comprising:

[0037] The session creation and initialization module is used to create a Model Context Protocol (MCP) session based on a user complaint work order and initialize the MCP context corresponding to the MCP session.

[0038] The semantic structured query module, connected to the session creation and initialization module, is used to call the Natural Language Processing to Structured Query Language (NLP-to-SQL) intelligent agent through the MCP protocol based on the initialized MCP context. This agent converts the natural language description of the user's complaint into a structured query statement for the O domain database, executes the structured query statement, and writes the query results into the MCP context in a standard data format.

[0039] The agent-specific diagnostic module is connected to the semantic structured query module. It is used to call the multi-dimensional network analysis agent in parallel to perform specific diagnostics based on the query results in the MCP context, and write the specific diagnostic results into the MCP context.

[0040] The comprehensive judgment and root cause localization module is connected to the agent-specific diagnosis module. It is used to summarize the various specific diagnosis results in the MCP context through the root cause reasoning agent, perform comprehensive judgment and root cause localization, and write the root cause localization conclusion into the MCP context.

[0041] The report generation and session termination module is connected to the comprehensive analysis and root cause localization module, and is used to generate a structured analysis report based on the root cause localization conclusion and terminate the MCP session.

[0042] Thirdly, this disclosure provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores one or more computer programs executable by the at least one processor, the one or more computer programs being executed by the at least one processor to enable the at least one processor to perform the user complaint root cause analysis method described in the first aspect above.

[0043] Fourthly, this disclosure provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the user complaint root cause analysis method described in the first aspect above.

[0044] The user complaint root cause analysis method, apparatus, electronic device, and storage medium provided in this disclosure, by creating an MCP session and initializing a unified MCP context, and relying on the MCP protocol to achieve automated querying and standardized data writing of the O-domain database by an NLP-to-SQL agent, combines parallel specialized diagnosis by a multi-dimensional network analysis agent with comprehensive judgment by a root cause reasoning agent. The entire process of complaint root cause analysis is automated through standardized protocol communication and global context management, reducing reliance on manual intervention and improving the efficiency of complaint root cause analysis processing. Simultaneously, it achieves full automation of the complaint root cause analysis process through multi-agent collaborative fusion of multi-dimensional evidence. Root cause analysis eliminates over-reliance on human experience, providing standardized judgment criteria and significantly improving the objectivity and accuracy of results. Furthermore, the unified communication standard and context management mechanism based on the MCP protocol breaks down information barriers between analysis modules, enabling efficient data sharing and deep collaborative analysis among agents, greatly enhancing the overall system's collaborative capabilities. In addition, the open intelligent analysis architecture built on the MCP protocol allows for easy integration of various new analytical agents without requiring large-scale customization of existing systems, effectively improving the system's intelligence and scalability, and adapting to rapidly changing network analysis needs. This addresses the problems of low processing efficiency, strong subjectivity in analysis results, poor system collaboration, and insufficient intelligence and scalability in existing user complaint root cause analysis methods. Attached Figure Description

[0045] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the embodiments of the present disclosure to explain the disclosure and do not constitute a limitation thereof. In the drawings:

[0046] Figure 1 A flowchart of a user complaint root cause analysis method provided in this disclosure embodiment;

[0047] Figure 2 A flowchart of yet another user complaint root cause analysis method provided in this disclosure embodiment;

[0048] Figure 3 A block diagram of a user complaint root cause analysis device provided in an embodiment of this disclosure;

[0049] Figure 4 This is a block diagram of an electronic device provided in an embodiment of the present disclosure. Detailed Implementation

[0050] To enable those skilled in the art to better understand the technical solutions of this disclosure, exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments of this disclosure to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0051] Where there is no conflict, the various embodiments of this disclosure and the features thereof in the embodiments may be combined with each other.

[0052] As used herein, the term “and / or” includes any and all combinations of one or more related enumerated entries.

[0053] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. As used herein, the singular forms “a” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that when the terms “comprising” and / or “made of” are used in this specification, the presence of the stated feature, integral, step, operation, element, and / or component is specified, but the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof is not excluded. Words such as “connected” or “linked” are not limited to physical or mechanical connections but can include electrical connections, whether direct or indirect.

[0054] Unless otherwise specified, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and this disclosure, and will not be interpreted as having an idealized or overly formal meaning, unless expressly so defined herein.

[0055] To enable those skilled in the art to better understand the technical solutions in the embodiments of this disclosure, some technical terms involved in the embodiments of this disclosure will be briefly explained below.

[0056] (1) O domain database: including 4 / 5G mobile network signaling records, PM (Performance Management) data, and MR (Measurement Report) data.

[0057] A. 4 / 5G Mobile Network Signaling Records: These records the signaling interactions between the mobile user and the mobile network each time, including user attach, detach, service request, service detach request, bearer resource application, bearer resource release, terminal capability reporting, tracking area updates, etc. Fields involved include Msisdn (Mobile Subscriber International ISDN), start time, end time, CITY_ID, NodeB, CELL_ID, longitude, latitude, network standard, IMEI (International Mobile Equipment Identity), Sdrtype (signaling type), DCNR (indicating whether the UE supports NR dual connectivity and 5G core network), S1APID, SVRNAME (service name), uplink traffic, downlink traffic, uplink latency, downlink latency, etc.

[0058] B. PM Data: Primarily used for monitoring and optimizing network performance. PM data is used to assess the utilization of existing network resources and to make reasonable resource allocation and planning accordingly. Fields involved include: network element number, ECI, day partition, hour partition, cell name, vendor ID, vendor name, province ID, province name, city ID, city name, district / county ID, district / county name, affiliated base station, start time, granularity, insertion time, number of successful handovers within eNodeB, number of successful handovers via X2 interface, number of successful handovers via S1 interface, number of handover requests within eNodeB, number of handover requests via X2 interface, number of handover requests via S1 interface, number of successful handovers within gNodeB, number of successful handovers via Xn interface, number of successful handovers via Ng interface, number of handover requests within g node, number of handover requests via Xn interface, number of handover requests via Ng interface, number of successful NR (New Radio) to LTE (Long Term Evolution) handovers, number of NR to LTE handover requests, number of RRC (Radio Resource Control) connection re-establishment requests, and number of UE (User Equipment) handovers. The data includes: the number of successful RRC connection establishments initiated by user equipment (UE), the number of successful RRC connection reconstructions, the number of successful RRC connection establishments initiated by the network, the number of successful RRC connection establishments received, the number of RRC reconstruction requests received, the total number of downlink PRBs (Physical Resource Blocks) occupied in 4G cells, the total number of available downlink PRBs in 4G cells, the total number of downlink PRBs occupied in 5G cells, the total number of available downlink PRBs in 5G cells, the RRC connection establishment success rate, the E-RAB establishment success rate (QCI1), the E-RAB establishment success rate (QCI5), and the VoNR (Voice Over New Radio) voice QoSFlow drop rate.

