A short video network poor area identification and root cause positioning method and related equipment

Abstract

This invention relates to the field of communication network optimization technology, specifically disclosing a method and related equipment for identifying and locating the root causes of poor-quality areas in short video networks. The technical solution of this invention integrates short video service experience data collected from a top-level application platform, signaling-level user behavior data collected from a service experience quality platform, and key performance indicator data of the wireless network collected from an operation and maintenance center system. Based on timestamps and spatial identifiers, it generates multi-source heterogeneous fusion records through spatiotemporal alignment. Then, it identifies poor-quality hotspot areas through geographic grid aggregation and dynamic spatiotemporal clustering. Finally, it uses a lightweight classification model to automatically diagnose and prioritize these hotspot areas. This solves the problems of lagging identification of poor-quality scenes and reliance on manual experience for root cause location in existing technologies, which suffer from single data dimensions, coarse positioning granularity, and a lack of intelligent closed-loop mechanisms. This improves the accuracy of identifying poor-quality areas in short video networks and the automation level of root cause location.

Description

Technical Field

[0001] This invention relates to the field of communication network optimization technology, and in particular to a method and related equipment for identifying and locating the root causes of poor-quality areas in short video networks. Background Technology

[0002] With the explosive growth of mobile internet services, short video applications have become the main carriers of mobile network traffic, and network quality directly affects user experience and platform stickiness. Currently, mobile network optimization mainly relies on traditional Measurement Reports (MRs) and Key Performance Indicators (KPIs). Indicators such as Reference Signal Receiving Power (RSRP), Signal to Interference plus Noise Ratio (SINR), and Physical Resource Block (PRB) utilization are used to assess coverage, capacity, and interference. However, this type of data has inherent limitations, including coarse granularity, disconnect from actual user perception, and delayed passive response. Meanwhile, over-the-top (OTT) platforms can collect quality of experience (QoE) data such as video bitrate, initial buffering time, and number of stutters through client software development kits (SDKs), while service experience quality (SEQ) platforms can obtain signaling-level user behavior data such as International Mobile Subscriber Identity (IMSI), cell ID, terminal model, and service type. However, these data sources are isolated from each other and lack an effective correlation mechanism.

[0003] Currently, existing technologies suffer from three core shortcomings: First, they lack a single data dimension, failing to integrate OTT real QoE, SEQ signaling, and wireless KPI data, thus failing to comprehensively depict the relationship between users, services, and the network. Second, their positioning granularity is coarse, generally using residential communities as the smallest unit of analysis, unable to identify deeply poor quality scenarios such as within buildings or street segments, and often relying on static thresholds or manual complaint heatmaps, lacking dynamic spatiotemporal clustering capabilities based on latitude and longitude. Third, they lack an intelligent closed-loop mechanism; even when problems are identified, remediation suggestions still rely on expert experience, lacking an automated diagnostic process for feature extraction, root cause classification, and priority dispatch, making it difficult to support efficient operation and maintenance of large-scale networks.

[0004] Therefore, there is an urgent need to provide a technical solution to address the above problems. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method and related equipment for identifying and locating poor-quality areas in short video networks.

[0006] In a first aspect, the present invention provides a method for identifying and locating the root causes of poor-quality regions in short video networks. The technical solution of this method is as follows: Acquire short video service experience data collected by the top application platform, signaling-level user behavior data collected by the service experience quality platform, and key performance indicator data of wireless network collected by the operation and maintenance center system; Using timestamps and spatial identifiers as association keys, the short video service experience data, the signaling-level user behavior data, and the wireless network key performance indicator data of the same user within the same service session period are spatiotemporally aligned to generate multiple multi-source heterogeneous fusion records. Each multi-source heterogeneous fusion record corresponds to one user's service session period. The multi-source heterogeneous fusion records are aggregated according to geographic grids. The proportion of users with poor quality is determined based on the number of users in each geographic grid that meet the preset poor quality rules and the total number of users. Geographic grids whose proportion of users with poor quality meets the density threshold and whose spatial distance and temporal continuity between them meet the preset threshold are clustered to obtain poor quality hotspot areas. Based on the fusion records contained in the poor quality hotspot region, feature vectors for characterizing network problem types are extracted, and the feature vectors are input into a lightweight classification model. The root cause diagnosis results corresponding to the poor quality hotspot region are output according to a preset priority order.

[0007] The beneficial effects of the short video network poor quality region identification and root cause localization method of the present invention are as follows: The method of this invention integrates short video service experience data collected by the over-the-top application platform, signaling-level user behavior data collected by the service experience quality platform, and key performance indicator data of the wireless network collected by the operation and maintenance center system. Based on timestamps and spatial identifiers, it generates multi-source heterogeneous fusion records through spatiotemporal alignment. Then, it identifies poor-quality hotspot areas through geographic grid aggregation and dynamic spatiotemporal clustering. Finally, it uses a lightweight classification model to automatically diagnose the root causes and prioritize poor-quality hotspot areas. This solves the problems of lagging identification of poor-quality scenes and reliance on human experience for root cause localization caused by the single data dimension, coarse positioning granularity, and lack of intelligent closed-loop mechanism in the prior art. It improves the identification accuracy of poor-quality areas in short video networks and the degree of automation of root cause localization.

[0008] Based on the above scheme, the short video network poor quality area identification and root cause localization method of the present invention can be further improved as follows.

[0009] In one alternative approach, the steps for obtaining short video service experience data collected by the over-the-top application platform, signaling-level user behavior data collected by the service experience quality platform, and key performance indicator data of the wireless network collected by the operation and maintenance center system include: The over-the-top application platform receives the short video service experience data reported by the client software development kit. The short video service experience data includes the time when the user service occurs, video bitrate, initial buffering time, number of stutters, application type, indoor and outdoor markers, and user latitude and longitude. The service experience quality platform receives user behavior data obtained from signaling plane parsing. The signaling-level user behavior data includes International Mobile Subscriber Identity (IMSI), service start timestamp, service end timestamp, first cell identifier, and terminal model. The operation and maintenance center system receives the key performance indicator data of the wireless network output from the northbound interface of the network management system. The key performance indicator data of the wireless network includes the indicator collection time, the second cell identifier, and the average value of the reference signal received power at the cell level, the signal-to-interference-plus-noise ratio, the physical resource block utilization rate, and the call drop rate.

[0010] The advantages of adopting the above-mentioned optional methods are as follows: it further refines the collection mechanism of multi-source heterogeneous data, and by clarifying the specific data fields and interface sources of the top application platform, business experience quality platform and operation and maintenance center system, it solves the problems of ambiguous data sources and unclear field definitions, and improves the standardization and traceability of data collection.

[0011] In one optional approach, the step of spatiotemporally aligning the short video service experience data, the signaling-level user behavior data, and the wireless network key performance indicator data of the same user within the same service session period, using timestamps and spatial identifiers as association keys, to generate multiple multi-source heterogeneous fusion records includes: Extract the time of the user's service and the user's latitude and longitude from each piece of short video service experience data; Extract the service start timestamp, service end timestamp, first cell identifier, and International Mobile Subscriber Identity from each of the aforementioned signaling-level user behavior data; Extract the index collection time and the second cell identifier from each of the aforementioned key performance index data of the wireless network; For each piece of short video service experience data, the target cell corresponding to the short video service experience data is determined based on the user's latitude and longitude. The signaling-level user behavior data that has the same International Mobile Subscriber Identity (IMSI), the user service occurrence time is between the service start time stamp and the service end time stamp, and the first cell identifier is consistent with the target cell is searched in the signaling-level user behavior data and used as the matching signaling data; The KPI data that matches the second cell identifier of the target cell and whose time difference between the time of the indicator collection and the time of the user service occurrence is less than a preset time threshold is found in the wireless network key performance indicator data and is used as the matching KPI data. The short video service experience data, the matched signaling data, and the matched KPI data are merged into a single multi-source heterogeneous fusion record; Repeat the steps of determining the target cell corresponding to each short video service experience data based on the user's latitude and longitude until all short video service experience data is processed, resulting in multiple multi-source heterogeneous fusion records.

[0012] The advantages of adopting the above optional approach are: further defining the specific association rules and matching logic for spatiotemporal alignment, and solving the problems of inaccurate association of heterogeneous data sources and redundant fusion records through multiple constraints of user identifier, time window and cell identifier, thereby improving the accuracy of data fusion and the quality of records.

[0013] In one optional approach, the step of aggregating multiple multi-source heterogeneous fusion records according to geographic rasters, and determining the proportion of users with poor quality based on the number of users in each geographic raster that meet a preset quality defect rule and the total number of users, includes: The area to be analyzed is divided into multiple geographic grids, and each geographic grid corresponds to a spatial range of a preset size. For each of the aforementioned multi-source heterogeneous fusion records, the target geographic raster to which the multi-source heterogeneous fusion record belongs is determined based on the user's latitude and longitude; For each target geographic raster, obtain all multi-source heterogeneous fusion records belonging to the target geographic raster, and determine whether the multi-source heterogeneous fusion record meets the preset quality difference rule based on the short video service experience data in each multi-source heterogeneous fusion record. The total number of users is calculated as the number of different International Mobile Subscriber Identity (IMSI) codes appearing in all multi-source heterogeneous fusion records belonging to the target geographic raster, and the number of different IMSI codes appearing in multi-source heterogeneous fusion records belonging to the target geographic raster and satisfying the preset quality difference rule is calculated as the number of users with poor quality. The proportion of poor-quality users for each target geographic raster is determined based on the ratio of the number of poor-quality users in each target geographic raster to the total number of users.

[0014] The beneficial effects of adopting the above-mentioned optional methods are as follows: the process of geographic raster aggregation and calculation of the proportion of poor quality users is further refined. By clearly defining the raster division rules and the criteria for judging poor quality, the problems of inconsistent statistical standards for poor quality users and coarse spatial granularity are solved, and the precision of identifying poor quality areas is improved.

[0015] In one optional approach, the step of clustering geographical rasters where the proportion of users with poor quality meets a density threshold and the spatial distance and temporal continuity between them meet a preset threshold to obtain poor quality hotspot areas includes: Divide the time period to be analyzed into multiple consecutive time windows; For each time window, determine the proportion of poor-quality users for each target geographic raster within that time window, and mark all target geographic rasters whose proportion of poor-quality users is not less than the density threshold as candidate rasters. For each candidate raster, search for target candidate rasters belonging to the same time window or other time windows among spatially adjacent target geographic rasters, and determine whether the time window index difference between the target candidate raster and the current candidate raster is not greater than the time window interval threshold. All candidate rasters that meet the spatial adjacency condition and whose time window index difference is not greater than the time window interval threshold are grouped into the same candidate cluster. Candidate clusters containing a number of candidate grids not less than a preset cluster size threshold are identified as the poor-quality hotspot regions.

