Audio control application defect clustering and iterative priority generation method, system, storage medium and electronic device

Abstract

The application relates to the technical field of software optimization management, and discloses an audio control application defect clustering and iteration priority generation method, system, storage medium and electronic equipment. The method comprises the following steps: acquiring multi-source user feedback data; processing the user feedback data, extracting a plurality of problem units, clustering the problem units to form candidate defect clusters; evaluating each candidate defect cluster to obtain defect severity information; analyzing each candidate defect cluster to identify the association relationship between defects, versions and devices; calculating the iteration priority of the defects based on the feature information of the candidate defect clusters, and generating defect processing tasks and product iteration suggestions according to the iteration priority. The application can automatically convert user complaints into structured defect tasks and iteration priorities, effectively improves defect identification efficiency and priority judgment accuracy, shortens the product iteration response cycle, and improves the user experience of audio products.

Description

Technical Field

[0001] This invention relates to the field of software optimization and management technology, and in particular to a method, system, storage medium, and electronic device for defect clustering and iterative priority generation in audio control applications. Background Technology

[0002] With the rapid development of smart audio devices, users' demands for audio playback experience are increasing. Mobile applications, as the main control terminal for smart audio devices, undertake core functions such as device addition, playback control, input source switching, multi-room grouping, audio calibration, surround sound configuration, parametric equalization adjustment, local media library browsing, and third-party music service licensing. User reviews in app stores have become an important source of data reflecting the real user experience of audio control applications, directly affecting application ranking, user acquisition costs, and brand reputation.

[0003] Currently, product teams for audio control applications typically review app store reviews, technical forums, customer support tickets, and in-app feedback regularly to identify product defects and areas for improvement. However, negative app store reviews are characterized by diverse language, colloquial expressions, mixed and overlapping issues, lack of contextual information, prominent emotional expressions, and scattered version distribution. For example, users might use different expressions such as "device not found," "constantly spinning," "cannot connect to network," "too many permissions," "playback freezes," or "music service unusable" to point to the same underlying defect. They might also report multiple independent issues such as network configuration failure, playback interruption, and difficulty in interface operation in a single review. Relying solely on manual reading and categorization is not only inefficient but also prone to inconsistencies in categorization standards due to subjective judgment differences, and the omission of low-frequency but serious issues.

[0004] In existing technologies, some review analysis tools can perform simple filtering and statistics of negative reviews based on ratings or keyword matching, but they lack mechanisms for in-depth correlation analysis between negative review texts and application versions, device firmware versions, device models, user behavior tracking data, system logs, and historical repair records. Especially in the field of audio control, defects often have obvious device and version dependencies; that is, the same problem may only occur on specific device models, with specific firmware versions, or under specific network conditions, and existing methods struggle to achieve such refined defect attribution. Furthermore, the actual impact of different defects on user experience varies significantly, making it difficult to accurately determine processing priorities based solely on the number of reviews. For example, privacy issues may cause users to abandon their first use, third-party music service anomalies may affect specific paying user groups, while interface layout problems may only cause minor inconvenience. Existing technologies lack a priority scoring system that comprehensively considers defect frequency, emotional intensity, functional impact, version trends, and user behavior evidence.

[0005] Therefore, there is an urgent need in this field for a technical solution that can automatically identify defect topics, perform semantic clustering, assess the scope of impact, and generate iterative priorities for multi-source user negative reviews of audio control applications, in order to solve the technical problems of low efficiency of manual processing, subjective priority judgment, and inaccurate defect attribution. Summary of the Invention

[0006] The purpose of this invention is to address the technical shortcomings of existing audio control applications, such as inefficiency, inconsistent defect classification standards, and subjective iteration priority judgment, which rely on manual processing of user feedback. This invention provides a method, system, storage medium, and electronic device for defect clustering and iteration priority generation in audio control applications based on user feedback. By constructing a multi-source feedback data acquisition mechanism, semantic clustering and defect topic identification are performed on the negative review text. Combined with sentiment intensity analysis, version attribution, and behavioral tracking, a comprehensive priority scoring model is established, automatically converting user negative reviews into a defect task list and iteration suggestions, forming a data-driven product iteration closed loop.

[0007] On one hand, the present invention provides a method for defect clustering and iterative priority generation in audio control applications, comprising the following steps: Obtain user feedback data from multiple sources; The user feedback data is preprocessed to generate several problem units. Based on the semantic vectors and defect topic dictionaries corresponding to the problem units, the problem units are clustered to form candidate defect clusters. For each of the candidate defect clusters, a severity assessment is performed, and a defect severity score is calculated; Attribution analysis is performed on each of the candidate defect clusters to identify the version correlation and device distribution characteristics of the defects; Based on the feature information of the candidate defect cluster, the defect severity score, the attribution analysis results, and user behavior tracking data, an iteration priority score is calculated for each defect, and a defect task list and product iteration suggestions are generated based on the iteration priority score.

[0008] Furthermore, the acquisition of multi-source user feedback data includes: By using app store open interfaces, exporting data from the backend, or synchronizing authorized data, we collect low-star ratings, user comments, scores, comment times, comment languages, application versions, device system versions, countries / regions, and developer response status from multiple app store platforms. In addition to app store data, the system accesses feedback channels including official websites, community forum posts, customer service tickets, email feedback, in-app feedback, return reason texts, and questionnaire texts. Feedback data from different sources are stored using a unified data field, which includes source channel identifier, feedback text content, timestamp, user anonymity identifier, and associated device information. Collect data fields specific to the audio control application, including but not limited to device model, user type identifier, multiple device ownership status, audio extension device connection status, audio return channel enabled status, audio calibration function enabled status, local media library usage status, and third-party music service binding status.