[0059] C. MR Data: Records network-level signal reception strength, uplink interference, etc., which can be used to trace the coverage performance of users in the network. The fields involved include network element number, ECI, day zone, hour zone, cell name, vendor ID, vendor name, province ID, province name, city ID, city name, district / county ID, district / county name, base station, start time, granularity, insertion time, 4G overlap coverage rate, RSRP (Reference Signal Receiving Power) ≥ -112 sampling points proportion, total number of 4G MR sampling points, -105 sampling points proportion / total number of MR sampling points, total number of 5G MR sampling points, average uplink interference per PRB bandwidth, 4G modulo 3 interference sampling point proportion, 5G modulo 30 interference sampling point proportion, and number of 5G SINR (Signal to Interference plus Noise Ratio) >= 0dB MR sampling points.

[0060] (2) B domain order data: records order data related to user business, including payment, change of package, new package, account cancellation, etc. The fields involved include Msisdn, business type, execution time, completion time, user name, package information, etc.

[0061] (3) Kafka data acquisition tool: a high-throughput distributed publish-subscribe messaging system that is fast, scalable, and persistent. Its greatest feature is that it can process large amounts of data in real time to meet various needs, such as Hadoop-based batch processing systems, low-latency real-time systems, and Spark streaming engines.

[0062] Figure 1 A flowchart illustrating a user complaint root cause analysis method provided in this embodiment of the disclosure. (Refer to...) Figure 1 The method includes:

[0063] Step S101: Create an MCP (Model Context Protocol) session based on the user complaint work order, and initialize the MCP context corresponding to the MCP session.

[0064] Specifically, a user complaint ticket must include at least the user identifier (e.g., Msisdn), the complaint timestamp, the geographical location of the complaint (latitude and longitude or cell ID), and the user's description of the fault (e.g., "video buffering" or "unable to access"). The system creates a unique MCP session for each complaint ticket and initializes an MCP context. This context serves as the global data space for this analysis, used to store and transmit all shared information during the analysis process.

[0065] In some embodiments, prior to creating a Model Context Protocol (MCP) session based on a user complaint ticket, the method further includes:

[0066] We use Kafka data collection tools to receive user complaint tickets from customer service systems, user self-service terminals, or third-party channels.

[0067] Specifically, the system uses the Kafka data collection tool to receive user complaint tickets from the customer service system, user self-service terminals, or third-party channels in real time, and each ticket event triggers a root cause analysis process.

[0068] Step S102: Based on the initialized MCP context, call the NLP-to-SQL (Natural Language Processing to Structured Query Language) agent through the MCP protocol to convert the natural language description of the user complaint into a structured query statement for the O domain database, execute the structured query statement and write the query result into the MCP context in a standard data format.

[0069] Specifically, the system sends a call request to the NLP-to-SQL agent via the MCP protocol, using the context of the current MCP session as input. This agent utilizes its built-in natural language understanding model to perform natural language understanding and dynamically generates structured SQL query statements for the O-domain database. After executing the query, the NLP-to-SQL agent performs preliminary cleaning and structuring of the acquired raw network data (i.e., the query results), and writes the results into the MCP context in a standard data format.

[0070] In some embodiments, the initialized MCP context includes the user's original complaint text and complaint channel metadata. Based on the initialized MCP context, the Natural Language Processing to Structured Query Language (NLP-to-SQL) intelligent agent is invoked through the MCP protocol to convert the natural language description of the user's complaint into a structured query statement for the O-domain database. The structured query statement is executed, and the query results are written to the MCP context in a standard data format. Specifically, this includes:

[0071] Using the MCP protocol, the original user complaint text and complaint channel metadata in the MCP context are input into the NLP-to-SQL agent. The NLP-to-SQL agent is used to parse the semantics of the user complaint to obtain key field information, and a structured query statement for the O domain database is generated based on the key field information. After executing the structured query statement, the query result is written into the MCP context in a standard data format.

[0072] Specifically, the user's original complaint text (natural language), such as "Around 8 pm last night, I was experiencing severe lag while watching Douyin (TikTok) near Central Street in Harbin, it kept spinning in circles," includes at least the user's complaint ticket ID in the complaint channel metadata. The NLP-to-SQL intelligent agent obtains relevant key field information through entity recognition, intent classification, and symptom keyword extraction. Based on this key field information, it generates a structured query statement for the O-domain database. After executing the structured query statement, the query results are written to the MCP context in a standard data format.

[0073] In some embodiments, the step of using the NLP-to-SQL agent to parse the semantics of user complaints to obtain key field information, and generating a structured query statement for the O-domain database based on the key field information, specifically includes:

[0074] The NLP-to-SQL agent is used to perform the following sub-tasks to obtain key field information including user identifier, time window, geographic location range, business type and symptom label: extracting user identifier based on complaint channel metadata, extracting the time window and geographic location range of user complaints from the original user complaint text, classifying the original user complaint text to identify the corresponding business type, and extracting fault symptom labels from the original user complaint text.

[0075] The NLP-to-SQL agent is used to encapsulate the user identifier, time window, geographic location range, business type, and symptom tags in a structured manner, and then writes them into the MCP context.

[0076] Based on the user identifier, time window, geographic location range, business type, and symptom tags encapsulated in the MCP context, a structured query statement for the O-domain database is generated.

[0077] Specifically, the NLP-to-SQL agent obtains the structured, encapsulated key field information by performing the following subtasks:

[0078] Entity recognition: Extract MSISDN (associated through work orders), time phrase ("around 8 pm yesterday" → convert to UTC time window [2025-11-23T19:30:00, 2025-11-23T20:30:00]), and geographic location ("Harbin Central Avenue" → call GIS service to map to latitude and longitude range or candidate cell list);

[0079] Intent Classification: Identify business type ("Scrolling through Douyin" → classified as "high-throughput short video business");

[0080] Symptom keyword extraction: "stuck" and "looping" → mapped to abnormal QoE indicators (such as low downlink speed, high latency, and high retransmission rate).

