Problem positioning method, device and electronic equipment for wireless network complaint scene

Abstract

The application provides a problem positioning method and device for a wireless network complaint scene and electronic equipment, relates to the technical field of wireless networks, adopts a bottom-up layered progressive architecture, processes mass original expansion detail record data layer by layer, reduces data scale layer by layer, reduces calculation complexity and subsequent calculation pressure, significantly reduces resource occupation, reduces the amount of processing resources required, improves overall operation efficiency and response speed to user complaints. Especially in the scene of mass data of provincial networks, this method is more easily implemented and can efficiently support larger-scale real-time or quasi-real-time complaint backtracking requirements. Moreover, relying on the dynamic logical reasoning capability of the large language model, the occurrence time period of each structured abnormal event is adaptively merged, which can greatly improve the positioning accuracy of the target problem time period, improve the accurate positioning capability of user complaints, and realize automatic accurate backtracking positioning for user complaint scenes.

Description

Technical Field

[0001] This invention relates to the field of wireless network technology, and in particular to a problem localization method, apparatus, and electronic device for wireless network complaint scenarios. Background Technology

[0002] With the continuous development of mobile communication technology, handling wireless network complaints and detecting performance anomalies in poor-quality cells have become crucial aspects of network operation and maintenance. Faced with a massive volume of user complaints, network operation and maintenance systems need to quickly trace and accurately pinpoint network issues causing a decline in user experience.

[0003] In terms of technical implementation, one approach is to use a multi-agent collaborative model to handle the diagnostic tasks of poor-quality wireless network cells. However, this approach can only process key performance indicators (KPIs) or alarm information at the cell level, resulting in coarse data processing granularity and an inability to accurately locate network problems.

[0004] Another approach is to perform quality difference analysis by precisely matching extended detail record (XDR) data and measurement report (MR) data at the user level. However, this approach suffers from high computational complexity and high resource requirements, making it inefficient for supporting real-time or near-real-time complaint backtracking analysis. Summary of the Invention

[0005] This invention provides a method, apparatus, and electronic device for locating problems in wireless network complaint scenarios, in order to address the deficiencies existing in related technologies.

[0006] This invention provides a problem localization method for wireless network complaint scenarios, including: Obtain the original extended details record data of the wireless network prior to the user's complaint, and determine the preliminary abnormal event based on the original extended details record data; The preliminary abnormal events are grouped and statistically analyzed to obtain a structured list of abnormal events; Based on the large language model, the occurrence time periods of each structured abnormal event in the structured abnormal event list are aggregated to obtain a candidate problem time period set; Each candidate issue time period in the candidate issue time period set is scored, and the target issue time period for the user complaint is determined based on the score of each candidate issue time period.

[0007] According to the present invention, a problem localization method for wireless network complaint scenarios includes, based on a large language model, aggregating the occurrence time periods of each structured abnormal event in the structured abnormal event list to obtain a candidate problem time period set, including: Based on the large language model, the time window aggregation tool and the event time period merging tool are invoked to merge the time periods of each of the structured abnormal events by cell to obtain each initial merged time period; For each of the initial merging time periods that meet the merging conditions and correspond to different cells, the geographic location calculation tool is invoked based on the large language model to perform spatial continuity verification on the cells corresponding to the candidate merging time periods. If the spatial continuity verification passes, the event time period merging tool is invoked to merge the candidate merging time periods to obtain the candidate problem time period set.

[0008] According to the present invention, a problem localization method for wireless network complaint scenarios includes, based on the large language model, invoking a time window aggregation tool and an event time period merging tool to merge the structured abnormal events by cell to obtain initial merged time periods, including: Based on the large language model, the time window aggregation tool is invoked to aggregate the structured abnormal events by cell into time windows, thereby obtaining each initial aggregation period. For each of the initial aggregation time periods that meet the merging conditions and correspond to the same cell, based on the large language model, the anomaly density analysis tool is invoked to perform anomaly event density verification on the candidate aggregation time periods. If the anomaly event density verification passes, the event time period merging tool is invoked to merge the candidate aggregation time periods of the same cell to obtain each of the initial merging time periods.

[0009] According to the present invention, a problem localization method for wireless network complaint scenarios includes, based on the large language model, invoking a geographic location calculation tool to perform spatial continuity verification on the cells corresponding to the candidate merging time periods, comprising: Based on the large language model, the geographic location calculation tool is invoked to calculate the distance between the cells corresponding to the candidate merging time periods; If the distance is less than the spatial distance threshold, then the spatial continuity check is deemed to have passed. If the distance is greater than or equal to the spatial distance threshold, then the spatial continuity check is determined to fail.

[0010] According to the present invention, a problem localization method for wireless network complaint scenarios includes, based on the large language model, invoking a geolocation calculation tool to perform spatial continuity verification on the cells corresponding to the candidate merging time periods, followed by: Based on the large language model, an anomaly density analysis tool is invoked to verify the anomaly event density of the candidate merging time periods; If both the abnormal event density check and the spatial continuity check pass, the event time period merging tool is invoked to merge the candidate merging time periods to obtain the candidate problem time period set.