[0016] The beneficial effects of adopting the above-mentioned optional methods are as follows: the clustering determination method for poor-quality hotspot areas is further expanded. By introducing time window continuity constraints and spatial adjacency conditions, the problem of fragmentation of poor-quality areas and spatiotemporal feature separation caused by static thresholds is solved, and the continuity and integrity of hotspot area clustering are improved.

[0017] In one alternative approach, the steps of extracting feature vectors characterizing network problem types based on fusion records contained within the poor-quality hotspot regions, inputting the feature vectors into a lightweight classification model, and outputting root cause diagnosis results corresponding to the poor-quality hotspot regions according to a preset priority order include: For each of the poor-quality hotspot regions, obtain all multi-source heterogeneous fusion records belonging to that poor-quality hotspot region; Based on the key performance index data of the wireless network in each multi-source heterogeneous fusion record, the weak coverage ratio, the high load ratio and the interference intensity are calculated respectively, and the weak coverage ratio, the high load ratio and the interference intensity are combined into the feature vector. The feature vector is input into the pre-trained lightweight classification model, which has a first root cause judgment rule, a second root cause judgment rule and a third root cause judgment rule arranged in the preset priority order. The first root cause judgment rule, the second root cause judgment rule, and the third root cause judgment rule are executed sequentially through the lightweight classification model. When the proportion of weak coverage is greater than the first preset threshold, the first root cause judgment rule is triggered and the weak coverage root cause type is output. When the proportion of weak coverage is not greater than the first preset threshold and the proportion of high load is greater than the second preset threshold, the second root cause judgment rule is triggered and the high load root cause type is output. When the proportion of weak coverage is not greater than the first preset threshold, the proportion of high load is not greater than the second preset threshold, and the interference intensity is greater than the third preset threshold, the third root cause judgment rule is triggered and the interference root cause type is output. The root cause type output by the triggered root cause judgment rule is used as the root cause diagnosis result.

[0018] The beneficial effects of adopting the above-mentioned optional methods are as follows: it further standardizes the priority determination process of root cause diagnosis, and solves the problems of determination conflict and diagnosis logic confusion when multiple root causes coexist by using the step-by-step rule execution order built into the lightweight classification model, thereby improving the orderliness and accuracy of root cause localization.

[0019] In one alternative approach, it also includes: Obtain the root cause type from the root cause diagnosis results, wherein the root cause type is: the weak coverage root cause type, the high load root cause type, or the interference root cause type; Based on the preset mapping relationship between root cause types and remediation suggestions, remediation suggestions corresponding to the root cause types are determined; wherein, the weak coverage root cause type corresponds to remediation suggestions for co-construction and sharing or supplementing sites, the high load root cause type corresponds to remediation suggestions for capacity expansion or load balancing, and the interference root cause type corresponds to remediation suggestions for interference investigation or physical cell identifier optimization. The location information of the poor quality hotspot area, the root cause type, the remediation suggestion, and the preset priority identifier are combined to generate an electronic work order, and the electronic work order is pushed to the network optimization system.

[0020] The beneficial effects of adopting the above-mentioned optional methods are: further extending the application of diagnostic results, solving the problem of disconnect between diagnostic results and operation and maintenance actions through the mapping relationship between root cause type and remediation suggestions and the generation of electronic work orders, and improving the automation level and handling efficiency of network optimization response.

[0021] Secondly, the present invention provides a device for identifying and locating the root causes of poor-quality areas in short video networks. The technical solution of this device is as follows: The data acquisition module is used to acquire short video service experience data collected by the top application platform, signaling-level user behavior data collected by the service experience quality platform, and key performance indicator data of wireless network collected by the operation and maintenance center system. The alignment and fusion module is used to perform spatiotemporal alignment of the short video service experience data, the signaling-level user behavior data and the wireless network key performance indicator data of the same user within the same service session period using timestamps and spatial identifiers as association keys, and to generate multiple multi-source heterogeneous fusion records, each of which corresponds to one service session period of a user. The region identification module is used to aggregate the multi-source heterogeneous fusion records according to geographic grids, determine the proportion of poor quality users based on the number of users in each geographic grid that meet the preset poor quality rules and the total number of users, and cluster the geographic grids whose proportion of poor quality users meets the density threshold and whose spatial distance and temporal continuity between them meet the preset threshold to obtain poor quality hotspot areas. The root cause localization module is used to extract feature vectors that characterize the network problem type based on the fusion records contained in the poor quality hotspot area, and input the feature vectors into a lightweight classification model to output the root cause diagnosis results corresponding to the poor quality hotspot area according to a preset priority order.

[0022] The beneficial effects of the short video network poor quality region identification and root cause localization device of the present invention are as follows: The device of this invention integrates short video service experience data collected from the over-the-top application platform, signaling-level user behavior data collected from the service experience quality platform, and key performance indicator data of the wireless network collected from the operation and maintenance center system. Based on timestamps and spatial identifiers, it generates multi-source heterogeneous fusion records through spatiotemporal alignment. Then, it identifies poor-quality hotspot areas through geographic grid aggregation and dynamic spatiotemporal clustering. Finally, it uses a lightweight classification model to automatically diagnose the root causes and prioritize these poor-quality hotspot areas. This solves the problems of lagging identification of poor-quality scenes and reliance on human experience for root cause localization caused by the single data dimension, coarse positioning granularity, and lack of intelligent closed-loop mechanism in the prior art. It improves the identification accuracy of poor-quality areas in short video networks and the degree of automation in root cause localization.

[0023] Thirdly, the technical solution of an electronic device according to the present invention is as follows: It includes a memory, a processor, and a program stored in the memory and running on the processor, wherein the processor executes the program to implement the steps of the short video network poor quality region identification and root cause localization method of the present invention.

[0024] Fourthly, the technical solution of a computer-readable storage medium provided by the present invention is as follows: The computer-readable storage medium stores instructions that, when read, cause the computer-readable storage medium to perform the steps of the short video network poor quality area identification and root cause localization method of the present invention.

[0025] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0026] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a flowchart illustrating an embodiment of a method for identifying and locating poor-quality areas in short video networks according to the present invention. Figure 2 This is an overall architecture diagram of a short video network poor quality region identification and root cause localization method according to the present invention; Figure 3 This is a schematic diagram of a root cause intelligent diagnostic decision tree; Figure 4 This is a schematic diagram of an embodiment of a short video network poor quality region identification and root cause localization device according to the present invention; Figure 5 This is a schematic diagram of an embodiment of an electronic device according to the present invention. Detailed Implementation

[0027] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein.

[0028] Figure 1This diagram illustrates a flowchart of an embodiment of a method for identifying and locating poor-quality areas in short video networks according to the present invention. This method can be executed by electronic devices such as terminal devices or servers. The terminal device can be any fixed or mobile terminal, such as user equipment (UE), mobile device, user terminal, terminal, cellular phone, cordless phone, personal digital assistant (PDA), handheld device, computing device, in-vehicle device, or wearable device. The server can be a single server or a server cluster consisting of multiple servers. Any electronic device can implement the method for identifying and locating poor-quality areas in short video networks by having its processor call computer-readable instructions stored in its memory. Figure 1 As shown, it includes the following steps: S1. Obtain short video service experience data collected by the top application platform, signaling-level user behavior data collected by the service experience quality platform, and key performance indicator data of wireless network collected by the operation and maintenance center system.

[0029] Among them, the over-the-top application platform refers to a system platform deployed on the application service provider's side that collects user experience data during the use of specific applications through client software development kits. For example, the short video application platform D is an over-the-top application platform. Platform D collects user experience data such as video bitrate and buffering time when watching short videos through the D client software development kit integrated into the user's mobile phone. Short video business experience data refers to the set of data reflecting business quality reported by the client during the use of short video applications. For example, a record reported by the D client may contain data such as video bitrate of 360P, initial buffering time of 2.5 seconds, number of stutters of 3, application type of D, indoor / outdoor identification as outdoor, and user latitude and longitude of (119.3°E, 26.1°N). This data constitutes short video business experience data.

[0030] The service experience quality platform refers to a system deployed on the operator's core network that acquires user service signaling data through deep packet inspection or signaling collection methods. For example, operator S's service experience quality platform can collect signaling records when user C watches short videos on operator D. Signaling-level user behavior data refers to information related to user service sessions, such as user identification, time, location, and device, obtained by parsing network signaling. For example, a signaling record collected by the service experience quality platform may contain the International Mobile Subscriber Identity (IMSI) 460011234567890, service start timestamp January 1, 2026, 20:15:30, service end timestamp January 1, 2026, 20:16:30, first cell identifier CELL_A001, and terminal model PHONE_X. This information constitutes signaling-level user behavior data.

[0031] The operation and maintenance center system refers to the comprehensive network management system used by telecommunications operators to manage and monitor wireless networks, responsible for collecting performance statistics at the base station and cell levels. For example, operator S's operation and maintenance center system can periodically obtain wireless indicator data for cell CELL_A001 from the base station. Key performance indicator data for the wireless network refers to quantitative indicators obtained from the operation and maintenance center system that reflect the coverage, capacity, interference, and other conditions of the wireless network. For example, an indicator record for cell CELL_A001 output by the operation and maintenance center system includes the indicator collection time January 1, 2026, 20:15:00, second cell identifier CELL_A001, average reference signal received power of -112dBm, signal-to-interference-plus-noise ratio of 5dB, physical resource block utilization rate of 78%, and call drop rate of 0.1%. These values ​​are key performance indicator data for the wireless network.

[0032] S2. Using timestamps and spatial identifiers as association keys, the short video service experience data, the signaling-level user behavior data, and the wireless network key performance indicator data of the same user within the same service session period are spatiotemporally aligned to generate multiple multi-source heterogeneous fusion records. Each multi-source heterogeneous fusion record corresponds to one service session period of a user.

[0033] In this context, a timestamp refers to a marker used to identify the precise moment an event occurred, typically accurate to the second or millisecond. For example, the user's service occurrence time in short video service experience data, January 1, 2026, 20:15:30, is a timestamp. A spatial identifier refers to a marker used to locate the user or event's location, including latitude and longitude coordinates or cell identifiers. For example, the user's latitude and longitude (119.3°E, 26.1°N) in short video service experience data is a spatial identifier, and the first cell identifier CELL_A001 in signaling-level user behavior data is also a spatial identifier. A correlation key refers to a combination of common fields used to link records from different data sources. For example, using a timestamp and spatial identifier as correlation keys, three sources of data for the same user within the same service session period can be matched. For instance, the user's service occurrence time, January 1, 2026, 20:15:30, and the user's latitude and longitude (119.3°E, 26.1°N) can be used as correlation keys to match signaling data and wireless network key performance indicator data.