[0081] Structured encapsulation: The above results are organized into a standard JSON object with the following fields: user identifier, time window, geographic location range, business type, and symptom tag.

[0082] {

[0083] "msisdn": "138XXXX1234",

[0084] "time_window": {"start": "2025-11-23T19:30:00Z", "end": "2025-11-23T20:30:00Z"},

[0085] "geo_hint": {"lat_min": 45.75, "lat_max": 45.77, "lon_min": 126.62, "lon_max": 126.64},

[0086] "service_type": "video_streaming",

[0087] "symptom_tags": ["low_throughput", "high_latency"]

[0088] }

[0089] Step S103: Based on the query results in the MCP context, the multi-dimensional network analysis agent is invoked in parallel to perform specialized diagnosis, and the specialized diagnosis results are written into the MCP context.

[0090] Specifically, based on the query results in the MCP context, the system calls multiple predefined network analysis agents in parallel to perform specialized diagnostics, and each network analysis agent communicates with the main control system through the MCP protocol.

[0091] In some embodiments, the step of invoking a multi-dimensional network analysis agent in parallel to perform specialized diagnosis based on the query results in the MCP context, and writing the specialized diagnosis results into the MCP context, specifically includes:

[0092] Based on the query results in the MCP context, at least two network analysis agents from the following dimensions are invoked in parallel to perform corresponding specialized diagnoses, and the specialized diagnoses are written into the MCP context:

[0093] The weak coverage analysis agent is invoked to calculate the RSRP (Reference Signal Receiving Power) / RSRQ (Reference Signal Receiving Quality) statistical value of the spatiotemporal point where the user is located, and to determine whether the statistical value meets the weak coverage determination threshold.

[0094] Call the high-load analysis agent to determine whether the user's cell is under high load.

[0095] The uplink interference analysis agent is invoked to query the average uplink interference level on the PRB bandwidth of each physical resource block in the user's cell, and to determine whether the average uplink interference level exceeds a preset threshold.

[0096] Specifically, the following network analysis agents are preferred for performing specialized diagnostics:

[0097] 1) Call the weak coverage analysis agent, read the user location and time information in the MCP context, calculate the RSRP / RSRQ statistics of the spatiotemporal point where the user is located, and determine whether the statistics meet the weak coverage judgment threshold;

[0098] 2) Call the high-load analysis agent to read the base station and time information in the MCP context, calculate the PRB occupancy rate and number of online users in the 4G / 5G cell where the user is located, and determine whether the cell is in a high-load state;

[0099] 3) Invoke the uplink interference analysis agent, read the cell information in the MCP context, query the average uplink interference level per PRB bandwidth of the user's cell, and determine whether the average uplink interference level exceeds a preset threshold.

[0100] Specifically, the specific diagnostic results output by each network analysis agent can carry the corresponding confidence level, and the specific diagnostic results and their confidence levels are uniformly written into the MCP context.

[0101] Step S104: The root cause reasoning agent summarizes the diagnostic results of each specific item in the MCP context, performs comprehensive analysis and root cause localization, and writes the root cause localization conclusion into the MCP context.

[0102] Specifically, the root cause reasoning agent is the top-level decision-making unit. Its core function is to receive and integrate the output results of all specialized agents (while combining B-domain order data), and generate the final, high-confidence root cause localization conclusion through preset reasoning rules or lightweight fusion models (such as weighted voting or rule engines).

[0103] In some embodiments, the step of summarizing the specific diagnostic results of each item in the MCP context through the root cause reasoning agent and performing comprehensive judgment and root cause localization specifically includes:

[0104] Multiple specialized analysis agents are invoked in parallel via the MCP protocol to perform specialized task analysis, and the results of the specialized task analysis are written into the MCP context. Each specialized analysis agent performs specialized task analysis based on the data in the MCP context.

[0105] The root cause reasoning agent reads the specific diagnostic results and the specific task analysis results in the MCP context, combines them with B-domain order data to perform multi-dimensional evidence fusion analysis, performs comprehensive judgment, and obtains the root cause judgment result and the corresponding confidence level.

[0106] If the confidence level is greater than or equal to a preset threshold, the root cause determination result is determined as the final root cause localization conclusion; otherwise, the root cause determination result and related analysis data are pushed to a human agent for review.

[0107] Specifically, the specialized analytical agent includes:

[0108] The current complaint semantic analysis agent is used to analyze user sentiment and determine the suddenness of complaint issues.

[0109] A historical duplicate complaint analysis agent is used to determine whether a user has filed duplicate complaints and whether the locations of the duplicate complaints are the same; and,

[0110] A network problem root cause analysis agent is used to determine the fault status of the base station associated with the location grid of the complaint and generate handling strategy suggestions.

[0111] Specifically, the system triggers a hierarchical collaborative intelligent analysis process through the MCP protocol. First, it calls the three specialized analysis agents mentioned above in parallel to complete in-depth diagnosis and writes the structured intermediate results back to the MCP context. Then, the root cause reasoning agent integrates the analysis results of all specialized tasks and combines them with the specialized diagnostic results of multi-dimensional network analysis to complete the final root cause reasoning.

[0112] It should be noted that the MCP context can pre-store B-domain order data. Specifically, while generating a structured query statement for the O-domain database based on the key field information, a structured query statement for the B-domain order database is simultaneously generated based on the key field information. After executing the two types of structured query statements, the retrieved B-domain order data is written into the MCP context in a standard data format. Based on this, the root cause reasoning agent can directly read the specific diagnostic results, specific task analysis results, and B-domain order data in the MCP context, conduct multi-dimensional evidence fusion analysis, and perform comprehensive judgment to finally obtain the root cause judgment result and the corresponding confidence level.

[0113] Step S105: Generate a structured analysis report based on the root cause localization conclusion and end the MCP session.

[0114] Specifically, the structured analysis report is a comprehensive diagnostic document for operations and customer service personnel, and its content includes, but is not limited to, the following:

[0115] Basic complaint information: work order ID, complaint time, original complaint text, etc.;

[0116] Final root cause analysis: Root cause identified, confidence score, and root cause type (network / service side).