[0011] According to the present invention, a problem localization method for wireless network complaint scenarios includes, based on the large language model, invoking an anomaly density analysis tool to verify the anomaly event density of the candidate merging time periods, comprising: Based on the large language model, the anomaly density analysis tool is invoked to calculate the total length of the candidate merging time periods, and the anomaly event density is calculated based on the total length of the time periods and the number of anomalies in the candidate merging time periods. If the density of abnormal events is within a preset density range, then the abnormal event density verification is deemed to have passed. If the density of abnormal events is outside the preset density range, then the abnormal event density verification is determined to be unsuccessful.

[0012] According to a problem localization method for wireless network complaint scenarios provided by the present invention, the method further includes, after invoking the event time period merging tool to merge the candidate merging time periods to obtain the candidate problem time period set, the method further includes: For any candidate problem time period in the set of candidate problem time periods, a final fallback check is performed on the duration of any candidate problem time period, the distance between the corresponding cells, and the density of abnormal events.

[0013] According to the present invention, a method for locating problems in a wireless network complaint scenario includes scoring each candidate problem time period in the candidate problem time period set, which includes: The score for each candidate problem period is calculated based on the number of abnormal events, the length of the period, and the influence weight of the business type.

[0014] According to the present invention, a problem localization method for wireless network complaint scenarios is provided, wherein the influence weight of the service type is dynamically adjusted based on the large language model using the target service type of the user complaint.

[0015] According to the present invention, a problem localization method for wireless network complaint scenarios includes, in which the preliminary abnormal events are grouped and statistically analyzed to obtain a structured abnormal event list, including: From the initial abnormal events, select event records with clear abnormal markers; Using a grouping statistics tool, the event records are grouped and statistically analyzed according to service type and cell identifier to generate the structured abnormal event list.

[0016] The present invention also provides a problem location device for wireless network complaint scenarios, comprising: The raw data processing module is used to acquire the raw extended details record data of the wireless network before the user complaint, and to determine the preliminary abnormal event based on the raw extended details record data; The event extraction module is used to group and statistically analyze the preliminary abnormal events to obtain a structured list of abnormal events; The time period aggregation module is used to aggregate the occurrence time periods of each structured abnormal event in the structured abnormal event list based on the large language model to obtain a candidate problem time period set. The problem location module is used to score each candidate problem time period in the candidate problem time period set, and determine the target problem time period of the user complaint based on the score of each candidate problem time period.

[0017] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the problem localization method for wireless network complaint scenarios as described above.

[0018] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the problem localization method for wireless network complaint scenarios as described above.

[0019] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the problem localization method for wireless network complaint scenarios as described above.

[0020] The present invention provides a problem localization method, device, and electronic device for wireless network complaint scenarios. It employs a bottom-up, layered, progressive architecture, processing massive amounts of raw extended detail record data layer by layer to reduce data size, computational complexity, and subsequent computational pressure. This significantly reduces resource consumption and processing resource requirements, improving overall operational efficiency and response speed to user complaints. Especially in the context of massive data in provincial networks, this method is easier to implement and can efficiently support larger-scale real-time or near-real-time complaint backtracking needs. Furthermore, relying on the dynamic logical reasoning capabilities of a large language model, the method adaptively merges the occurrence periods of various structured abnormal events. This not only significantly improves the localization accuracy of the target problem period and enhances the precise problem localization capability for user complaints, enabling automated and accurate backtracking and localization for user complaint scenarios, ensuring that abnormal events are not missed or falsely reported, but also reduces human intervention, further improving problem localization efficiency and response speed to user complaints. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in this invention or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating the problem localization method for wireless network complaint scenarios provided by the present invention.

[0023] Figure 2 This is a schematic diagram of the pyramid-shaped hierarchical architecture in the problem localization method for wireless network complaint scenarios provided by the present invention.

[0024] Figure 3 This is a flowchart illustrating the application of a large language model in the problem localization method for wireless network complaint scenarios provided by this invention.

[0025] Figure 4 This is a schematic diagram of the problem location device for wireless network complaint scenarios provided by the present invention.

[0026] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0028] In existing technologies, for wireless network complaint handling scenarios, one solution for problem localization is to use a multi-agent collaborative mode to handle the diagnostic tasks of poor wireless network quality cells. This solution can only handle cell-level KPIs or alarm information, and the data processing granularity is coarse, making it difficult to quickly reconstruct the precise problem time period of the user's complaint.

[0029] Another approach involves precisely matching XDR and MR data at the user level for quality difference analysis. However, this method is complex, resource-intensive, and struggles to efficiently support large-scale real-time or near-real-time complaint retrospective analysis. This approach employs a single-layer data processing flow, resulting in high overall latency and failing to provide aggregation and scoring mechanisms tailored to specific time periods, thus limiting its applicability in real-time operations and maintenance scenarios. Furthermore, its static data matching method is ill-suited to scenarios with varying data densities, easily leading to missed or false positives.

[0030] In addition, another approach is based on a Retrieval-augmented Generation (RAG) database combined with a multi-layered knowledge graph. It uses collected Non-Access Stratum (NAS) or Radio Resource Control (RRC) signaling text for retrieval, inputting the search results along with prompts into a large language model to output signaling analysis conclusions. This approach only utilizes NAS or RRC signaling text and does not apply XDR data. Furthermore, the large language model can only provide macro-level signaling analysis conclusions and cannot perform fine-grained problem localization.

[0031] Based on this, this embodiment of the invention provides a problem localization method for wireless network complaint scenarios to solve the above-mentioned problems.