[0034] The business session cycle refers to the continuous time period from when a user starts using a service to when they stop using it. For example, if user C starts playing a short video D from 20:15:30 on January 1, 2026 to 20:16:30 on the same day, this 1-minute time period is one business session cycle.

[0035] Among them, a multi-source heterogeneous fusion record refers to a structured data source generated by merging short video service experience data, signaling-level user behavior data, and wireless network key performance indicator data of the same user within the same service session period through spatiotemporal alignment. For example, a record generated by merging user C's D experience data, service experience quality platform signaling data, and operation and maintenance center system cell data includes fields such as the user service occurrence time January 1, 2026, 20:15:30, user latitude and longitude (119.3°E, 26.1°N), video bitrate 360P, International Mobile Subscriber Identity 460011234567890, first cell identifier CELL_A001, and average reference signal received power -112dBm. This record is a multi-source heterogeneous fusion record.

[0036] S3. Aggregate the multi-source heterogeneous fusion records according to geographic grids. Determine the proportion of users with poor quality based on the number of users in each geographic grid that meet the preset poor quality rules and the total number of users. Cluster the geographic grids whose proportion of users with poor quality meets the density threshold and whose spatial distance and temporal continuity between them meet the preset threshold to obtain poor quality hotspot areas.

[0037] In this context, a geographic grid refers to a regular grid unit that divides a continuous geographic space into grid cells of preset sizes. For example, the area of ​​District B in City A can be divided into multiple 50m × 50m square grids, each grid being a geographic grid. Preset quality control rules refer to pre-defined criteria used to determine whether a user's service experience is poor. For example, a preset quality control rule could be a video bitrate lower than 720P, an initial buffering time greater than 2 seconds, or more than 2 instances of buffering; meeting any one of these criteria would classify the user as having poor quality experience.

[0038] The total number of users refers to the total number of different users appearing within a geographic raster, obtained by counting the different International Mobile Subscriber Identity (IMSI) codes. For example, if 30 different IMSI codes appear in all multi-source heterogeneous fusion records within a 50m × 50m geographic raster, the total number of users in this raster is 30. The proportion of low-quality users refers to the ratio of low-quality users to the total number of users within a geographic raster. For example, if 10 low-quality users exist in a raster and the total number of users is 30, the proportion of low-quality users in this raster is approximately 10 / 30 ≈ 33.3%.

[0039] Density threshold refers to the lower limit of the proportion of users with poor quality when determining whether a geographic raster is a candidate raster. For example, if the density threshold is set to 30%, and a raster has a poor quality user ratio of 33.3%, which is greater than 30%, this raster is marked as a candidate raster. Spatial distance refers to the Euclidean distance or grid adjacency between the center points of two geographic rasters. For example, two adjacent 50m×50m rasters whose center points are 50m apart satisfy the spatial distance adjacency condition. Temporal continuity refers to the temporal proximity of candidate rasters within different time windows, usually measured by the difference in time window indices. For example, candidate rasters in the first and second time windows are temporally continuous if the difference in their time window indices is 1.

[0040] The preset threshold refers to a pre-defined boundary value used to determine whether the spatial distance and temporal continuity between rasters meet the clustering requirements. For example, the spatial adjacency condition is that the rasters are directly adjacent (sharing an edge), and the temporal continuity condition is that the time window index difference is ≤1. These two conditions are the preset thresholds. A poor-quality hotspot region refers to a continuous region formed by clustering multiple candidate rasters that meet the density threshold and are spatiotemporally continuous. For example, in District B of City A, 15 adjacent candidate rasters meet the density threshold for three consecutive time windows; these rasters together constitute a poor-quality hotspot region.

[0041] S4. Based on the fusion records contained in the poor quality hotspot region, extract feature vectors to characterize the network problem type, and input the feature vectors into a lightweight classification model, outputting the root cause diagnosis results corresponding to the poor quality hotspot region according to a preset priority order.

[0042] Here, "fusion record" refers to a shorthand for "multi-source heterogeneous fusion record," which is a user-level service record that integrates three data sources. For example, the aforementioned multi-source heterogeneous fusion record for user C is a fusion record. "Network problem type" refers to the classification of wireless network-level causes leading to poor user service experience, including weak coverage, high load, and interference. For example, weak coverage, high load, and interference are three common network problem types. "Feature vector" refers to a vector form composed of multiple feature values ​​reflecting the network problem type, used as input to the classification model for root cause diagnosis. For example, for a poor-quality hotspot area, extracting the percentages of weak coverage (55%), high load (45%), and interference (20%), these three values ​​form a feature vector [55%, 45%, 20%).

[0043] Lightweight classification models refer to machine learning models that are simple in structure, computationally inexpensive, but highly accurate, used to output root cause diagnostic results based on feature vectors. For example, a pre-trained decision tree model is a lightweight classification model that incorporates priority rules. Preset priority order refers to the pre-defined order in which different network problem types are diagnosed. For example, prioritizing weak coverage, then high load, and finally interference is a preset priority order.

[0044] The root cause diagnosis result refers to the main problem type output after the classification model diagnoses the poor-quality hotspot areas. For example, for the aforementioned poor-quality hotspot areas, the model outputs a weakly covered root cause type, which is the root cause diagnosis result.

[0045] The technical solution of this embodiment integrates short video service experience data collected by the over-the-top application platform, signaling-level user behavior data collected by the service experience quality platform, and key performance indicator data of the wireless network collected by the operation and maintenance center system. Based on timestamps and spatial identifiers, it generates multi-source heterogeneous fusion records through spatiotemporal alignment. Then, it identifies poor-quality hotspot areas through geographic grid aggregation and dynamic spatiotemporal clustering. Finally, it uses a lightweight classification model to automatically diagnose the root causes and prioritize poor-quality hotspot areas. This solves the problems of lagging identification of poor-quality scenes and reliance on human experience for root cause localization caused by the single data dimension, coarse positioning granularity, and lack of intelligent closed-loop mechanism in the prior art. It improves the identification accuracy of poor-quality areas in short video networks and the degree of automation of root cause localization.

[0046] In one alternative approach, S1 specifically includes: The over-the-top application platform receives the short video service experience data reported by the client software development kit. The short video service experience data includes the time when the user service occurs, video bitrate, initial buffering time, number of stutters, application type, indoor and outdoor identifiers, and user latitude and longitude.

[0047] In this context, a client-side software development kit (SDK) refers to a set of program interfaces integrated into a mobile application, used to collect user-side data and report it to the application platform. For example, the SSD integrated into the D client can collect information such as the user's mobile phone model, network status, and video playback parameters, and report it to the D server.

[0048] The "User Service Instance Time" refers to the specific point in time when a user begins using a service or during a service interaction. For example, the moment user C clicks to play video D at 20:15:30 on January 1, 2026 is the "User Service Instance Time." "Video Bitrate" refers to the amount of data transmitted per unit of time in a video file, usually expressed in kbps or petabytes (P). For example, if user C is watching a video with a bitrate of 360P, it's considered low and may affect the viewing experience. "Initial Buffering Time" refers to the waiting time from when a user clicks to play a video until the video starts playing. For example, user C's initial buffering time for video D is 2.5 seconds, exceeding the normal experience threshold. "Number of Stutters" refers to the number of times video playback is interrupted due to buffering during a single playback session. For example, user C experienced three stutters while watching a one-minute video. "Application Type" refers to the name or category of the mobile application used by the user. For example, user C is using application type D. "Indoor / Outdoor Identifier" indicates whether the user's current environment is indoors or outdoors. For example, if user C is watching a video outdoors, indoor and outdoor are marked as outdoors. User latitude and longitude refer to the longitude and latitude coordinates of the user's current location. For example, user C's location is at 119.3 degrees east longitude and 26.1 degrees north latitude.

[0049] The service experience quality platform receives user behavior data obtained from signaling plane parsing. The signaling-level user behavior data includes International Mobile Subscriber Identity (IMSI), service start timestamp, service end timestamp, first cell identifier, and terminal model.

[0050] The signaling plane refers to the logical channel in a mobile communication network used for transmitting control signaling, distinct from the user plane that transmits user data. For example, a service experience quality platform can obtain signaling messages indicating the start and end of user services by parsing signaling plane data.

[0051] The International Mobile Subscriber Identity (IMSI) is a number used to uniquely identify a mobile user globally. For example, the IMSI for user C's mobile phone SIM card is 460011234567890. The service start timestamp is the precise moment a user's service session begins. For example, the time user C starts playing video D, January 1, 2026, at 20:15:30, is the service start timestamp. The service end timestamp is the precise moment a user's service session ends. For example, the time user C ends playing this video, January 1, 2026, at 20:16:30, is the service end timestamp. The first cell identifier is the unique identifier of the serving cell currently accessed by the user, recorded in the signaling-level user behavior data. For example, the cell accessed by user C, recorded in the service experience quality platform data, is CELL_A001, and CELL_A001 is the first cell identifier. The terminal model refers to the brand and model of the mobile device used by the user. For example, user C uses a phone called PHONE_X.

[0052] The operation and maintenance center system receives the key performance indicator data of the wireless network output from the northbound interface of the network management system. The key performance indicator data of the wireless network includes the indicator collection time, the second cell identifier, and the average value of the reference signal received power at the cell level, the signal-to-interference-plus-noise ratio, the physical resource block utilization rate, and the call drop rate.

[0053] The northbound interface of the network management system refers to the standardized interface through which the operation and maintenance center system provides data access to the upper-level network management system or third-party systems. For example, the operation and maintenance center system periodically pushes out key performance indicator data of the wireless network in cell CELL_A001 through the northbound interface.

[0054] The "Indicator Acquisition Time" refers to the point in time when the operation and maintenance center system collects the key performance indicator data of the wireless network. For example, if the operation and maintenance center system collects the indicator data of cell CELL_A001 at 20:15:00 on January 1, 2026, this time is the indicator acquisition time. The "Second Cell Identifier" refers to the cell identifier corresponding to the key performance indicator data of the wireless network. For example, in the key performance indicator data of the wireless network output by the operation and maintenance center system, the cell identifier field is CELL_A001, and CELL_A001 is the second cell identifier. The "Average Reference Signal Received Power" refers to the average value of the reference signal received power measured by all users in the cell within a statistical period, reflecting the cell coverage level. For example, the average reference signal received power of cell CELL_A001 at 20:15:00 is -112dBm. The "Average Signal-to-Interference-plus-Noise Ratio" refers to the average value of the signal-to-interference-plus-noise ratio within the cell within a statistical period, reflecting signal quality. For example, the average signal-to-interference-plus-noise ratio of cell CELL_A001 is 5dB. Physical resource block utilization rate refers to the proportion of physical resource blocks in a cell that are occupied within a statistical period, reflecting the cell's load status. For example, the physical resource block utilization rate of cell CELL_A001 is 78%. Call drop rate refers to the proportion of dropped calls within a cell to the total number of calls within a statistical period, reflecting network stability. For example, the call drop rate of cell CELL_A001 is 0.1%.