[0117] Multi-dimensional analysis of the evidence chain: key results and data support from semantic analysis, historical complaint analysis, network data analysis, and B-domain order data;

[0118] Recommended network optimization measures: network optimization measures or service processing guidelines targeting root causes;

[0119] Related resource information: auxiliary location information such as network topology and geographical location;

[0120] Report metadata: analysis time, agent version, report ID, etc.

[0121] Specifically, after generating a structured analysis report and archiving the analysis data, the system officially closes the current MCP session.

[0122] In one specific embodiment, a root cause analysis method for user complaints based on a multi-task intelligent agent is provided. By introducing AI capabilities and integrating large language model capabilities, and through multi-task parallel processing and pre-configured prompt word scripts, the large model analyzes complaint data and network data according to the analytical thinking of cloud network engineers. This enables real-time intelligent analysis of user complaints, improving the timeliness of complaint handling. Furthermore, it locates network problems based on O-domain data, and uses intelligent methods to handle user complaints, improving processing efficiency, increasing user satisfaction, and ultimately enhancing the company's reputation. Figure 2As shown, the root cause analysis method for user complaints may include the following steps:

[0123] Step S1: Receive user complaint events and initialize the MCP session using the Kafka data collection tool.

[0124] The system receives user complaint tickets from the customer service system, user self-service terminals, or third-party channels. The ticket information includes at least the user identifier (MSISDN), complaint timestamp, geographical location of the complaint (latitude and longitude or cell ID), and a description of the problem (e.g., "video buffering," "unable to access"). The system creates a unique MCP session for this complaint event and initializes an MCP context. This context serves as the global data space for this analysis, used to store and transmit all shared information during the analysis process.

[0125] The initialized MCP context is a structured data object (usually serialized in JSON format) that follows a predefined schema. Its core lies in encapsulating task-driven semantic information. Specific content includes: (1) the user's original complaint text and complaint channel metadata (such as user complaint ticket ID, access timestamp, and channel type); (2) standardized network entity identifiers; (3) a pre-defined, structured set of analysis task instructions; (4) a list of roles and capabilities of the AI ​​Agents participating in the analysis; (5) global session status and unique IDs for process tracking; and (6) metadata of the MCP protocol itself. These contents together constitute a context that guides and constrains the collaborative analysis behavior of all subsequent AI Agents.

[0126] Step S2: Invoke the NLP-to-SQL agent to perform the conversion from natural language to data query.

[0127] The system sends a call request to the NLP-to-SQL Agent via the MCP protocol, using the context of the current MCP session as input. This agent utilizes its built-in natural language understanding model to parse the semantics of the user's description, identify key field information (such as service type and symptom labels), and dynamically generate structured SQL query statements for the O-domain database. After executing the query, the NLP-to-SQL agent performs preliminary cleaning and structuring processing on the acquired raw network data (such as user-level signaling, MR measurement reports, and cell-level KPI indicators), and writes the results into the MCP context in a standard data format. The input and processing logic of the NLP-to-SQL agent is as follows:

[0128] (1) Input:

[0129] Original user complaint text (natural language), for example: "Around 8 pm last night, when I was browsing Douyin near Central Street in Harbin, it was extremely laggy and kept spinning in circles."

[0130] Complaint channel metadata, such as user complaint ticket ID, access timestamp, and channel type, is preferred. Among these, access timestamp and channel type are optional fields. However, in actual deployment, providing the access timestamp directly can significantly improve system robustness and analysis timeliness, and avoid cross-system query delays or failures. Therefore, complaint channel metadata preferably includes user complaint ticket ID and access timestamp.

[0131] (2) Processing logic:

[0132] Semantic parsing: The following subtasks are performed by the NLP-to-SQL agent:

[0133] Entity recognition: Extract MSISDN (associated through work orders), time phrase ("around 8 pm yesterday" → convert to UTC time window [2025-11-23T19:30:00, 2025-11-23T20:30:00]), and geographic location ("Harbin Central Avenue" → call GIS service to map to latitude and longitude range or candidate cell list);

[0134] Intent Classification: Identify business type ("Scrolling through Douyin" → classified as "high-throughput short video business");

[0135] Symptom keyword extraction: "stuck" and "looping" → mapped to abnormal QoE indicators (such as low downlink speed, high latency, and high retransmission rate).

[0136] Structured encapsulation: The above results are organized into a standard JSON object with the following fields: user identifier, time window, geographic location range, business type, and symptom tag.

[0137] {

[0138] "msisdn": "138XXXX1234",

[0139] "time_window": {"start": "2025-11-23T19:30:00Z", "end": "2025-11-23T20:30:00Z"},

[0140] "geo_hint": {"lat_min": 45.75, "lat_max": 45.77, "lon_min": 126.62, "lon_max": 126.64},

[0141] "service_type": "video_streaming",

[0142] "symptom_tags": ["low_throughput", "high_latency"]

[0143] }

[0144] It should be noted that the core of this step is to transform unstructured complaints into standardized state units in the MCP context, providing a unified and reliable starting point for reasoning for all subsequent agents.

[0145] Step S3: Invoke the multi-dimensional network analysis agent in parallel to perform specialized diagnostics.

[0146] Based on preliminary data in the MCP context, the system invokes multiple predefined network analysis agents in parallel. Each agent communicates with the main control system via the MCP protocol.

[0147] (1) Coverage Analysis Agent: Reads user location and time information (i.e. time window) from the MCP context, calculates the RSRP / RSRQ statistics of the spatiotemporal point, determines whether the weak coverage threshold is met (e.g., RSRP < -110dBm percentage > 30%), and writes the diagnosis results (Boolean value and confidence level) back to the context.

[0148] (2) Load Analysis Agent: Reads base station and time information in the MCP context, queries and calculates indicators such as downlink PRB average occupancy rate, uplink PRB average occupancy rate, and number of online users of 4G / 5G cells, determines whether the cell is in a high-load state, and writes the results back to the context.

[0149] (3) Uplink Interference Agent: Obtain cell information from the MCP context, query the average uplink interference level on each PRB bandwidth, determine whether it exceeds the preset threshold (e.g., >-98dBm), and write the interference level assessment result into the context.

[0150] It should be noted that the above-mentioned multi-dimensional network analysis agents can be extended according to actual network analysis needs, and may also include other specialized analysis agents such as handover analysis agents and core network signaling analysis agents. Each extended agent follows the MCP protocol to complete data interaction with the context and write back specialized diagnostic results.