[0032] Figure 1 This is a flowchart illustrating a problem localization method for wireless network complaint scenarios provided in an embodiment of the present invention, as shown below. Figure 1 As shown, the method includes: S1, Obtain the original extended details record data of the wireless network before the user complaint, and determine the preliminary abnormal event based on the original extended details record data; S2, group and statistically analyze the preliminary abnormal events to obtain a structured list of abnormal events; S3, based on the large language model, aggregate the occurrence time periods of each structured abnormal event in the structured abnormal event list to obtain a candidate problem time period set; S4, score each candidate issue time period in the candidate issue time period set, and determine the target issue time period for the user complaint based on the score of each candidate issue time period.

[0033] Specifically, the problem localization method for wireless network complaint scenarios provided in this embodiment of the invention is executed by a problem localization device for wireless network complaint scenarios. This device can be an intelligent agent that can be configured in a server, cloud computing platform, network management and maintenance system, or a dedicated quality diagnosis device, etc., and is not specifically limited here.

[0034] like Figure 2As shown, this method can process data layer by layer through a pyramid-shaped hierarchical architecture to ultimately determine the target time period of user complaints. The pyramid-shaped hierarchical architecture, from bottom to top, includes: Raw Data Layer, Event Extraction Layer, Time Segment Aggregation Layer, and Problem Localization Layer, corresponding to steps S1-S4 in the method steps, respectively.

[0035] First, step S1 is executed. After receiving a user complaint, the raw XDR data of the wireless network prior to the complaint can be obtained, for example, raw XDR data from several calendar days prior to the complaint. This raw XDR data is recorded in real-time or near real-time by Deep Packet Inspection (DPI) probes deployed in the core network and Gi-LAN. The raw XDR data can be stored at the minute or second level, used to record basic communication data of the user's underlying service activity status, providing the most basic massive data basis for problem localization. Each record includes at least the timestamp (time), cell identifier (cell_id), cell name (cell_name), terminal model, service type (business_type), and performance indicators. Among them, the service type can include video (VIDEO), web browsing (WEB), etc. Performance metrics may include Real-time Transport Protocol (RTP) packet loss, Mean Opinion Score (MOS) value, uplink traffic (KB), TCP connection establishment confirmation latency (ms), and downlink average round-trip time (RTT) (ms).

[0036] The raw XDR data undergoes preliminary filtering through a data cleaning mechanism. This preliminary filtering can include missing value imputation, outlier removal, and standardization. Standardization operations may include standardizing the time format and normalizing fields.

[0037] After initial filtering, basic rules are used to mark abnormal records, thus identifying preliminary anomalies. Here, preliminary anomalies refer to records identified as potentially having quality issues after verification using basic rules.

[0038] This step, by directly acquiring the raw XDR data from the underlying layer and performing initial screening, can effectively remove invalid and redundant information from the massive network data, while preserving refined second-level time-series characteristics and significantly reducing the data processing load of subsequent analysis processes.

[0039] Then, step S2 is executed to group and statistically analyze the preliminary abnormal events, resulting in a structured abnormal event list. Here, grouping and statistical analysis refers to the initial processing operation of classifying and aggregating discrete preliminary abnormal events according to specific dimensions. This maps the scattered preliminary abnormal events to corresponding physical or business dimensions, facilitating subsequent logical reasoning. The structured abnormal event list is a standardized data set formed after grouping and statistical analysis, used as a standardized input source for subsequent large language models to read and perform chain-like reasoning. The structured abnormal event list can include one or more structured abnormal events. Each structured abnormal event includes not only data such as event timestamp (time), cell identifier (cell_id), cell name (cell_name), terminal model, business type (business_type), and performance indicators, but also event type. This event type can include video stuttering and small packet latency anomalies, etc.

[0040] This step, through grouped statistics, enables the structured transformation of initially screened abnormal events, reduces the scale of redundant data, and provides a high-quality, well-structured data foundation for subsequent advanced reasoning using large language models.

[0041] Next, step S3 is executed, converting the structured list of abnormal events into parseable prompts that the Large Language Model (LLM) can parsed. These prompts are then input into the LLM. The LLM, by invoking various tools in its toolkit, aggregates the occurrence times of each structured abnormal event in the list, obtaining a set of candidate problem time periods. The LLM is an intelligent kernel with natural language understanding, logical chain reasoning, and tool invocation decision-making capabilities. Its role is to overcome the limitations of traditional fixed rules and dynamically decide on the aggregation and merging of event time periods.

[0042] Aggregation refers to the action of merging scattered structured anomaly events in a structured anomaly event list into a continuous problem interval on a time axis based on time, space, and business relevance. Its purpose is to reconstruct the actual occurrence cycle of network problems.

[0043] The large language model, through its inherent chain-like reasoning mechanism, comprehensively analyzes the occurrence time, business type, and spatial distribution of various structured anomaly events. During the reasoning process, it dynamically determines whether structured anomaly events with similar occurrence times, related business types, or spatial contiguousness need to be merged. For structured anomaly events that meet the merging criteria, an aggregation operation is performed on the occurrence time periods, merging fragmented, second-level anomaly events into continuous time periods with practical business significance. After one or more rounds of dynamic aggregation decision-making, a reasonable set of candidate problem time periods is output.

[0044] The merging criteria can be determined based on the time interval of each structured anomaly, the relevance of the service type, and the spatial continuity. For example, it can be expressed as at least one of the following: the time interval is less than a first threshold, the relevance of the service type is greater than a second threshold, and the distance between the corresponding cells is less than a spatial distance threshold. The candidate problem time period set can include one or more candidate problem time periods.