[0055] Among the above-mentioned optional methods, the collection process of multi-source heterogeneous data is further refined. By clarifying the specific data fields and interface sources of the top application platform, business experience quality platform and operation and maintenance center system, the problems of ambiguous data sources and unclear field definitions are solved, and the standardization and traceability of data collection are improved.

[0056] In one alternative approach, S2 specifically includes: Extract the time of the user's service and the user's latitude and longitude from each piece of short video service experience data.

[0057] Specifically, for each short video service experience data item, the system parses the recorded field structure to identify the preset user service occurrence time field and reads its value as the user service occurrence time corresponding to that short video service experience data item; simultaneously, it identifies the preset user latitude and longitude field and reads its value as the user latitude and longitude corresponding to that short video service experience data item. The user service occurrence time field stores the precise time when the user started using the short video service, and the user latitude and longitude field stores the user's current location's latitude and longitude coordinates.

[0058] Extract the service start timestamp, service end timestamp, first cell identifier, and International Mobile Subscriber Identity from each of the aforementioned signaling-level user behavior data.

[0059] Specifically, for each signaling-level user behavior data, by parsing the field structure recorded therein, the field value is read from the preset service start timestamp field as the service start timestamp, the field value is read from the preset service end timestamp field as the service end timestamp, the field value is read from the preset first cell identifier field as the first cell identifier, and the field value is read from the preset International Mobile Subscriber Identity field as the International Mobile Subscriber Identity.

[0060] Extract the data collection time and the second cell identifier from each of the key performance indicators of the wireless network.

[0061] Specifically, for each wireless network key performance indicator data, by parsing the field structure recorded therein, the field value is read from the preset indicator collection time field as the indicator collection time, and the field value is read from the preset second cell identifier field as the second cell identifier.

[0062] For each piece of short video service experience data, the target cell corresponding to the short video service experience data is determined based on the user's latitude and longitude.

[0063] The target cell refers to the cell to which the user actually belongs, calculated based on the user's latitude and longitude using a geographic information system or cell coverage area. For example, by matching user C's latitude and longitude (119.3°E, 26.1°N) on a map, user C is determined to belong to cell CELL_A001, which is the target cell.

[0064] Specifically, for each short video service experience data, based on the user's latitude and longitude, the pre-configured geographic information data of cell coverage area is used to retrieve the cell coverage area containing the user's latitude and longitude coordinates through spatial inclusion relationships, and the cell identifier corresponding to the retrieved cell coverage area is determined as the target cell corresponding to the short video service experience data.

[0065] The signaling-level user behavior data that has the same International Mobile Subscriber Identity (IMSI), the user service occurrence time is between the service start timestamp and the service end timestamp, and the first cell identifier is consistent with the target cell is searched in the signaling-level user behavior data and used as the matching signaling data.

[0066] In this context, matched signaling data refers to the signaling record within signaling-level user behavior data that successfully matches the International Mobile Subscriber Identity (IMSI), time range, and cell identifier. For example, if a signaling record is found in the service experience quality platform with IMSI 460011234567890, service time covering January 1, 2026, 20:15:30, and the first cell identifier CELL_A001 matches the target cell, this record is considered matched signaling data.

[0067] Specifically, among all signaling-level user behavior data, multiple signaling records with the same International Mobile Subscriber Identity (IMSI) are selected. From the selected signaling records, those with a user service occurrence time no less than the service start timestamp and no greater than the service end timestamp are selected. Then, from the selected signaling records, those with a first cell identifier consistent with the target cell are selected. Finally, the selected signaling records are determined as the matching signaling data.

[0068] The KPI data that matches the second cell identifier of the target cell and whose time difference between the time of data collection and the time of user service occurrence is less than a preset time threshold is searched in the KPI data of the wireless network.

[0069] In this context, the matched KPI data refers to the record in the wireless network key performance indicator data that is successfully matched by cell identifier and timestamp. For example, if a record of wireless network key performance indicator data is found in the operation and maintenance center system, and the second cell identifier of this record is CELL_A001, and the time difference between the indicator collection time of 20:15:00 on January 1, 2026 and the user service occurrence time of 20:15:30 on January 1, 2026 is less than a preset time threshold (such as 5 minutes), then this record is a matched wireless network key performance indicator data.

[0070] Specifically, among all the key performance indicator data of the wireless network, multiple indicator records with the same second cell identifier as the target cell are selected. From the selected indicator records, the absolute value of the time difference between the indicator collection time and the user service occurrence time is selected to be less than a preset time threshold. The finally selected indicator records are determined as the matching key performance indicator data of the wireless network.

[0071] The short video service experience data, the matched signaling data, and the matched KPI data are merged into a single multi-source heterogeneous fusion record.

[0072] Repeat the steps of determining the target cell corresponding to each short video service experience data based on the user's latitude and longitude until all short video service experience data is processed, resulting in multiple multi-source heterogeneous fusion records.

[0073] Among the above-mentioned optional methods, the specific association rules and matching logic for spatiotemporal alignment are further defined. Through multiple constraints such as user identifier, time window and cell identifier, the problems of inaccurate association of heterogeneous data sources and redundancy of fused records are solved, thereby improving the accuracy of data fusion and the quality of records.

[0074] In one alternative approach, S3 specifically includes: The area to be analyzed is divided into multiple geographic grids, each corresponding to a spatial range of a preset size.

[0075] The area to be analyzed refers to the geographical region where poor-quality areas need to be identified, typically specified by network optimization personnel. For example, the area analyzed in this case is the business district of District B in City A. The preset spatial size refers to the size of each grid cell when dividing the geographic grid, such as 50m × 50m.

[0076] Specifically, the area to be analyzed is divided into grids according to a preset size spatial range to generate multiple geographic grids that are evenly distributed and do not overlap. Each geographic grid corresponds to a preset size spatial range and is assigned a unique geographic grid identifier.

[0077] For each of the multi-source heterogeneous fusion records, the target geographic raster to which the multi-source heterogeneous fusion record belongs is determined based on the user's latitude and longitude.

[0078] The target geographic raster refers to the specific raster to which each multi-source heterogeneous fusion record belongs based on the user's latitude and longitude. For example, if user C's fusion record falls within the raster between 119.30°E and 119.31°E and 26.10°N based on (119.3°E, 26.1°N), this raster is the target geographic raster.

[0079] Specifically, for each multi-source heterogeneous fusion record, the target geographic raster to which the record belongs is calculated based on the user's latitude and longitude using a raster spatial indexing algorithm. The expression for the raster spatial indexing algorithm is as follows: Raster row number = floor((user latitude - minimum latitude of raster area) / raster height) + 1, raster column number = floor((user longitude - minimum longitude of raster area) / raster width) + 1, target geographic raster identifier = raster row number × maximum value of raster column number + raster column number; Among them, the minimum latitude of the raster area refers to the minimum latitude covered by the area to be analyzed, the minimum longitude of the raster area refers to the minimum longitude covered by the area to be analyzed, the raster height refers to the size of each geographic raster in the latitudinal direction, the raster width refers to the size of each geographic raster in the longitudinal direction, the raster row number refers to the sequence number of the geographic raster in the latitudinal direction, the raster column number refers to the sequence number of the geographic raster in the longitudinal direction, the maximum raster column number refers to the maximum raster column number in the latitudinal direction within the area to be analyzed, and the target geographic raster identifier refers to the number that uniquely identifies a geographic raster.

[0080] For each target geographic raster, obtain all multi-source heterogeneous fusion records belonging to the target geographic raster, and determine whether the multi-source heterogeneous fusion record meets the preset quality difference rule based on the short video service experience data in each multi-source heterogeneous fusion record.

[0081] Specifically, for each target geographic raster, all multi-source heterogeneous fusion records whose target geographic raster identifier is the same as that of the target geographic raster are selected from all multi-source heterogeneous fusion records and are regarded as multi-source heterogeneous fusion records belonging to the target geographic raster. For each multi-source heterogeneous fusion record belonging to the target geographic raster, short video service experience data is extracted from the multi-source heterogeneous fusion record, and it is determined whether the short video service experience data meets the quality defect conditions according to the preset quality defect rules. The preset quality defect rules include at least one of the following judgment conditions: video bitrate is lower than a preset bitrate threshold, initial buffer duration is greater than a preset buffer duration threshold, and the number of stutters is greater than a preset stutter number threshold. If the short video service experience data meets any one of the judgment conditions, it is determined that the multi-source heterogeneous fusion record meets the preset quality defect rules; otherwise, it is determined that the multi-source heterogeneous fusion record does not meet the preset quality defect rules.

[0082] The total number of different International Mobile Subscriber Identity (IMSI) codes appearing in all multi-source heterogeneous fusion records belonging to the target geographic raster is counted, and the number of different IMSI codes appearing in multi-source heterogeneous fusion records belonging to the target geographic raster and satisfying the preset quality difference rule is counted, and the number of poor quality users is counted.

[0083] The number of users with poor quality refers to the number of different users within a geographic raster who meet the preset quality-poor rules across all multi-source heterogeneous fusion records. For example, in a target geographic raster, there are 30 users, and all records of 10 of them meet the quality-poor rules; therefore, the number of users with poor quality in this raster is 10.

[0084] Specifically, for each target geographic raster, all multi-source heterogeneous fusion records belonging to that target geographic raster are collected. The International Mobile Subscriber Identity (IMSI) is extracted from each multi-source heterogeneous fusion record. The number of all different IMSIs is counted by deduplication, and this number is taken as the total number of users in the target geographic raster. At the same time, multi-source heterogeneous fusion records that meet the preset quality difference rules are selected from all multi-source heterogeneous fusion records belonging to that target geographic raster. The IMSIs are extracted from each selected multi-source heterogeneous fusion record. The number of all different IMSIs in these records is counted by deduplication, and this number is taken as the number of poor-quality users in the target geographic raster.

[0085] The proportion of poor-quality users for each target geographic raster is determined based on the ratio of the number of poor-quality users in each target geographic raster to the total number of users.

[0086] Among the above-mentioned optional methods, the process of geographic raster aggregation and calculation of the proportion of poor-quality users is further refined. By clearly defining the raster division rules and the criteria for judging poor quality, the problems of inconsistent statistical standards for poor-quality users and coarse spatial granularity are solved, and the precision of identifying poor-quality areas is improved.