[0151] Step S4: Invoke the root cause reasoning agent to perform comprehensive analysis and root cause localization.

[0152] The system invokes the Root Cause Inference Agent via the MCP protocol. This agent accesses the complete MCP context, summarizing the diagnostic results of all specialized analysis agents, raw network data, and user complaint descriptions. The agent incorporates a comprehensive inference engine based on an expert knowledge base and machine learning models. Employing algorithms such as weighted decision-making, rule-based reasoning, or graph neural networks, it fuses and analyzes multi-dimensional evidence, outputting the most probable root cause of the network problem (e.g., "5G cell uplink interference," "4G cell high load," "weak coverage"), and generating a confidence score for the root cause determination.

[0153] Task 1: The current complaint semantic analysis agent, based on the capabilities of a large model, processes several tasks in steps, including "complainant sentiment analysis" and "whether the user complaint issue is sudden."

[0154] Data input:

[0155] {

[0156] "user_id": "User ID",

[0157] "conversation": [

[0158] {"role": "user", "text": "What the user said"},

[0159] {"role": "agent", "text": "Customer service's response"}, ... ]

[0162] }

[0163] Task breakdown: The entire task is broken down into the following sub-tasks:

[0164] 1) Sentiment Analysis: Analyze the user's emotions (such as anger, anxiety, calmness, etc.).

[0165] 2) Is the problem sudden? Determine whether the problem reported by users is long-standing or sudden.

[0166] 3) Location Extraction: Extract the actual location of the user's complaint from the dialogue.

[0167] 4) Whether the location is within the province: Determine whether the location is within the province based on the extracted location (providing the province range is required).

[0168] 5) Classification of network problems: Classify network problems reported by users (such as weak signal, inability to access the Internet, disconnection, etc.).

[0169] The task is executed by calling a large model.

[0170] Task 2: Analyze duplicate complaints within the past 30 days using an intelligent agent. Based on a large model, process several tasks step-by-step, including "Does this user have duplicate complaints?" and "Are the user's duplicate complaints from the same location?".

[0171] Data input:

[0172] {

[0173] "user_id": "User ID",

[0174] "conversations": [

[0175] {

[0176] "timestamp": "time stamp", # Complaint time

[0177] "conversation": [

[0178] {"role": "user", "text": "What the user said"},

[0179] {"role": "agent", "text": "Customer service's response"}, ... ]

[0182] }, ... ]

[0185] }

[0186] Task breakdown: The entire task is broken down into the following sub-tasks:

[0187] 1) Whether there are repeated complaints: Determine whether the user has made multiple complaints within 30 days.

[0188] 2) Whether it is the same location: If there are duplicate complaints, determine whether these complaints occurred in the same location.

[0189] 3) Risk of escalation of emotions: Analyze the trend of users' emotional changes to determine whether there is a risk of escalating complaints.

[0190] 4) Whether it is the same type of network problem: Determine whether the repeated complaints belong to the same type of network problem.

[0191] The task is executed by calling a large model.

[0192] Task 3: Network problem root cause analysis intelligent agent, based on the large model capability, to process several NLP to SQL data query tasks such as "user complaint location grid" and "user complaint grid associated base station information" in steps;

[0193] Complaint data input:

[0194] {

[0195] "user_id": "User ID",

[0196] "conversation": [

[0197] {"role": "user", "text": "What the user said"},

[0198] {"role": "agent", "text": "Customer service's response"}, ... ]

[0201] }

[0202] Geographic location grid information:

[0203] {

[0204] "grid_id": "grid ID",

[0205] "associated_base_station_id": "Associated base station ID"

[0206] }

[0207] Network performance data:

[0208] {

[0209] "cell_id": "cell ID",

[0210] "base_station_id": "Base station ID",

[0211] "coverage": "Coverage status (normal / weak coverage)",

[0212] "load": "Load status (normal / high load)",

[0213] "quality": "Quality status (normal / poor)"

[0214] "fault_status": "Is there a fault (yes / no)?"

[0215] }

[0216] Task breakdown: The entire task is broken down into the following sub-tasks:

[0217] 1) Extract the raster information of the user complaint location.

[0218] 2) Obtain the base station information associated with the user complaint grid.

[0219] 3) Determine if the base station is faulty.

[0220] 4) Determine whether the cell in the base station has weak coverage.

[0221] 5) Determine if the cells in the base station are under high load.

[0222] 6) Determine if the cell in the base station is of poor quality.

[0223] 7) Generate strategy suggestions for problem handling.

[0224] The task is executed by calling a large model.

[0225] It should be noted that the root cause reasoning agent and three specialized analysis agents (the current complaint semantic analysis agent, the historical 30-day repeated complaint analysis agent, and the network problem root cause analysis agent) constitute a hierarchical and collaborative intelligent analysis architecture. The three specialized agents are parallel execution analysis units, each responsible for conducting in-depth diagnosis of complaints from three dimensions: semantic understanding, historical patterns, and network status. They output structured intermediate results and write them into the MCP context. The root cause reasoning agent is the top-level decision-making unit. Its core role is to receive and integrate the output results of all specialized agents (while also combining B-domain order data), and generate the final, high-confidence root cause localization conclusion through preset reasoning rules or lightweight fusion models (such as weighted voting or rule engines). The input data for each specialized analysis agent comes from the content already stored in the MCP context when executing this step.

[0226] Step S5: Generate a structured analysis report and push processing suggestions.

[0227] The system integrates the conclusions of the root cause reasoning agent, the analysis process and data basis of each specialized agent, and recommended network optimization measures (such as "It is recommended to check the interference source of XX cell" or "It is recommended to expand the capacity of XX cell") into a structured intelligent analysis report. The report content follows a standard template to ensure information completeness and logical clarity.

[0228] It should be noted that the root cause reasoning agent has a built-in expert knowledge base, which predefines the mapping relationship between common network problem root causes and corresponding network optimization measures. For example, when the root cause is determined to be "5G cell uplink interference," the agent will automatically match and associate it with the handling suggestion of "verifying the interference source"; when the root cause is determined to be "4G cell high load," the agent will automatically match and associate it with the handling suggestion of "evaluating capacity expansion or implementing load balancing." Based on this, while outputting the root cause conclusion and corresponding confidence level in step S4, the root cause reasoning agent will simultaneously output a list of matching network optimization measures. Step S5 simply integrates this list of suggestions along with the diagnostic evidence and analysis process of each specialized agent into the structured analysis report.