[0045] This step, by introducing a large language model for time-segment aggregation, can overcome the shortcomings of traditional fixed-window aggregation, which is prone to mis-aggregation or omission, and achieve adaptive and highly accurate time-segment extraction for different complaint scenarios with second-level precision.

[0046] Finally, step S4 is executed to score each candidate issue time period in the candidate issue time period set. For each candidate issue time period, multi-dimensional evaluation factors can be introduced for comprehensive scoring. Then, the candidate issue time periods are sorted from highest to lowest score, and the top specified number of high-scoring candidate issue time periods are selected as the target issue time periods for user complaints. The specified number can be set as needed, for example, three. If the candidate issue time period set contains fewer than three candidate issue time periods, all of them can be used as target issue time periods.

[0047] This step uses a scoring mechanism to scientifically quantify the actual impact of different abnormal time periods on user experience, ensuring that the final selected target problem time periods are highly consistent with the actual complaint scenarios, and greatly improving the accuracy of problem investigation priority.

[0048] In addition to determining the target problem time period, it can also determine the main cell and neighboring cell information involved in the target problem time period and generate a complaint analysis report.

[0049] The problem localization method for wireless network complaint scenarios provided in this invention adopts a bottom-up, layered, progressive architecture. By processing massive amounts of raw extended detail record data layer by layer, the data scale is reduced, computational complexity and subsequent computational pressure are decreased, significantly reducing resource consumption and processing resource requirements, and improving overall operational efficiency and response speed to user complaints. Especially in the massive data scenarios of provincial networks, this method is easier to implement and can efficiently support larger-scale real-time or near-real-time complaint backtracking needs. Furthermore, relying on the dynamic logical reasoning capabilities of a large language model, this method adaptively merges the occurrence periods of various structured abnormal events. This not only significantly improves the localization accuracy of the target problem period and enhances the precise problem localization capability for user complaints, enabling automated and accurate backtracking and localization for user complaint scenarios, ensuring that abnormal events are not missed or falsely reported, but also reduces human intervention, further improving problem localization efficiency and response speed to user complaints.

[0050] Based on the above embodiments, the aggregation of the occurrence time periods of each structured abnormal event in the structured abnormal event list based on the large language model to obtain a candidate problem time period set includes: Based on the large language model, the time window aggregation tool and the event time period merging tool are invoked to merge the time periods of each of the structured abnormal events by cell to obtain each initial merged time period; For each of the initial merging time periods that meet the merging conditions and correspond to different cells, the geographic location calculation tool is invoked based on the large language model to perform spatial continuity verification on the cells corresponding to the candidate merging time periods. If the spatial continuity verification passes, the event time period merging tool is invoked to merge the candidate merging time periods to obtain the candidate problem time period set.

[0051] Specifically, when using the large language model to aggregate the occurrence time periods of each structured abnormal event in the list of structured abnormal events, the large language model can be used to call the time window aggregation tool and the event time period merging tool to merge the time periods of each structured abnormal event by cell to obtain each initial merged time period.

[0052] The time window aggregation tool is a programmatic functional module used to group structured abnormal events corresponding to the same cell into time periods according to a set time interval threshold. Its function is to achieve preliminary coarse aggregation of abnormal events on the time axis, reducing the data complexity of a single processing operation by the large language model. First, the large language model is used to call the time window aggregation tool to aggregate each structured abnormal event by cell, obtaining the initial aggregation time periods.

[0053] For all structured abnormal events occurring within the same time window and corresponding to the same cell, their occurrence time periods are integrated into a continuous time period as an initial aggregation period, and the number of abnormal events within this initial aggregation period is counted. The length of the time window can be set as needed and is not specifically limited here. For example, the list of structured abnormal events may include 5 structured abnormal events occurring between 10:00:01 and 10:00:15 on a certain day, involving cells A and B. Cell A's cell ID is 123456789, and cell B's cell ID is 987654321. The service types include VIDEO and WEB, and the event types include video stuttering and small packet latency anomalies.

[0054] Assuming a time window length of 10 seconds, the time window aggregation tool can output three initial aggregation periods, which are: 1) The initial aggregation period 1 is 10:00:01-10:00:05, the corresponding cell ID is 123456789, and the number of abnormal events is 3; 2) The initial aggregation period 2 is 10:00:10-10:00:10, the corresponding cell ID is 987654321, and the number of abnormal events is 1; 3) The initial aggregation period 3 is 10:00:15-10:00:15, the corresponding cell ID is 987654321, and the number of abnormal events is 1.

[0055] Since initial aggregation periods 2 and 3 correspond to the same cell, are 5 seconds apart, and have a high correlation between service type and time type, they can be further merged by calling the event period merging tool. The event period merging tool is a programmatic functional module used to perform the initial aggregation period fusion operation for abnormal events. Its function is to continue integrating two or more initial aggregation periods that meet the merging conditions after the large language model makes a time window aggregation decision. In other words, initial aggregation periods 2 and 3 are candidate aggregation periods that meet the merging conditions and correspond to the same cell. Their occurrence periods can be integrated into a continuous period as an initial merging period, while simultaneously counting the number of abnormal events within this initial merging period.

[0056] The event time period merging tool can output two initial aggregated time periods, namely: 1) The initial merging period 1 is 10:00:01-10:00:05, the corresponding cell ID is 123456789, and the number of abnormal events is 3; 2) The initial merging period 2 is 10:00:10-10:00:15, the corresponding cell ID is 987654321, and the number of abnormal events is 2.