[0087] In one alternative approach, S4 specifically includes: The time period to be analyzed is divided into multiple consecutive time windows.

[0088] The analysis period refers to the time range within which poor-quality areas need to be identified and clustered. This is typically a continuous time period specified by network optimization personnel based on business needs. For example, when performing a poor-quality analysis on the B district of City A, the analysis period might be from 17:00 to 22:00 on January 1, 2026, to analyze the dynamic changes in poor-quality hotspots within this period. A time window refers to a segment of a continuous time axis divided into fixed durations, used to analyze the change in the proportion of poor-quality areas over time. For example, the period from 17:00 to 22:00 might be divided into five time windows, with 20:00 to 21:00 being one of them.

[0089] For each time window, determine the proportion of poor-quality users for each target geographic raster within that time window, and mark all target geographic rasteres whose proportion of poor-quality users is not less than the density threshold as candidate rasters.

[0090] Candidate rasters are geographic rasters whose proportion of users with poor quality is greater than or equal to the density threshold within a certain time window. For example, within the time window of 20:00 to 21:00, if a certain raster has a proportion of users with poor quality of 33.3%, which is greater than the density threshold of 30%, this raster is marked as a candidate raster.

[0091] Specifically, for each time window, the proportion of poor-quality users for each target geographic raster within that time window is obtained, and the proportion of poor-quality users for each target geographic raster is compared with a preset density threshold. If the proportion of poor-quality users for a target geographic raster is not less than the density threshold, then the target geographic raster is marked as a candidate raster.

[0092] For each candidate raster, search for target candidate rasters belonging to the same time window or other time windows among spatially adjacent target geographic rasters, and determine whether the time window index difference between the target candidate raster and the current candidate raster is not greater than the time window interval threshold.

[0093] The target candidate raster refers to the currently selected candidate raster during the clustering process, used for spatiotemporal comparison with neighboring rasteres. For example, when traversing candidate rasteres, the raster currently being processed is the target candidate raster.

[0094] The time window index difference refers to the absolute value of the difference between the time window indices of two candidate rasters. For example, the time window index difference between a raster in the first time window and a raster in the second time window is 1. The time window interval threshold refers to the maximum time window index difference that allows two candidate rasters to cluster. For example, setting the time window interval threshold to 1 allows rasters in adjacent time windows to cluster.

[0095] Specifically, for each candidate raster, all target geographic rasters that are spatially adjacent to the current candidate raster are obtained. From these spatially adjacent target geographic rasters, target candidate rasters that are marked as candidate rasters and belong to the same time window or other time windows are selected. The absolute value of the difference between the time window index of the time window where each target candidate raster is located and the time window index of the time window where the current candidate raster is located is calculated. The calculated absolute value of the difference is compared with the preset time window interval threshold. If the absolute value of the difference is not greater than the time window interval threshold, the target candidate raster is determined to meet the time continuity condition.

[0096] All candidate rasters that meet the spatial adjacency condition and whose time window index difference is not greater than the time window interval threshold are grouped into the same candidate cluster.

[0097] Spatial adjacency refers to the criteria for determining whether two geographic rasters are spatially adjacent, such as sharing an edge or a vertex. For example, two 50m × 50m rasters satisfy the spatial adjacency condition if they share an edge. A candidate cluster is a set of multiple candidate rasters that satisfy the spatial adjacency condition and whose temporal window index difference is no greater than the interval threshold. For example, 15 spatially adjacent and temporally consecutive candidate rasters are grouped into the same candidate cluster.

[0098] Specifically, an empty candidate cluster set is established, and all candidate rasters are traversed. For each candidate raster that has not yet been assigned to any candidate cluster, a breadth-first search or depth-first search is performed starting from the current candidate raster. During the search, other candidate rasters that meet the spatial adjacency condition and whose time window index difference is not greater than the time window interval threshold are found. All the searched candidate rasters are assigned to the same candidate cluster as the current candidate raster, and these candidate rasters are marked as assigned to a cluster. The above process is repeated until all candidate rasters are assigned to the corresponding candidate cluster.

[0099] Candidate clusters containing a number of candidate grids not less than a preset cluster size threshold are identified as the poor-quality hotspot regions.

[0100] The preset cluster size threshold refers to the minimum number of candidate rasters a candidate cluster must contain to be ultimately identified as a poor-quality hotspot region. For example, if the cluster size threshold is set to 10, a candidate cluster containing 15 rasters meets the condition and is identified as a poor-quality hotspot region.

[0101] Among the above-mentioned optional methods, the clustering determination process for poor-quality hotspot areas is further extended. By introducing time window continuity constraints and spatial adjacency conditions, the problem of fragmentation of poor-quality areas and spatiotemporal feature separation caused by static thresholds is solved, thereby improving the continuity and integrity of hotspot area clustering.

[0102] In one alternative approach, S5 specifically includes: For each of the poor-quality hotspot regions, obtain all multi-source heterogeneous fusion records belonging to that poor-quality hotspot region.

[0103] Based on the key performance index data of the wireless network in each multi-source heterogeneous fusion record, the weak coverage ratio, high load ratio and interference intensity are calculated respectively, and the weak coverage ratio, the high load ratio and the interference intensity are combined into the feature vector.

[0104] The weak coverage percentage refers to the proportion of records with an average received reference signal power below the weak coverage threshold among all multi-source heterogeneous fusion records within a poor-quality hotspot area. For example, if there are 100 records in a hotspot area, and 55 of them have an average received reference signal power less than -110 dBm, the weak coverage percentage for this hotspot area is 55%. The high load percentage refers to the proportion of records with a physical resource block utilization rate above the high load threshold among all multi-source heterogeneous fusion records within a poor-quality hotspot area. For example, if 45 of the 100 records have a physical resource block utilization rate greater than 70%, the high load percentage for this hotspot area is 45%. The interference intensity refers to the proportion of records with a signal-to-interference-plus-noise ratio below the interference threshold among all multi-source heterogeneous fusion records within a poor-quality hotspot area. For example, if 20 of the 100 records have a signal-to-interference-plus-noise ratio less than 3 dB, the interference intensity for this hotspot area is 20%.

[0105] Specifically, for each poor-quality hotspot region, all multi-source heterogeneous fusion records belonging to that poor-quality hotspot region are obtained, and the total number of multi-source heterogeneous fusion records is counted and denoted as . The number of records in the multi-source heterogeneous fusion record where the average received power of the reference signal is less than a preset weak coverage threshold is recorded as follows: According to the formula Calculate the proportion of weak coverage The number of records in the multi-source heterogeneous fusion record where the physical resource block utilization rate is greater than the preset high load threshold is recorded as follows: According to the formula Calculate the proportion of high load The number of records in the multi-source heterogeneous fusion recording where the signal-to-interference plus-noise ratio is less than a preset interference threshold is recorded as follows: According to the formula Calculate interference strength The proportion of weak coverage High load ratio With interference intensity Combined into feature vectors eigenvectors Represented as .

[0106] The feature vector is input into the pre-trained lightweight classification model, which has built-in first root cause judgment rules, second root cause judgment rules and third root cause judgment rules arranged in the preset priority order.

[0107] The first root cause rule refers to the highest priority root cause rule in a lightweight classification model, typically corresponding to weak coverage. For example, the first condition in a decision tree is whether the percentage of weak coverage is greater than 50%. The second root cause rule refers to the next highest priority root cause rule in a lightweight classification model, typically corresponding to high load. For example, the second condition in a decision tree is whether the percentage of high load is greater than 40%. The third root cause rule refers to the lowest priority root cause rule in a lightweight classification model, typically corresponding to interference. For example, the third condition in a decision tree is whether the interference intensity is greater than 5%.

[0108] The lightweight classification model sequentially executes the first root cause judgment rule, the second root cause judgment rule, and the third root cause judgment rule.

[0109] When the proportion of weak coverage is greater than the first preset threshold, the first root cause judgment rule is triggered and the weak coverage root cause type is output.

[0110] Here, the first preset threshold refers to the threshold used in the first root cause judgment rule, such as 50%. For example, if the weak coverage ratio is greater than 50%, the first root cause judgment rule is triggered. The weak coverage root cause type refers to the root cause output by the model when the weak coverage ratio exceeds the first preset threshold. For example, for the aforementioned hotspot area, the model outputs the weak coverage root cause type.

[0111] When the proportion of weak coverage is not greater than the first preset threshold and the proportion of high load is greater than the second preset threshold, the second root cause judgment rule is triggered and the high load root cause type is output.

[0112] The second preset threshold refers to the threshold used in the second root cause judgment rule, such as 40%. For example, if the weak coverage ratio is no more than 50% and the high load ratio is greater than 40%, the second root cause judgment rule is triggered. The high load root cause type refers to the root cause output by the model when the high load ratio exceeds the second preset threshold. For example, if a hotspot area has a weak coverage ratio of 30% and a high load ratio of 45%, the model outputs a high load root cause type.

[0113] When the proportion of weak coverage is not greater than the first preset threshold, the proportion of high load is not greater than the second preset threshold, and the interference intensity is greater than the third preset threshold, the third root cause judgment rule is triggered and the interference root cause type is output.

[0114] The third preset threshold refers to the threshold used in the third root cause judgment rule, such as 5%. For example, if the weak coverage ratio is no more than 50%, the high load ratio is no more than 40%, and the interference intensity is greater than 5%, the third root cause judgment rule is triggered. The interference root cause type refers to the root cause output by the model when the interference intensity exceeds the third preset threshold. For example, if a hotspot area has a weak coverage ratio of 30%, a high load ratio of 30%, and an interference intensity of 10%, the model outputs the interference root cause type.

[0115] The root cause type output by the triggered root cause judgment rule is used as the root cause diagnosis result.

[0116] Among the above-mentioned optional methods, the priority determination process for root cause diagnosis is further standardized. By using the step-by-step rule execution order built into the lightweight classification model, the problems of determination conflict and diagnostic logic confusion when multiple root causes coexist are solved, thereby improving the orderliness and accuracy of root cause localization.

[0117] In one alternative approach, it also includes: Obtain the root cause type from the root cause diagnosis results.

[0118] Root cause type refers to the main network problem category identified in the root cause diagnosis results, including weak coverage root cause type, high load root cause type, and interference root cause type. For example, for the hotspot area in District B of City A, the root cause type is weak coverage root cause type.

[0119] Based on the preset mapping relationship between root cause types and remediation suggestions, remediation suggestions corresponding to the root cause types are determined; wherein, the weak coverage root cause type corresponds to remediation suggestions for co-construction and sharing or supplementing sites, the high load root cause type corresponds to remediation suggestions for capacity expansion or load balancing, and the interference root cause type corresponds to remediation suggestions for interference investigation or physical cell identifier optimization.