[0229] The structured analysis report is a comprehensive diagnostic document for operations and customer service personnel, and its content includes, but is not limited to, the following:

[0230] Basic complaint information: work order ID, complaint time, original complaint text, etc.;

[0231] Final root cause analysis: Root cause identified, confidence score, and root cause type (network / service side).

[0232] Multi-dimensional analysis of the evidence chain: key results and data support from semantic analysis, historical complaint analysis, network data analysis, and B-domain order data;

[0233] Recommended network optimization measures: network optimization measures or service processing guidelines targeting root causes;

[0234] Related resource information: auxiliary location information such as network topology and geographical location;

[0235] Report metadata: analysis time, agent version, report ID, etc.

[0236] Step S6: Result persistence and visualization.

[0237] The system parses and stores the structured analysis report generated by S5 (including the final root cause localization conclusion, multi-dimensional analysis evidence chain, recommended network optimization measures, etc.) into a local results database. Network optimization personnel can query the corresponding analysis results through the web front-end interface based on the work order ID, user identifier, or network entity (such as cell ID) to achieve efficient review and handling tracking.

[0238] Step S7: End the MCP session and archive the analysis data.

[0239] After the analysis results are successfully pushed, the system officially closes the current MCP session. At the same time, all context data, agent interaction logs, final report, and other key information from this analysis process are anonymized and stored in the historical analysis database for subsequent model training, system optimization, and iterative updates of the root cause knowledge base.

[0240] It should be noted that the root cause analysis method for user complaints provided in this disclosure has the following beneficial effects:

[0241] a) Improve processing efficiency and automation: By introducing AI capabilities and using NLP technology, the natural language descriptions of user complaints are automatically converted into structured data queries (NLP-to-SQL), replacing the traditional cumbersome manual cross-system query operations, greatly reducing the human intervention process, effectively shortening the analysis cycle of the root causes of complaints, and improving the overall processing efficiency.

[0242] b) Enhance the objectivity and accuracy of the analysis: Construct multiple network analysis agents such as weak coverage analysis, high load analysis, and uplink interference analysis. Through the MCP protocol, realize the efficient parallel and collaborative analysis of each agent, carry out root cause reasoning by integrating multi-dimensional evidence, get rid of excessive reliance on human experience, reduce human interference, and improve the objectivity and accuracy of root cause analysis results.

[0243] c) Solving the challenges of system collaboration and context management: Innovatively adopting the MCP protocol as a unified communication standard between intelligent agents, relying on its built-in context mechanism, it globally and uniformly manages user identifiers, time windows, geographical locations, and intermediate results of each stage of analysis in the analysis session, breaking down information silos in traditional systems and ensuring information consistency and analysis process continuity throughout the entire process.

[0244] d) Achieving system openness and scalability: Based on the MCP protocol, an open intelligent analysis architecture is built. For new analysis needs such as new services and new network standards, the corresponding analysis capabilities can be implemented in the form of standardized intelligent agents in a "plug-and-play" manner. Without the need for large-scale modification of the existing system, it can be quickly integrated into the existing analysis system, which greatly reduces the cost of system upgrades and function expansion, and lays a solid technical foundation for the future intelligent evolution of network operation and maintenance.

[0245] The user complaint root cause analysis method provided in this disclosure creates an MCP session and initializes a unified MCP context. It leverages the MCP protocol to automate NLP-to-SQL agent queries and standardizes data writing to the O-domain database. Combined with parallel specialized diagnosis by a multi-dimensional network analysis agent and comprehensive judgment by a root cause reasoning agent, the entire process of complaint root cause analysis is automated through standardized protocol communication and global context management. This reduces reliance on manual intervention and improves the efficiency of complaint root cause analysis. Simultaneously, it utilizes multi-agent collaborative fusion of multi-dimensional evidence to locate the root cause. This approach eliminates over-reliance on human experience, providing unified judgment criteria for analytical conclusions and significantly improving the objectivity and accuracy of root cause analysis results. Furthermore, the unified communication standard and context management mechanism based on the MCP protocol breaks down information barriers between analysis modules, enabling efficient data sharing and deep collaborative analysis among agents, greatly enhancing the overall system's collaborative capabilities. In addition, the open intelligent analysis architecture built on the MCP protocol can easily integrate various new analytical agents without requiring large-scale customization of existing systems, effectively improving the system's intelligence level and scalability, and adapting to rapidly changing network analysis needs. It solves the problems of low processing efficiency, strong subjectivity in analysis results, poor system collaboration, and insufficient intelligence and scalability in existing user complaint root cause analysis methods.

[0246] It is understood that the various method embodiments mentioned above in this disclosure can be combined with each other to form combined embodiments without violating the principle and logic. Due to space limitations, this disclosure will not elaborate further. Those skilled in the art will understand that in the above methods of specific implementation, the specific execution order of each step should be determined by its function and possible internal logic.

[0247] Figure 3 This is a block diagram of a user complaint root cause analysis device provided in an embodiment of the present disclosure.

[0248] Reference Figure 3 This disclosure provides a user complaint root cause analysis apparatus for performing the above-described user complaint root cause analysis method. The apparatus includes:

[0249] The session creation and initialization module 11 is used to create a Model Context Protocol (MCP) session based on a user complaint work order and initialize the MCP context corresponding to the MCP session.

[0250] The semantic structured query module 12 is connected to the session creation and initialization module 11. It is used to call the natural language processing to structured query language NLP-to-SQL intelligent agent through the MCP protocol based on the initialized MCP context, to convert the natural language description of the user complaint into a structured query statement for the O domain database, execute the structured query statement and write the query result into the MCP context in a standard data format.

[0251] The intelligent agent-specific diagnostic module 13 is connected to the semantic structured query module 12. It is used to call the multi-dimensional network analysis intelligent agent in parallel to perform specific diagnostics based on the query results in the MCP context, and write the specific diagnostic results into the MCP context.

[0252] The comprehensive judgment and root cause localization module 14 is connected to the agent-specific diagnosis module 13. It is used to summarize the specific diagnosis results in the MCP context through the root cause reasoning agent, perform comprehensive judgment and root cause localization, and write the root cause localization conclusion into the MCP context.

[0253] The report generation and session termination module 15 is connected to the comprehensive analysis and root cause localization module 14, and is used to generate a structured analysis report based on the root cause localization conclusion and terminate the MCP session.