[0057] This step, through the use of time window aggregation tools and event time period merging tools, enables the large language model to perform coarse screening and merging of time periods within the same cell at the underlying level. This aggregates the occurrence time periods of massive discrete structured abnormal events into continuous time periods, which can significantly reduce the computational dimensionality of subsequent complex reasoning.

[0058] For candidate merging periods that meet the merging conditions within each initial merging period, the large language model can be used to call the geolocation calculation tool to perform spatial continuity verification on the cells corresponding to the candidate merging periods. Here, the geolocation calculation tool is a programmatic functional module that calculates the distance between different cells. Its role is to provide the large language model with accurate spatial distance data to help it determine whether the user has made a reasonable physical movement.

[0059] Spatial continuity verification is the process of verifying whether the distance between cells corresponding to the candidate merging time period meets the preset constraints. Its purpose is to prevent the erroneous aggregation of abnormal events that do not belong to the same user complaint scenario due to excessive spatial span, thereby eliminating the risk of misjudgment in mobile user scenarios.

[0060] If the spatial continuity check passes, it means that the user has reasonable spatial mobility continuity during the period of the anomaly. The large language model then calls the event time period merging tool to merge the candidate merging time periods corresponding to different cells to obtain a set of candidate problem time periods.

[0061] A candidate problem time period in the candidate problem time period set can be represented as: The time period for the candidate questions is 10:00:01-10:00:15; The affected residential communities are: Community A and Community B. Number of abnormal events: 5; Business type: VIDEO, WEB; Event type: Lag, small packet delay greater than 5s.

[0062] If the spatial continuity check fails, the candidate merging periods will not be merged.

[0063] In this embodiment of the invention, by introducing spatial continuity verification, the merging of time periods across cells must meet strict spatial continuity, which can effectively filter out erroneous aggregation caused by spatial jumps and greatly improve the authenticity and accuracy of candidate problem time periods in mobile user scenarios.

[0064] Based on the above embodiments, the step of using the large language model to call the time window aggregation tool and the event time period merging tool to merge the structured abnormal events by cell to obtain each initial merged time period includes: Based on the large language model, the time window aggregation tool is invoked to aggregate the structured abnormal events by cell into time windows, thereby obtaining each initial aggregation period. For each of the initial aggregation time periods that meet the merging conditions and correspond to the same cell, based on the large language model, the anomaly density analysis tool is invoked to perform anomaly event density verification on the candidate aggregation time periods. If the anomaly event density verification passes, the event time period merging tool is invoked to merge the candidate aggregation time periods of the same cell to obtain each of the initial merging time periods.

[0065] Specifically, before invoking the event time period merging tool, the large language model can first invoke the anomaly density analysis tool to verify the anomaly event density of the candidate aggregation time periods. This anomaly density analysis tool is a programmed functional module that calculates the anomaly event density within a specific time period, and its function is to quantitatively assess the density of network problems within that specific time period. The anomaly event density verification process determines whether the anomaly event density of the merged time period is within a preset density range; its function is to prevent the forced merging of irrelevant time periods with excessively large time spans but extremely sparse events.

[0066] Here, the abnormal event density can be determined by the ratio of the total length of the candidate merging period to the number of abnormal events in the candidate merging period. For example, if the total length of the candidate merging period is 14 seconds and the number of abnormal events in the candidate merging period is 5, the abnormal event density is approximately 0.357 events / second. The preset density range can be set as needed, for example, it can be set to 0.2-0.4 events / second, or it can be set to other ranges; no specific limitation is made here.

[0067] If the density of abnormal events is within the preset density range, and the abnormal event density verification is passed, the event time period merging tool is invoked to merge the candidate aggregation time periods in the same cell to obtain each initial merged time period.

[0068] If the density of abnormal events is not within the preset density range, and the abnormal event density verification fails, then the candidate aggregation time periods in the same cell will not be merged.

[0069] In this embodiment of the invention, by introducing anomaly event density verification, density rationality constraints are imposed on the merging of candidate aggregation periods that meet the merging conditions in the same cell. This avoids forcibly piecing together occasional, sporadic and long-interval anomalies, ensuring that the initial merging period after merging can truly reflect the concentrated period of poor network quality, thereby ensuring high fidelity in locating the target problem period.

[0070] Based on the above embodiments, the step of calling a geolocation calculation tool based on the large language model to perform spatial continuity verification on the cells corresponding to the candidate merging time periods includes: Based on the large language model, the geographic location calculation tool is invoked to calculate the distance between the cells corresponding to the candidate merging time periods; If the distance is less than the spatial distance threshold, then the spatial continuity check is deemed to have passed. If the distance is greater than or equal to the spatial distance threshold, then the spatial continuity check is determined to fail.

[0071] Specifically, during spatial continuity verification, a large language model can be used to call geographic location calculation tools to obtain the latitude and longitude or physical coordinates of the cells corresponding to the candidate merging time periods, and then calculate the distance between these cells. This distance can be a straight-line distance or an actual topological distance.

[0072] Next, this distance is compared with a spatial distance threshold. The spatial distance threshold is a physical distance limit used to determine whether a user's movement between different cells has spatial continuity. Its function is to provide a quantified spatial scale to eliminate abnormal cross-cell events that could not possibly have actually occurred by the same user within a short period. The spatial distance threshold can be set as needed, for example, it can be set to 300m.