[0120] Among them, "rectification recommendations" refer to network optimization measures proposed for specific root cause types. For example, the rectification recommendations for the weak coverage root cause type are co-construction and sharing or adding sites. "Mapping relationship" refers to a pre-established correspondence table between root cause types and rectification recommendations. For example, in the preset mapping relationship, weak coverage corresponds to co-construction and sharing or adding sites, high load corresponds to capacity expansion or load balancing, and interference corresponds to interference investigation or physical cell identifier optimization.

[0121] The location information of the poor quality hotspot area, the root cause type, the remediation suggestion, and the preset priority identifier are combined to generate an electronic work order, and the electronic work order is pushed to the network optimization system.

[0122] Priority flags refer to the urgency level of work orders, categorized by root cause type or severity of hotspot areas, such as high, medium, and low. For example, a large hotspot area caused by weak coverage could be assigned a high priority flag. Electronic work orders are digital task orders containing information on poor-quality hotspot areas, root cause type, remediation recommendations, and priority flags, used to dispatch network optimization personnel for execution. For example, a generated electronic work order might include location (City A, District B), root cause type (weak coverage), remediation recommendation (adding sites), and priority (high), and be automatically pushed to the network optimization system.

[0123] In this context, the network optimization system refers to a comprehensive platform used by telecommunications operators to manage, schedule, and execute network optimization tasks. For example, after receiving an electronic work order, operator S's network optimization system will assign it to an engineer for on-site inspection and rectification.

[0124] Among the above-mentioned optional methods, the application of diagnostic results is further extended. By mapping the root cause type with the remediation recommendations and generating electronic work orders, the problem of the disconnect between diagnostic results and operation and maintenance actions is solved, and the automation level and handling efficiency of network optimization response are improved.

[0125] Figure 2 The overall architecture diagram of the technical solution in this embodiment is shown. Figure 2 The three external data sources include the Over-the-Top (OTT) platform, the Service Experience Quality (SeQ) platform, and the Operation and Maintenance Center (OMC) system. The OTT platform provides user-level short video service experience data, including application type, video bitrate, initial buffering time, number of buffering events, and user latitude and longitude from the user's geographic location information. The SeQ platform provides signaling-level user behavior data, including the International Mobile Subscriber Identity (IMSI), service start timestamp, service end timestamp, first cell identifier, and terminal model. The OMC system provides key performance indicator (KPI) data for the wireless network, including the indicator collection time, second cell identifier, and cell-level average reference received power, signal-to-interference-plus-noise ratio (SNR), physical resource block utilization, and call drop rate.

[0126] The multi-source data fusion engine receives short video service experience data from OTT platforms, signaling-level user behavior data from SeQ platforms, and wireless network key performance indicator data from OMC systems. Using timestamps and spatial identifiers as association keys, the engine spatiotemporally aligns short video service experience data, signaling-level user behavior data, and wireless network key performance indicator data belonging to the same user, the same service session period, and the same spatiotemporal location, generating a multi-source heterogeneous fusion record. The timestamps include the user service occurrence time in the short video service experience data, the service start and end timestamps in the signaling-level user behavior data, and the indicator collection time in the wireless network key performance indicator data. The spatial identifiers include the user's latitude and longitude in the short video service experience data, the first cell identifier in the signaling-level user behavior data, and the second cell identifier in the wireless network key performance indicator data. Each multi-source heterogeneous fusion record corresponds to one user's service session period and includes information such as the user service occurrence time, user latitude and longitude, video bitrate, International Mobile Subscriber Identity (IMSI), first cell identifier, and average reference signal received power.

[0127] The dynamic poor-quality region clusterer receives multi-source heterogeneous fusion records output by the multi-source data fusion engine and aggregates these records according to geographic rasters. A geographic raster is a grid unit that divides the area to be analyzed into spatial ranges of preset sizes. The dynamic poor-quality region clusterer determines the proportion of users with poor quality based on the number of users within each geographic raster that meet preset poor-quality rules and the total number of users. The preset poor-quality rules include at least one of the following criteria: video bitrate lower than a preset bitrate threshold, initial buffer duration greater than a preset buffer duration threshold, and number of stutters greater than a preset stutter count threshold. The dynamic poor-quality region clusterer uses a spatiotemporal density clustering algorithm to cluster geographic rasters where the proportion of users with poor quality is not less than a density threshold and the spatial distance and temporal continuity between them meet preset thresholds, obtaining poor-quality hotspot regions and outputting a hotspot map. Spatial adjacency is defined as geographic rasters sharing edges or vertices, and temporal continuity is defined as the difference in time window indices of candidate rasters not exceeding a time window interval threshold.

[0128] The root cause intelligent diagnostic model receives multi-source heterogeneous fusion records corresponding to poor-quality hotspot regions output by a dynamic poor-quality region clusterer. For each poor-quality hotspot region, it calculates the weak coverage ratio, high load ratio, and interference intensity, and combines these three factors into a feature vector. The weak coverage ratio is determined by the ratio of the number of multi-source heterogeneous fusion records in the poor-quality hotspot region whose average received power of the reference signal is less than a preset weak coverage threshold to the total number of records. The high load ratio is determined by the ratio of the number of multi-source heterogeneous fusion records whose physical resource block utilization is greater than a preset high load threshold to the total number of records. The interference intensity is determined by the ratio of the number of multi-source heterogeneous fusion records whose signal-to-interference-plus-noise ratio is less than a preset interference threshold to the total number of records. The root cause intelligent diagnostic model inputs the feature vector into a pre-trained lightweight classification model. The lightweight classification model has built-in first root cause judgment rules, second root cause judgment rules, and third root cause judgment rules arranged in a preset priority order.

[0129] Data from three data sources—the OTT platform, the SeQ platform, and the OMC system—is input into a multi-source data fusion engine in parallel. The multi-source data fusion engine then outputs the generated multi-source heterogeneous fusion records to a dynamic poor-quality region clusterer and a root cause intelligent diagnostic model. The dynamic poor-quality region clusterer and the root cause intelligent diagnostic model process the data in parallel. The dynamic poor-quality region clusterer outputs poor-quality hotspot regions, and the root cause intelligent diagnostic model outputs root cause diagnostic results based on these hotspot regions. Finally, the root cause intelligent diagnostic model pushes the root cause diagnostic results to the network optimization system for remediation.

[0130] like Figure 3As shown in the diagram, the root cause intelligent diagnosis decision tree illustrates the logical flow of root cause diagnosis. The root cause intelligent diagnosis model receives feature vectors of poor-quality hotspot areas as input. These feature vectors include the proportion of weak coverage, the proportion of high load, and the interference intensity. The root cause intelligent diagnosis model executes root cause judgment rules in a preset priority order. The first step executes the first root cause judgment rule, determining whether the proportion of weak coverage is greater than a first preset threshold, which is set to 50%. If the proportion of weak coverage is greater than 50%, the first root cause judgment rule is triggered and outputs the weak coverage root cause type, with corresponding remediation suggestions of co-construction and sharing or adding sites. If the proportion of weak coverage is not greater than 50%, the second step executes the second root cause judgment rule, determining whether the proportion of high load is greater than a second preset threshold, which is set to 40%. If the proportion of high load is greater than 40%, the second root cause judgment rule is triggered and outputs the high load root cause type, with corresponding remediation suggestions of capacity expansion or load balancing. If the proportion of high load is not greater than 40%, the third step executes the third root cause judgment rule, determining whether the interference intensity is greater than a third preset threshold, which is set to 5%. If the interference intensity exceeds 5%, the third root cause judgment rule is triggered, and the type of interference root cause is output. The corresponding remediation suggestion is interference investigation or physical cell identifier optimization. By gradually eliminating the main causes through the above priority order, the root cause diagnosis results and remediation suggestions are ensured to be both operable and cost-effective.

[0131] In another embodiment of the short video network poor quality region identification and root cause localization method of the present invention, the specific steps include: S10. Obtain short video service experience data collected by the top application platform, signaling-level user behavior data collected by the service experience quality platform, and key performance indicator data of wireless network collected by the operation and maintenance center system.

[0132] S20. Using timestamps and spatial identifiers as association keys, the short video service experience data, signaling-level user behavior data, and wireless network key performance indicator data of the same user within the same service session period are spatiotemporally aligned to generate multiple multi-source heterogeneous fusion records. Each multi-source heterogeneous fusion record corresponds to one service session period of a user.

[0133] S30. Aggregate multiple multi-source heterogeneous fusion records according to geographic rasters. Determine the proportion of users with poor quality based on the number of users in each geographic raster that meet the preset poor quality rules and the total number of users. Then, cluster geographic rasters whose proportion of users with poor quality meets the density threshold and whose spatial distance and temporal continuity between them meet the preset thresholds, and obtain poor quality hotspot areas.

[0134] S40. For each poor-quality hotspot area, extract all multi-source heterogeneous fusion records belonging to the poor-quality hotspot area. Calculate the weak coverage ratio, high load ratio, and interference intensity based on the wireless network key performance indicator data in each multi-source heterogeneous fusion record. Combine the weak coverage ratio, high load ratio, and interference intensity into a feature vector. Input the feature vector into a lightweight classification model. The lightweight classification model has built-in first root cause judgment rules, second root cause judgment rules, and third root cause judgment rules arranged in a preset priority order. After executing the first root cause judgment rules, second root cause judgment rules, and third root cause judgment rules in sequence, output the root cause type corresponding to the poor-quality hotspot area as the root cause diagnosis result.

[0135] S50. Based on the total number of users, weak coverage ratio, high load ratio, and interference intensity in the poor quality hotspot area, calculate the priority score for each poor quality hotspot area by weighted summation. In the weighted summation process, the total number of users, weak coverage ratio, high load ratio, and interference intensity are multiplied by their respective preset weight coefficients and then added together to obtain the priority score. The preset weight coefficients are periodically adjusted based on the historical work order resolution efficiency.

[0136] S60. Based on the root cause type in the root cause diagnosis results, match the corresponding remediation suggestions from the preset root cause type and remediation suggestion mapping relationship. The weak coverage root cause type corresponds to the co-construction and sharing or supplementary site remediation suggestion, the high load root cause type corresponds to the capacity expansion or load balancing remediation suggestion, and the interference root cause type corresponds to the interference investigation or physical cell identifier optimization remediation suggestion. Combine the location information, root cause type, remediation suggestion and priority score of the poor quality hotspot area to generate an electronic work order, and attach a priority identifier based on the priority score to the electronic work order.