[0254] Optionally, the device further includes:

[0255] The data acquisition module is used to receive user complaint tickets from customer service systems, user self-service terminals, or third-party channels through the Kafka data acquisition tool.

[0256] Optionally, the initialized MCP context includes the user's original complaint text and complaint channel metadata, and the semantic structured query module 12 specifically includes:

[0257] The semantic structured query unit is used to input the original user complaint text and complaint channel metadata from the MCP context into the NLP-to-SQL agent via the MCP protocol, use the NLP-to-SQL agent to parse the semantics of the user complaint to obtain key field information, generate a structured query statement for the O domain database based on the key field information, and write the query result into the MCP context in a standard data format after executing the structured query statement.

[0258] Optionally, the semantic structured query unit includes:

[0259] The key field information acquisition unit is used to perform the following sub-tasks using the NLP-to-SQL agent to obtain key field information including user identifier, time window, geographical location range, business type and symptom label: extracting user identifier based on complaint channel metadata, extracting the time window and geographical location range of user complaints from the original user complaint text, classifying the original user complaint text to identify the corresponding business type, and extracting fault symptom labels from the original user complaint text.

[0260] The key field structured encapsulation unit is used to encapsulate the user identifier, time window, geographic location range, business type and symptom label in a structured manner using the NLP-to-SQL intelligent agent and then write it into the MCP context;

[0261] The structured query statement generation unit is used to generate structured query statements for the O-domain database based on the user identifier, time window, geographical location range, business type and symptom label encapsulated in the MCP context.

[0262] Optionally, the intelligent agent-specific diagnostic module 13 is specifically used for:

[0263] Based on the query results in the MCP context, at least two network analysis agents from the following dimensions are invoked in parallel to perform corresponding specialized diagnoses, and the specialized diagnoses are written into the MCP context:

[0264] The weak coverage analysis agent is invoked to calculate the statistical values ​​of the reference signal received power (RSRP) and the reference signal received quality (RSRQ) at the user's spatiotemporal point, and to determine whether the statistical values ​​meet the weak coverage determination threshold.

[0265] Call the high-load analysis agent to determine whether the user's cell is under high load.

[0266] The uplink interference analysis agent is invoked to query the average uplink interference level on the PRB bandwidth of each physical resource block in the user's cell, and to determine whether the average uplink interference level exceeds a preset threshold.

[0267] Optionally, the comprehensive analysis and root cause localization module 14 includes:

[0268] The special task analysis unit is used to call multiple special task analysis agents in parallel through the MCP protocol to perform special task analysis and write the special task analysis results into the MCP context. Each special task analysis agent performs special task analysis based on the data in the MCP context.

[0269] The multi-dimensional evidence fusion analysis unit is used to read the specific diagnosis results and the specific task analysis results in the MCP context through the root cause reasoning agent, combine them with B-domain order data to perform multi-dimensional evidence fusion analysis, perform comprehensive judgment, and obtain the root cause judgment result and the corresponding confidence level.

[0270] The root cause localization conclusion determination unit is used to determine the root cause judgment result as the final root cause localization conclusion in response to the confidence level being greater than or equal to a preset threshold; otherwise, the root cause judgment result and related analysis data are pushed to a human agent for review.

[0271] Optionally, the specialized analytical agent includes:

[0272] The current complaint semantic analysis agent is used to analyze user sentiment and determine the suddenness of complaint issues.

[0273] A historical duplicate complaint analysis agent is used to determine whether a user has filed duplicate complaints and whether the locations of the duplicate complaints are the same; and,

[0274] A network problem root cause analysis agent is used to determine the fault status of the base station associated with the location grid of the complaint and generate handling strategy suggestions.

[0275] Figure 4 This is a block diagram of an electronic device provided in an embodiment of the present disclosure.

[0276] Reference Figure 4 This disclosure provides an electronic device comprising: at least one processor 701; at least one memory 702; and one or more I / O interfaces 703 connected between the processor 701 and the memory 702; wherein the memory 702 stores one or more computer programs executable by the at least one processor 701, the one or more computer programs being executed by the at least one processor 701 to enable the at least one processor 701 to perform the aforementioned user complaint root cause analysis method.

[0277] This disclosure also provides a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the aforementioned user complaint root cause analysis method. The computer-readable storage medium may be volatile or non-volatile.

[0278] In summary, the user complaint root cause analysis method, apparatus, electronic device, and storage medium provided in this disclosure, by creating an MCP session and initializing a unified MCP context, and relying on the MCP protocol to achieve automated querying and standardized data writing of the O-domain database by the NLP-to-SQL agent, and combining the parallel specialized diagnosis of the multi-dimensional network analysis agent and the comprehensive judgment of the root cause reasoning agent, achieves fully automated progress of the complaint root cause analysis process through standardized protocol communication and global context management, reducing reliance on manual intervention and improving the efficiency of complaint root cause analysis processing; at the same time, through multi-agent collaborative fusion of multi-dimensional... This method utilizes high-level evidence for root cause analysis, reducing over-reliance on human experience and establishing unified judgment standards for analytical conclusions, significantly improving the objectivity and accuracy of root cause analysis results. Furthermore, the unified communication standard and context management mechanism based on the MCP protocol breaks down information barriers between analysis modules, enabling efficient data sharing and deep collaborative analysis among agents, greatly enhancing the overall system's collaborative capabilities. In addition, the open intelligent analysis architecture built on the MCP protocol can easily integrate various new analytical agents without requiring large-scale customization of existing systems, effectively improving the system's intelligence level and scalability, and adapting to rapidly changing network analysis needs. This addresses the problems of low processing efficiency, strong subjectivity in analysis results, poor system collaboration, and insufficient intelligence and scalability in existing user complaint root cause analysis methods.

[0279] Those skilled in the art will understand that all or some of the steps, systems, and apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software can be distributed on a computer-readable storage medium, which may include computer storage media (or non-transitory media) and communication media (or transient media).

[0280] As is known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable program instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), static random access memory (SRAM), flash memory or other memory technologies, portable compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, it is known to those skilled in the art that communication media typically contain computer-readable program instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0281] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0282] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0283] This disclosure has disclosed exemplary embodiments, and although specific terminology has been used, it is for general illustrative purposes only and should not be construed as limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in conjunction with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in conjunction with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of this disclosure as set forth by the appended claims.