[0073] If the distance between the cells corresponding to the candidate merging time periods is less than the spatial distance threshold, it means that the user's spatial movement range within the two candidate merging time periods is within a reasonable range, and the large language model determines that the spatial continuity verification has passed.

[0074] If the distance between cells corresponding to the candidate merging time period is greater than or equal to the spatial distance threshold, it indicates that the spatial continuity check has failed.

[0075] Conversely, if the calculated distance is greater than or equal to the spatial distance threshold, it means that the location span of the cells corresponding to the two candidate merging time periods is too large, exceeding the movement limit of a single user in a short period of time. In this case, the large language model determines that the spatial continuity verification fails.

[0076] In this embodiment of the invention, by setting specific spatial distance thresholds and performing precise comparisons, a rigid geographical constraint rule is provided for merging event time periods across cells. This can effectively filter out false clustering of events caused by pseudo-movement or location drift, ensuring the rationality of time period positioning in the spatial dimension.

[0077] Based on the above embodiments, the step of calling a geolocation calculation tool based on the large language model to perform spatial continuity verification on the cells corresponding to the candidate merging time periods includes: Based on the large language model, the anomaly density analysis tool is invoked to verify the anomaly event density of the candidate merging time periods. If both the abnormal event density check and the spatial continuity check pass, the event time period merging tool is invoked to merge the candidate merging time periods to obtain the candidate problem time period set.

[0078] Specifically, after performing spatial continuity verification on the cells corresponding to the candidate merging time periods, the large language model can be used to call the anomaly density analysis tool to perform anomaly event density verification on the candidate merging time periods, that is, to determine whether the anomaly event density of the merged time period is within the preset density range.

[0079] When both the abnormal event density verification and the spatial continuity verification pass, the large language model will determine that the candidate merging time period corresponds to the same poor network quality scenario, and then call the event time period merging tool to merge the candidate merging time periods to obtain a set of candidate problem time periods.

[0080] In this embodiment of the invention, a dual check mechanism of spatial continuity verification and abnormal event density verification is adopted, so that the merging decision of cross-regional time periods not only considers the proximity of physical distance, but also takes into account the distribution density of abnormal events on the time axis, which greatly reduces the false clustering rate caused by coincidence factors.

[0081] Based on the above embodiments, the step of calling the anomaly density analysis tool based on the large language model to verify the anomaly event density of the candidate merging time periods includes: Based on the large language model, the anomaly density analysis tool is invoked to calculate the total length of the candidate merging time periods, and the anomaly event density is calculated based on the total length of the time periods and the number of anomalies in the candidate merging time periods. If the density of abnormal events is within a preset density range, then the abnormal event density verification is deemed to have passed. If the density of abnormal events is outside the preset density range, then the abnormal event density verification is determined to be unsuccessful.

[0082] In this embodiment of the invention, by accurately calculating the density of abnormal events and comparing it with threshold intervals, the sparsity of fuzzy events is quantified, providing reliable digital support for the merging decision of large language models and improving the rigor of defining problem time periods.

[0083] Based on the above embodiments, the step of invoking the event time period merging tool to merge the candidate merging time periods to obtain the candidate problem time period set further includes: For any candidate problem time period in the set of candidate problem time periods, a final fallback check is performed on the duration of any candidate problem time period, the distance between the corresponding cells, and the density of abnormal events.

[0084] Specifically, after merging the candidate merging time periods, a large language model can be used to perform a final fallback check on each candidate problem time period in the candidate problem time period set. The final fallback check includes determining whether the duration of each candidate problem time period is less than a preset duration, whether the distance between the cells involved in each candidate problem time period is less than a spatial distance threshold, and whether the abnormal event density of each candidate problem time period is within a preset density range.

[0085] For any candidate problem time period in the candidate problem time period set, if all the final fallback checks pass (i.e., the duration of any candidate problem time period is less than a preset duration, the distance between cells involved in any candidate problem time period is less than a spatial distance threshold, and the abnormal event density of any candidate problem time period is within a preset density range), then any candidate problem time period is considered reasonable and does not require further merging; further processing of any candidate problem time period ceases. If any final check fails, then any candidate problem time period is processed according to its generation process until all the final fallback checks pass.

[0086] A candidate problem time period after the final fallback check passes can be represented as: { "start_time": "2025-06-24 10:00:01", "end_time": "2025-06-24 10:00:15", "cells": [ {"cell_id": "123456789", "cell_name": "Beijing xx Community"}, {"cell_id": "987654321", "cell_name": "Beijing YY Community"} ], "event_count": 5, "business_types": ["VIDEO", "WEB"], "event_types": ["Lag", "Small packet latency greater than 5 seconds"] }

[0087] In this embodiment of the invention, a fault-tolerance and anti-bias mechanism is established by performing a full-element final check on the generated candidate problem time periods. This ensures that the final output candidate problem time periods still strictly conform to the time, space and density logic of network quality difference analysis, completely eliminates excessive aggregation caused by algorithm divergence, and ensures that the complaint location results delivered to the operation and maintenance personnel are highly precise and credible.

[0088] Based on the above embodiments, the scoring of each candidate question time period in the candidate question time period set includes: The score for each candidate problem period is calculated based on the number of abnormal events, the length of the period, and the influence weight of the business type.