[0137] S70. Push the electronic work order to the network optimization system. The network optimization system automatically assigns the electronic work order to the corresponding engineer terminal according to the priority identifier. Receive the work order processing status and actual on-site rectification measures record returned by the network optimization system, and compare and verify the actual on-site rectification measures with the root cause diagnosis results.

[0138] S80. Store the comparison and verification results as feedback samples in the historical database, and use the feedback samples to perform incremental training on the lightweight classification model periodically to optimize the model parameters; at the same time, dynamically adjust the preset weight coefficients used in the priority score calculation process based on the historical work order processing time and resolution effect statistics.

[0139] This embodiment acquires and spatiotemporally aligns multi-source heterogeneous data to generate fused records. The fused records are then aggregated using geographic raster and subjected to spatiotemporal density clustering to identify poor-quality hotspot areas. A lightweight classification model is used for root cause diagnosis. Electronic work orders are generated based on priority scores and pushed to the network optimization system. Finally, the on-site remediation measures are compared and verified with the diagnostic results, and used as feedback samples to optimize model parameters and weight coefficients. This solves the problem in existing technologies where diagnostic results cannot be continuously optimized, leading to a decrease in accuracy. It achieves continuous improvement in the accuracy of root cause location and dynamic optimization of remediation resource scheduling.

[0140] Figure 4 This diagram illustrates a structural schematic of an embodiment of a short video network poor quality region identification and root cause localization device 200 provided by the present invention. Figure 4 As shown, the short video network poor quality area identification and root cause localization device 200 includes: The data acquisition module 201 is used to acquire short video service experience data collected by the over-the-top application platform, signaling-level user behavior data collected by the service experience quality platform, and wireless network key performance indicator data collected by the operation and maintenance center system. The alignment and fusion module 202 is used to perform spatiotemporal alignment of the short video service experience data, the signaling-level user behavior data and the wireless network key performance indicator data of the same user within the same service session period using timestamps and spatial identifiers as association keys, and generate multiple multi-source heterogeneous fusion records, each of which corresponds to one service session period of a user. The region identification module 203 is used to aggregate the multi-source heterogeneous fusion records according to geographic grids, determine the proportion of poor quality users based on the number of users in each geographic grid that meet the preset poor quality rules and the total number of users, and cluster the geographic grids whose proportion of poor quality users meets the density threshold and whose spatial distance and temporal continuity between them meet the preset threshold to obtain poor quality hotspot areas. The root cause localization module 204 is used to extract feature vectors that characterize the network problem type based on the fusion records contained in the poor quality hotspot area, and input the feature vectors into a lightweight classification model, and output the root cause diagnosis results corresponding to the poor quality hotspot area according to a preset priority order.

[0141] In one alternative embodiment, the data acquisition module 201 is specifically used for: The over-the-top application platform receives the short video service experience data reported by the client software development kit. The short video service experience data includes the time when the user service occurs, video bitrate, initial buffering time, number of stutters, application type, indoor and outdoor markers, and user latitude and longitude. The service experience quality platform receives user behavior data obtained from signaling plane parsing. The signaling-level user behavior data includes International Mobile Subscriber Identity (IMSI), service start timestamp, service end timestamp, first cell identifier, and terminal model. The operation and maintenance center system receives the key performance indicator data of the wireless network output from the northbound interface of the network management system. The key performance indicator data of the wireless network includes the indicator collection time, the second cell identifier, and the average value of the reference signal received power at the cell level, the signal-to-interference-plus-noise ratio, the physical resource block utilization rate, and the call drop rate.

[0142] In one alternative embodiment, the alignment and fusion module 202 is specifically used for: Extract the time of the user's service and the user's latitude and longitude from each piece of short video service experience data; Extract the service start timestamp, service end timestamp, first cell identifier, and International Mobile Subscriber Identity from each of the aforementioned signaling-level user behavior data; Extract the index collection time and the second cell identifier from each of the aforementioned key performance index data of the wireless network; For each piece of short video service experience data, the target cell corresponding to the short video service experience data is determined based on the user's latitude and longitude. The signaling-level user behavior data that has the same International Mobile Subscriber Identity (IMSI), the user service occurrence time is between the service start time stamp and the service end time stamp, and the first cell identifier is consistent with the target cell is searched in the signaling-level user behavior data and used as the matching signaling data; The KPI data that matches the second cell identifier of the target cell and whose time difference between the time of the indicator collection and the time of the user service occurrence is less than a preset time threshold is found in the wireless network key performance indicator data and is used as the matching KPI data. The short video service experience data, the matched signaling data, and the matched KPI data are merged into a single multi-source heterogeneous fusion record; Repeat the steps of determining the target cell corresponding to each short video service experience data based on the user's latitude and longitude until all short video service experience data is processed, resulting in multiple multi-source heterogeneous fusion records.

[0143] In one alternative embodiment, the region identification module 203 is specifically used for: The area to be analyzed is divided into multiple geographic grids, and each geographic grid corresponds to a spatial range of a preset size. For each of the aforementioned multi-source heterogeneous fusion records, the target geographic raster to which the multi-source heterogeneous fusion record belongs is determined based on the user's latitude and longitude; For each target geographic raster, obtain all multi-source heterogeneous fusion records belonging to the target geographic raster, and determine whether the multi-source heterogeneous fusion record meets the preset quality difference rule based on the short video service experience data in each multi-source heterogeneous fusion record. The total number of users is calculated as the number of different International Mobile Subscriber Identity (IMSI) codes appearing in all multi-source heterogeneous fusion records belonging to the target geographic raster, and the number of different IMSI codes appearing in multi-source heterogeneous fusion records belonging to the target geographic raster and satisfying the preset quality difference rule is calculated as the number of users with poor quality. The proportion of poor-quality users for each target geographic raster is determined based on the ratio of the number of poor-quality users in each target geographic raster to the total number of users.

[0144] In one alternative embodiment, the region identification module 203 is specifically used for: Divide the time period to be analyzed into multiple consecutive time windows; For each time window, determine the proportion of poor-quality users for each target geographic raster within that time window, and mark all target geographic rasters whose proportion of poor-quality users is not less than the density threshold as candidate rasters. For each candidate raster, search for target candidate rasters belonging to the same time window or other time windows among spatially adjacent target geographic rasters, and determine whether the time window index difference between the target candidate raster and the current candidate raster is not greater than the time window interval threshold. All candidate rasters that meet the spatial adjacency condition and whose time window index difference is not greater than the time window interval threshold are grouped into the same candidate cluster. Candidate clusters containing a number of candidate grids not less than a preset cluster size threshold are identified as the poor-quality hotspot regions.

[0145] In one alternative embodiment, the root cause localization module 204 is specifically used for: For each of the poor-quality hotspot regions, obtain all multi-source heterogeneous fusion records belonging to that poor-quality hotspot region; Based on the key performance index data of the wireless network in each multi-source heterogeneous fusion record, the weak coverage ratio, the high load ratio and the interference intensity are calculated respectively, and the weak coverage ratio, the high load ratio and the interference intensity are combined into the feature vector. The feature vector is input into the pre-trained lightweight classification model, which has a first root cause judgment rule, a second root cause judgment rule and a third root cause judgment rule arranged in the preset priority order. The first root cause judgment rule, the second root cause judgment rule, and the third root cause judgment rule are executed sequentially through the lightweight classification model. When the proportion of weak coverage is greater than the first preset threshold, the first root cause judgment rule is triggered and the weak coverage root cause type is output. When the proportion of weak coverage is not greater than the first preset threshold and the proportion of high load is greater than the second preset threshold, the second root cause judgment rule is triggered and the high load root cause type is output. When the proportion of weak coverage is not greater than the first preset threshold, the proportion of high load is not greater than the second preset threshold, and the interference intensity is greater than the third preset threshold, the third root cause judgment rule is triggered and the interference root cause type is output. The root cause type output by the triggered root cause judgment rule is used as the root cause diagnosis result.

[0146] In one alternative approach, the module further includes: a work order generation module; the work order generation module is used for: Obtain the root cause type from the root cause diagnosis results, wherein the root cause type is: the weak coverage root cause type, the high load root cause type, or the interference root cause type; Based on the preset mapping relationship between root cause types and remediation suggestions, remediation suggestions corresponding to the root cause types are determined; wherein, the weak coverage root cause type corresponds to remediation suggestions for co-construction and sharing or supplementing sites, the high load root cause type corresponds to remediation suggestions for capacity expansion or load balancing, and the interference root cause type corresponds to remediation suggestions for interference investigation or physical cell identifier optimization. The location information of the poor quality hotspot area, the root cause type, the remediation suggestion, and the preset priority identifier are combined to generate an electronic work order, and the electronic work order is pushed to the network optimization system.

[0147] It should be noted that the beneficial effects of the short video network poor quality region identification and root cause localization device 200 provided in the above embodiments are the same as those of the short video network poor quality region identification and root cause localization method described above, and will not be repeated here. Furthermore, the device provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the device can be divided into different functional modules according to the actual situation to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, and will not be repeated here.

[0148] The short video network poor quality region identification and root cause localization device 200 of the present invention can be a computer program (including program code) running on a computer device. For example, the short video network poor quality region identification and root cause localization device 200 of the present invention is an application software that can be used to execute the corresponding steps in the short video network poor quality region identification and root cause localization method of the present invention.

[0149] In some embodiments, the short video network poor quality region identification and root cause localization device 200 of the present invention can be implemented in a combination of hardware and software. As an example, the short video network poor quality region identification and root cause localization device 200 of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the short video network poor quality region identification and root cause localization method of the present invention. For example, the processor in the form of a hardware decoding processor can adopt one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0150] The modules described in the embodiments of this invention can be implemented in software or hardware. The names of the modules are not, in some cases, limiting the scope of the module itself.

[0151] An electronic device according to an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements any of the above-mentioned methods for identifying and locating poor-quality regions in short video networks. That is, an electronic device according to an embodiment of the present invention may include, but is not limited to: a processor and a memory; the memory is used to store the computer program; the processor is used to execute the method for identifying and locating poor-quality regions in short video networks according to any embodiment of the present invention by calling the computer program.

[0152] In one alternative embodiment, an electronic device is provided, such as Figure 5 As shown, Figure 5The illustrated electronic device 4000 includes a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of the present invention.

[0153] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 4001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0154] Bus 4002 may include a path for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 4002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 5 The bus 4002 is represented by only one thick line, but this does not mean that there is only one bus or one type of bus.

[0155] The memory 4003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0156] The memory 4003 stores application code (computer program) for executing the present invention, and its execution is controlled by the processor 4001. The processor 4001 executes the application code stored in the memory 4003 to implement the content shown in the foregoing method embodiments.