Claims

1. A method for root cause analysis of user complaints, characterized in that, The method includes: Create a Model Context Protocol (MCP) session based on the user complaint work order, and initialize the MCP context corresponding to the MCP session; Based on the initialized MCP context, the Natural Language Processing to Structured Query Language (NLP-to-SQL) intelligent agent is called through the MCP protocol to convert the natural language description of the user's complaint into a structured query statement for the O domain database, execute the structured query statement, and write the query result into the MCP context in a standard data format. Based on the query results in the MCP context, the multi-dimensional network analysis agent is invoked in parallel to perform specialized diagnosis, and the specialized diagnosis results are written into the MCP context. The root cause reasoning agent summarizes the specific diagnostic results in the MCP context, performs comprehensive judgment and root cause localization, and writes the root cause localization conclusion into the MCP context. A structured analysis report is generated based on the root cause localization conclusions, and the MCP session is terminated.

2. The method according to claim 1, characterized in that, Before creating a Model Context Protocol (MCP) session based on a user complaint ticket, the method further includes: We use Kafka data collection tools to receive user complaint tickets from customer service systems, user self-service terminals, or third-party channels.

3. The method according to claim 1, characterized in that, The initialized MCP context includes the user's original complaint text and complaint channel metadata. Based on the initialized MCP context, the Natural Language Processing to Structured Query Language (NLP-to-SQL) intelligent agent is invoked through the MCP protocol to convert the natural language description of the user complaint into a structured query statement for the O-domain database. The structured query statement is executed, and the query results are written to the MCP context in a standard data format. Specifically, this includes: Using the MCP protocol, the original user complaint text and complaint channel metadata in the MCP context are input into the NLP-to-SQL agent. The NLP-to-SQL agent is used to parse the semantics of the user complaint to obtain key field information, and a structured query statement for the O domain database is generated based on the key field information. After executing the structured query statement, the query result is written into the MCP context in a standard data format.

4. The method according to claim 3, characterized in that, The process of using the NLP-to-SQL agent to parse the semantics of user complaints to obtain key field information, and generating a structured query statement for the O-domain database based on the key field information, specifically includes: The NLP-to-SQL agent is used to perform the following sub-tasks to obtain key field information including user identifier, time window, geographic location range, business type and symptom label: extracting user identifier based on complaint channel metadata, extracting the time window and geographic location range of user complaints from the original user complaint text, classifying the original user complaint text to identify the corresponding business type, and extracting fault symptom labels from the original user complaint text. The NLP-to-SQL agent is used to encapsulate the user identifier, time window, geographic location range, business type, and symptom tags in a structured manner, and then writes them into the MCP context. Based on the user identifier, time window, geographic location range, business type, and symptom tags encapsulated in the MCP context, a structured query statement for the O-domain database is generated.

5. The method according to claim 1, characterized in that, The process of using the query results in the MCP context to invoke a multi-dimensional network analysis agent in parallel to perform specialized diagnostics, and then writing the specialized diagnostic results into the MCP context, specifically includes: Based on the query results in the MCP context, at least two network analysis agents from the following dimensions are invoked in parallel to perform corresponding specialized diagnoses, and the specialized diagnoses are written into the MCP context: The weak coverage analysis agent is invoked to calculate the statistical values ​​of the reference signal received power (RSRP) and the reference signal received quality (RSRQ) at the user's spatiotemporal point, and to determine whether the statistical values ​​meet the weak coverage determination threshold. Call the high-load analysis agent to determine whether the user's cell is under high load. The uplink interference analysis agent is invoked to query the average uplink interference level on the PRB bandwidth of each physical resource block in the user's cell, and to determine whether the average uplink interference level exceeds a preset threshold.

6. The method according to claim 1, characterized in that, The process of summarizing the specific diagnostic results from the MCP context through a root cause reasoning agent and performing comprehensive analysis and root cause localization specifically includes: Multiple specialized analysis agents are invoked in parallel via the MCP protocol to perform specialized task analysis, and the results of the specialized task analysis are written into the MCP context. Each specialized analysis agent performs specialized task analysis based on the data in the MCP context. The root cause reasoning agent reads the specific diagnostic results and the specific task analysis results in the MCP context, combines them with B-domain order data to perform multi-dimensional evidence fusion analysis, performs comprehensive judgment, and obtains the root cause judgment result and the corresponding confidence level. If the confidence level is greater than or equal to a preset threshold, the root cause determination result is determined as the final root cause localization conclusion; otherwise, the root cause determination result and related analysis data are pushed to a human agent for review.

7. The method according to claim 6, characterized in that, The specialized analytical agent includes: The current complaint semantic analysis agent is used to analyze user sentiment and determine the suddenness of complaint issues. A historical duplicate complaint analysis agent is used to determine whether a user has filed duplicate complaints and whether the locations of the duplicate complaints are the same; and, A network problem root cause analysis agent is used to determine the fault status of the base station associated with the location grid of the complaint and generate handling strategy suggestions.

8. A user complaint root cause analysis device, characterized in that, The device includes: The session creation and initialization module is used to create a Model Context Protocol (MCP) session based on a user complaint work order and initialize the MCP context corresponding to the MCP session. The semantic structured query module, connected to the session creation and initialization module, is used to call the Natural Language Processing to Structured Query Language (NLP-to-SQL) intelligent agent through the MCP protocol based on the initialized MCP context. This agent converts the natural language description of the user's complaint into a structured query statement for the O domain database, executes the structured query statement, and writes the query results into the MCP context in a standard data format. The agent-specific diagnostic module is connected to the semantic structured query module. It is used to call the multi-dimensional network analysis agent in parallel to perform specific diagnostics based on the query results in the MCP context, and write the specific diagnostic results into the MCP context. The comprehensive judgment and root cause localization module is connected to the agent-specific diagnosis module. It is used to summarize the various specific diagnosis results in the MCP context through the root cause reasoning agent, perform comprehensive judgment and root cause localization, and write the root cause localization conclusion into the MCP context. The report generation and session termination module is connected to the comprehensive analysis and root cause localization module, and is used to generate a structured analysis report based on the root cause localization conclusion and terminate the MCP session.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores one or more computer programs that can be executed by the at least one processor, the one or more computer programs being executed by the at least one processor to enable the at least one processor to perform the user complaint root cause analysis method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the root cause analysis method for user complaints as described in any one of claims 1-7.

Images