[0089] Specifically, when scoring each candidate issue time period in the candidate issue time period set, multi-dimensional scoring can be used. Scoring dimensions can include the number of abnormal events in the candidate issue time period, the time period length, and the influence weight of the business type. The score for each candidate issue time period can be obtained by weighted summing of these factors. The weights used for weighted summation can be set as needed and are not specifically limited here.

[0090] The number of abnormal events directly reflects the density of network anomalies within a given time period; a higher number usually indicates a more severe anomaly. The duration of the period measures the persistence of the abnormal event, helping to determine whether it's a momentary fluctuation or a persistent network failure. The service type impact weight differentiates the sensitivity of different services to network quality fluctuations and their impact on user experience; for example, the impact weight for video services is typically higher than that for web browsing.

[0091] In this embodiment of the invention, the scoring is carried out by comprehensively considering the number of abnormal events, the length of the time period, and the impact weight of the business type. This abandons the single-dimensional evaluation method and makes the quantification of the severity of the problem period more scientific and comprehensive, which can accurately reflect the actual negative impact of different time periods on user experience.

[0092] Based on the above embodiments, the influence weight of the business type is obtained by dynamically adjusting the target business type of the user complaint using the large language model.

[0093] Specifically, the target business type is defined as the specific business type that is affected, either explicitly stated in the user's complaint or inferred from the content of the complaint. The large language model dynamically adjusts the influence weight of the business type based on the target business type. For example, when the target business type is video service, the large language model will automatically increase the influence weight of the business type; when the target business type is web browsing, the large language model will automatically decrease the influence weight of the business type.

[0094] In this embodiment of the invention, a large language model is used to dynamically adjust the influence weight of the business type according to the target business type, thereby realizing the scenario adaptability of the scoring mechanism. This makes the positioning results highly consistent with the user's actual pain points, significantly enhancing the pertinence and effectiveness of anomaly positioning.

[0095] Based on the above embodiments, the preliminary abnormal event grouping and statistical analysis to obtain a structured abnormal event list includes: From the initial abnormal events, select event records with clear abnormal markers; Using a grouping statistics tool, the event records are grouped and statistically analyzed according to service type and cell identifier to generate the structured abnormal event list.

[0096] Specifically, when grouping and statistically analyzing preliminary anomalies, event records with clear anomaly markers can be selected from the preliminary anomalies. This means filtering out records that are merely marginal fluctuations or false alarms, and retaining the truly problematic event records. Clear anomaly markers are assigned during the preliminary anomaly identification stage to indicate that the event record indeed exhibits network performance indicators exceeding limits or a fault state. Their function is to serve as a hard standard for selecting valid anomalies.

[0097] Using grouping and statistical tools, and taking business type and cell identifier as the two dimensions of classification, event records are grouped and statistically analyzed according to business type and cell identifier, generating a well-organized and clearly categorized list of abnormal events.

[0098] In this embodiment of the invention, by filtering based on explicit anomaly markers and grouping statistics based on business type and spatial dimension, noise in the underlying data can be effectively removed, and scattered data can be transformed into high-dimensional structured information, greatly reducing the computational load of subsequent complex models and providing a clear framework for accurate source tracing.

[0099] Based on the above embodiments, such as Figure 3 As shown in the embodiment of the present invention, the problem localization method for wireless network complaint scenarios can first input a structured list of abnormal events into the large language model when using the large language model to call tools. The large language model generates a task flow through a chain-like reasoning mechanism and calls tools in the tool library according to the task flow to iteratively execute tasks. When all tasks in the task flow are completed, a set of candidate problem time periods is obtained and output through the large language model.

[0100] like Figure 4 As shown, based on the above embodiments, this embodiment of the invention provides a problem location device for wireless network complaint scenarios, including: The raw data processing module 41 is used to acquire the raw extended details record data of the wireless network before the user complaint, and to determine the preliminary abnormal event based on the raw extended details record data; Event extraction module 41 is used to group and statistically analyze the preliminary abnormal events to obtain a structured abnormal event list; The time period aggregation module 43 is used to aggregate the occurrence time periods of each structured abnormal event in the structured abnormal event list based on the large language model to obtain a candidate problem time period set. The problem location module 44 is used to score each candidate problem time period in the candidate problem time period set, and determine the target problem time period of the user complaint based on the score of each candidate problem time period.

[0101] Specifically, the functions of each module in the problem location device for wireless network complaint scenarios provided in this embodiment of the invention correspond one-to-one with the operation flow of each step in the above method-like embodiments, and the achieved effects are also the same. For details, please refer to the above embodiments, and this will not be repeated in this embodiment of the invention.

[0102] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5 As shown, the electronic device may include a processor 810, a communications interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute the problem localization method for wireless network complaint scenarios provided in the above embodiments.

[0103] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to related technologies, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0104] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the problem localization method for wireless network complaint scenarios provided in the above embodiments.

[0105] In another aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the problem localization method for wireless network complaint scenarios provided in the above embodiments. This computer-readable storage medium can be either a non-transitory computer-readable storage medium or a transient computer-readable storage medium, and is not specifically limited here.

[0106] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0107] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of software products. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for locating problems in wireless network complaint scenarios, characterized in that, include: Obtain the original extended details record data of the wireless network prior to the user's complaint, and determine the preliminary abnormal event based on the original extended details record data; The preliminary abnormal events are grouped and statistically analyzed to obtain a structured list of abnormal events; Based on the large language model, the occurrence time periods of each structured abnormal event in the structured abnormal event list are aggregated to obtain a candidate problem time period set; Each candidate issue time period in the candidate issue time period set is scored, and the target issue time period for the user complaint is determined based on the score of each candidate issue time period.