[0157] Among them, electronic devices can also be terminal devices. A terminal device can be any terminal device that can install applications and access web pages through applications, including at least one of smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, smart TVs, and smart in-vehicle devices.

[0158] It should be noted that, Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0159] An embodiment of the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements any of the above-mentioned methods for identifying and locating poor-quality areas in short video networks.

[0160] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0161] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the aforementioned method for identifying and locating poor-quality areas in short video networks.

[0162] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can be executed 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 remote computers, the remote computer can 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 can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0163] It should be understood that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using dedicated hardware-based means to perform the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0164] The computer-readable storage medium provided in this invention can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0165] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the method shown in the above embodiments.

[0166] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0167] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0168] Those skilled in the art will recognize that this invention can be implemented as an apparatus, method, or computer program product. Therefore, this invention can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "apparatus." Furthermore, in some embodiments, this invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0169] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for identifying and locating the root cause of poor-quality regions in short video networks, characterized in that, include: Acquire short video service experience data collected by the top application platform, signaling-level user behavior data collected by the service experience quality platform, and key performance indicator data of wireless network collected by the operation and maintenance center system; Using timestamps and spatial identifiers as association keys, the short video service experience data, the signaling-level user behavior data, and the wireless network key performance indicator data of the same user within the same service session period are spatiotemporally aligned to generate multiple multi-source heterogeneous fusion records. Each multi-source heterogeneous fusion record corresponds to one user's service session period. Multiple multi-source heterogeneous fusion records are aggregated according to geographic grids. The proportion of users with poor quality is determined based on the number of users in each geographic grid that meet the preset poor quality rules and the total number of users. Geographic grids whose proportion of users with poor quality meets the density threshold and whose spatial distance and temporal continuity between them meet the preset threshold are clustered to obtain poor quality hotspot areas. Based on the fusion records contained in the poor quality hotspot region, feature vectors for characterizing network problem types are extracted, and the feature vectors are input into a lightweight classification model. The root cause diagnosis results corresponding to the poor quality hotspot region are output according to a preset priority order.

2. The method for identifying and locating poor-quality regions in short video networks according to claim 1, characterized in that, The steps for obtaining short video service experience data collected by the top application platform, signaling-level user behavior data collected by the service experience quality platform, and key performance indicator data of the wireless network collected by the operation and maintenance center system include: The over-the-top application platform receives the short video service experience data reported by the client software development kit. The short video service experience data includes the time when the user service occurs, video bitrate, initial buffering time, number of stutters, application type, indoor and outdoor markers, and user latitude and longitude. The service experience quality platform receives user behavior data obtained from signaling plane parsing. The signaling-level user behavior data includes International Mobile Subscriber Identity (IMSI), service start timestamp, service end timestamp, first cell identifier, and terminal model. The operation and maintenance center system receives the key performance indicator data of the wireless network output from the northbound interface of the network management system. The key performance indicator data of the wireless network includes the indicator collection time, the second cell identifier, and the average value of the reference signal received power at the cell level, the signal-to-interference-plus-noise ratio, the physical resource block utilization rate, and the call drop rate.

3. The method for identifying and locating poor-quality regions in short video networks according to claim 1, characterized in that, The steps of generating multiple multi-source heterogeneous fusion records by spatiotemporally aligning the short video service experience data, the signaling-level user behavior data, and the wireless network key performance indicator data of the same user within the same service session period using timestamps and spatial identifiers as association keys include: Extract the time of the user's service and the user's latitude and longitude from each piece of short video service experience data; Extract the service start timestamp, service end timestamp, first cell identifier, and International Mobile Subscriber Identity from each of the aforementioned signaling-level user behavior data; Extract the index collection time and the second cell identifier from each of the aforementioned key performance index data of the wireless network; For each piece of short video service experience data, the target cell corresponding to the short video service experience data is determined based on the user's latitude and longitude. The signaling-level user behavior data that has the same International Mobile Subscriber Identity (IMSI), the user service occurrence time is between the service start time stamp and the service end time stamp, and the first cell identifier is consistent with the target cell is searched in the signaling-level user behavior data and used as the matching signaling data; The KPI data that matches the second cell identifier of the target cell and whose time difference between the time of the indicator collection and the time of the user service occurrence is less than a preset time threshold is found in the wireless network key performance indicator data and is used as the matching KPI data. The short video service experience data, the matched signaling data, and the matched KPI data are merged into a single multi-source heterogeneous fusion record; Repeat the steps of determining the target cell corresponding to each short video service experience data based on the user's latitude and longitude until all short video service experience data is processed, resulting in multiple multi-source heterogeneous fusion records.

4. The method for identifying and locating poor-quality regions in short video networks according to claim 3, characterized in that, The step of aggregating multiple multi-source heterogeneous fusion records according to geographic rasters and determining the proportion of users with poor quality based on the number of users in each geographic raster that meet the preset quality difference rule and the total number of users includes: The area to be analyzed is divided into multiple geographic grids, and each geographic grid corresponds to a spatial range of a preset size. For each of the aforementioned multi-source heterogeneous fusion records, the target geographic raster to which the multi-source heterogeneous fusion record belongs is determined based on the user's latitude and longitude; For each target geographic raster, obtain all multi-source heterogeneous fusion records belonging to the target geographic raster, and determine whether the multi-source heterogeneous fusion record meets the preset quality difference rule based on the short video service experience data in each multi-source heterogeneous fusion record. The total number of users is calculated as the number of different International Mobile Subscriber Identity (IMSI) codes appearing in all multi-source heterogeneous fusion records belonging to the target geographic raster, and the number of different IMSI codes appearing in multi-source heterogeneous fusion records belonging to the target geographic raster and satisfying the preset quality difference rule is calculated as the number of users with poor quality. The proportion of poor-quality users for each target geographic raster is determined based on the ratio of the number of poor-quality users in each target geographic raster to the total number of users.

5. The method for identifying and locating poor-quality regions in short video networks according to claim 4, characterized in that, The step of clustering geographical rasters where the proportion of poor-quality users meets a density threshold and the spatial distance and temporal continuity between them meet a preset threshold to obtain poor-quality hotspot areas includes: Divide the time period to be analyzed into multiple consecutive time windows; For each time window, determine the proportion of poor-quality users for each target geographic raster within that time window, and mark all target geographic rasters whose proportion of poor-quality users is not less than the density threshold as candidate rasters. For each candidate raster, search for target candidate rasters belonging to the same time window or other time windows among spatially adjacent target geographic rasters, and determine whether the time window index difference between the target candidate raster and the current candidate raster is not greater than the time window interval threshold. All candidate rasters that meet the spatial adjacency condition and whose time window index difference is not greater than the time window interval threshold are grouped into the same candidate cluster. Candidate clusters containing a number of candidate grids not less than a preset cluster size threshold are identified as the poor-quality hotspot regions.

6. The method for identifying and locating poor-quality regions in short video networks according to claim 5, characterized in that, Based on the fusion records contained within the poor-quality hotspot region, the steps of extracting feature vectors to characterize the network problem type, inputting the feature vectors into a lightweight classification model, and outputting root cause diagnosis results corresponding to the poor-quality hotspot region according to a preset priority order include: For each of the poor-quality hotspot regions, obtain all multi-source heterogeneous fusion records belonging to that poor-quality hotspot region; Based on the key performance index data of the wireless network in each multi-source heterogeneous fusion record, the weak coverage ratio, the high load ratio and the interference intensity are calculated respectively, and the weak coverage ratio, the high load ratio and the interference intensity are combined into the feature vector. The feature vector is input into the pre-trained lightweight classification model, which has a first root cause judgment rule, a second root cause judgment rule and a third root cause judgment rule arranged in the preset priority order. The first root cause judgment rule, the second root cause judgment rule, and the third root cause judgment rule are executed sequentially through the lightweight classification model. When the proportion of weak coverage is greater than the first preset threshold, the first root cause judgment rule is triggered and the weak coverage root cause type is output. When the proportion of weak coverage is not greater than the first preset threshold and the proportion of high load is greater than the second preset threshold, the second root cause judgment rule is triggered and the high load root cause type is output. When the proportion of weak coverage is not greater than the first preset threshold, the proportion of high load is not greater than the second preset threshold, and the interference intensity is greater than the third preset threshold, the third root cause judgment rule is triggered and the interference root cause type is output. The root cause type output by the triggered root cause judgment rule is used as the root cause diagnosis result.

7. The method for identifying and locating poor-quality regions in short video networks according to claim 6, characterized in that, Also includes: Obtain the root cause type from the root cause diagnosis results, wherein the root cause type is: the weak coverage root cause type, the high load root cause type, or the interference root cause type; Based on the preset mapping relationship between root cause types and remediation suggestions, remediation suggestions corresponding to the root cause types are determined; wherein, the weak coverage root cause type corresponds to remediation suggestions for co-construction and sharing or supplementing sites, the high load root cause type corresponds to remediation suggestions for capacity expansion or load balancing, and the interference root cause type corresponds to remediation suggestions for interference investigation or physical cell identifier optimization. The location information of the poor quality hotspot area, the root cause type, the remediation suggestion, and the preset priority identifier are combined to generate an electronic work order, and the electronic work order is pushed to the network optimization system.

8. A device for identifying and locating the root cause of poor-quality areas in short video networks, characterized in that, include: The data acquisition module is used to acquire short video service experience data collected by the top application platform, signaling-level user behavior data collected by the service experience quality platform, and key performance indicator data of wireless network collected by the operation and maintenance center system. The alignment and fusion module is used to perform spatiotemporal alignment of the short video service experience data, the signaling-level user behavior data and the wireless network key performance indicator data of the same user within the same service session period using timestamps and spatial identifiers as association keys, and to generate multiple multi-source heterogeneous fusion records, each of which corresponds to one service session period of a user. The region identification module is used to aggregate the multi-source heterogeneous fusion records according to geographic grids, determine the proportion of poor quality users based on the number of users in each geographic grid that meet the preset poor quality rules and the total number of users, and cluster the geographic grids whose proportion of poor quality users meets the density threshold and whose spatial distance and temporal continuity between them meet the preset threshold to obtain poor quality hotspot areas. The root cause localization module is used to extract feature vectors that characterize the network problem type based on the fusion records contained in the poor quality hotspot area, and input the feature vectors into a lightweight classification model to output the root cause diagnosis results corresponding to the poor quality hotspot area according to a preset priority order.

9. An electronic device, characterized in that, The electronic device includes a processor coupled to a memory, the memory storing at least one computer program, which is loaded and executed by the processor to enable the electronic device to implement the short video network poor quality region identification and root cause localization method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which, when executed by a processor, implements the method for identifying and locating poor-quality areas in short video networks as described in any one of claims 1 to 7.

Images