2. The problem localization method for wireless network complaint scenarios according to claim 1, characterized in that, The method based on the large language model aggregates the occurrence time periods of each structured abnormal event in the structured abnormal event list to obtain a candidate problem time period set, including: Based on the large language model, the time window aggregation tool and the event time period merging tool are invoked to merge the time periods of each of the structured abnormal events by cell to obtain each initial merged time period; For each of the initial merging time periods that meet the merging conditions and correspond to different cells, the geographic location calculation tool is invoked based on the large language model to perform spatial continuity verification on the cells corresponding to the candidate merging time periods. If the spatial continuity verification passes, the event time period merging tool is invoked to merge the candidate merging time periods to obtain the candidate problem time period set.

3. The problem localization method for wireless network complaint scenarios according to claim 2, characterized in that, Based on the large language model, the time window aggregation tool and the event time period merging tool are invoked to merge the time periods of each structured abnormal event by cell, resulting in each initial merged time period, including: Based on the large language model, the time window aggregation tool is invoked to aggregate the structured abnormal events by cell into time windows, thereby obtaining each initial aggregation period. For each of the initial aggregation time periods that meet the merging conditions and correspond to the same cell, based on the large language model, the anomaly density analysis tool is invoked to perform anomaly event density verification on the candidate aggregation time periods. If the anomaly event density verification passes, the event time period merging tool is invoked to merge the candidate aggregation time periods of the same cell to obtain each of the initial merging time periods.

4. The problem localization method for wireless network complaint scenarios according to claim 2, characterized in that, The step of calling a geolocation calculation tool based on the large language model to perform spatial continuity verification on the cells corresponding to the candidate merging time periods includes: Based on the large language model, the geographic location calculation tool is invoked to calculate the distance between the cells corresponding to the candidate merging time periods; If the distance is less than the spatial distance threshold, then the spatial continuity check is deemed to have passed. If the distance is greater than or equal to the spatial distance threshold, then the spatial continuity check is determined to fail.

5. The problem localization method for wireless network complaint scenarios according to claim 2, characterized in that, Based on the large language model, a geolocation calculation tool is invoked to perform spatial continuity verification on the cells corresponding to the candidate merging time periods, followed by: Based on the large language model, an anomaly density analysis tool is invoked to verify the anomaly event density of the candidate merging time periods; If both the abnormal event density check and the spatial continuity check pass, the event time period merging tool is invoked to merge the candidate merging time periods to obtain the candidate problem time period set.

6. The problem localization method for wireless network complaint scenarios according to claim 5, characterized in that, The step of calling an anomaly density analysis tool based on the large language model to verify the anomaly event density of the candidate merging time periods includes: Based on the large language model, the anomaly density analysis tool is invoked to calculate the total length of the candidate merging time periods, and the anomaly event density is calculated based on the total length of the time periods and the number of anomalies in the candidate merging time periods. If the density of abnormal events is within a preset density range, then the abnormal event density verification is deemed to have passed. If the density of abnormal events is outside the preset density range, then the abnormal event density verification is determined to be unsuccessful.

7. The problem localization method for wireless network complaint scenarios according to claim 2, characterized in that, The process of invoking the event time period merging tool to merge the candidate merge time periods to obtain the candidate issue time period set further includes: For any candidate problem time period in the set of candidate problem time periods, a final fallback check is performed on the duration of any candidate problem time period, the distance between the corresponding cells, and the density of abnormal events.

8. The problem localization method for wireless network complaint scenarios according to any one of claims 1-7, characterized in that, The scoring of each candidate question time period in the candidate question time period set includes: The score for each candidate problem period is calculated based on the number of abnormal events, the length of the period, and the influence weight of the business type.

9. The problem localization method for wireless network complaint scenarios according to claim 8, characterized in that, The influence weight of the business type is obtained by dynamically adjusting the target business type of the user complaint based on the large language model.

10. The problem localization method for wireless network complaint scenarios according to any one of claims 1-7, characterized in that, The preliminary abnormal events are grouped and statistically analyzed to obtain a structured list of abnormal events, including: From the initial abnormal events, select event records with clear abnormal markers; Using a grouping statistics tool, the event records are grouped and statistically analyzed according to service type and cell identifier to generate the structured abnormal event list.

11. A problem location device for wireless network complaint scenarios, characterized in that, include: The raw data processing module is used to acquire the raw extended details record data of the wireless network before the user complaint, and to determine the preliminary abnormal event based on the raw extended details record data; The event extraction module is used to group and statistically analyze the preliminary abnormal events to obtain a structured list of abnormal events; The time period aggregation module is used to aggregate the occurrence time periods of each structured abnormal event in the structured abnormal event list based on the large language model to obtain a candidate problem time period set. The problem location module is used to score each candidate problem time period in the candidate problem time period set, and determine the target problem time period of the user complaint based on the score of each candidate problem time period.

12. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the problem localization method for wireless network complaint scenarios as described in any one of claims 1-10.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the problem localization method for wireless network complaint scenarios as described in any one of claims 1-10.

14. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the problem localization method for wireless network complaint scenarios as described in any one of claims 1-10.

Images