[0009] Furthermore, the text preprocessing of the user feedback data includes: Perform a cleaning operation on the feedback text to remove meaningless symbols, duplicate characters and non-problem expressions, and retain negative words, degree adverbs, function names and error descriptions; The language type of the feedback text is identified, and machine translation is performed on the non-analytical language text to convert it into the unified analysis language, while preserving the original expression of the technical terms; A single feedback text is split into multiple independent question units based on sentence boundaries and semantic segments, with each question unit corresponding to a single defect topic.

[0010] Preferably, the defect topic dictionary is pre-loaded with defect topics and their synonyms belonging to the domain of the audio control application, wherein, The defects include issues related to permissions and privacy, missing functions, playback control, network connection failure, device discovery failure, local media library anomalies, interface availability, playback interruptions, third-party music service anomalies, application crashes, firmware update failures, short-range wireless communication pairing anomalies, volume control anomalies, audio return channel anomalies, account login anomalies, and multi-device grouping anomalies. The synonym expressions are mapped to the same defect topic through a synonym expression mapping table; The method also includes establishing a functional module mapping table, which maps each defect topic to one or more functional modules of an audio control application.

[0011] Furthermore, clustering the problem units includes: Each of the problem units is converted into a semantic vector using a natural language processing model; Clustering is performed using the semantic vector as the dominant factor, keyword matching, user rating, time window and functional module mapping as auxiliary factors, and application version and device model as constraint factors. Problem units with similar semantics and that meet the constraint factor conditions are aggregated into candidate defect clusters. The constraint factors are used to prevent problem units with version specificity or device specificity from being aggregated across versions or device models, thereby preventing the defect features corresponding to the problem units from being diluted. The cluster name of the candidate defect cluster is automatically generated based on the high-frequency words, representative feedback text, and associated functional modules within the candidate defect cluster.

[0012] Furthermore, the calculation of the defect severity score includes: The problem unit is subjected to emotion intensity recognition, which identifies emotion words, negative expressions, expressions of complaint intensity, and expressions of return intention in the feedback text; The severity of a feature is determined based on how much it affects the user’s core tasks. Issues that cause devices to be unable to connect to the network, play content, crash applications, become unauthorized, or become undiscoverable are classified as high severity, while issues with the interface or text display are classified as low severity. The severity score of a defect is obtained by weighting and fusing the emotional intensity, functional severity, user rating, whether it leads to a return, whether it affects the first-time user experience, and whether it affects the core playback function of the problem unit in the candidate defect cluster.

[0013] Furthermore, the attribution analysis includes: Align the time of negative reviews with the time of application version release, firmware version release, server configuration change, and third-party service interface change to determine whether the defect is related to a specific update. Stratified analysis is conducted based on equipment model, hardware capabilities, and whether a display screen is provided to identify whether defects occur only in specific equipment models or combinations of equipment. Perform layered analysis based on operating system platforms to identify the unique defects of each system platform; Based on the statistical significance of the defect frequency variation, the consistency between the defective equipment distribution and the equipment sales distribution, and the cross-validation results of the embedded data, a confidence level is generated for the attribution analysis.

[0014] Furthermore, the method also includes: While meeting privacy compliance requirements, user feedback is matched with application tracking data, device logs, and cloud status data to identify whether any abnormal events occurred before or after the user feedback. If the defect topic described in the feedback text is consistent with the abnormal event of the tracking point, the corresponding defect will be marked as a high-confidence defect. If a defective topic receives frequent feedback but the tracking data does not capture the corresponding anomaly, a missing tracking information message will be generated.

[0015] Furthermore, the calculation of the iteration priority score for each defect includes: The frequency and proportion of each candidate defect cluster within a specified time window are statistically analyzed, and the frequency changes between the current period and the previous period are compared to identify problems whose growth trend exceeds a preset threshold. Estimate the user scale affected by each defect by combining active user count, device sales, and feature utilization rate; Based on the feature information of the candidate defect cluster, the defect severity score, the attribution analysis results, and user behavior tracking data, multiple priority scoring factors are set, and the iteration priority score is obtained through weighted calculation. The weight configuration of each factor is dynamically adjusted according to the product stage.

[0016] Furthermore, the generation of the defect task list and product iteration suggestions includes: For defects whose iteration priority score is higher than a preset threshold, a defect task card is generated. The defect task card includes the defect name, representative feedback text, scope of impact, associated version, associated device, suspected cause, evidence data, priority level and verification indicators. Based on the defect type, product optimization suggestions are generated. These suggestions include: explanations of permission usage and step-by-step authorization suggestions for privacy-related defects; fault diagnosis and compatibility tips for network connectivity-related defects; hierarchical simplification and operation guidance suggestions for interface usability-related defects; and authorization status detection and re-authentication path suggestions for third-party service-related defects. Test scenarios are automatically generated based on candidate defect clusters. These test scenarios cover specific network environments, specific music services, specific operating system versions, specific combinations of input sources, and specific combinations of multiple devices.

[0017] Furthermore, the method also includes: Defects with iteration priority scores higher than a preset threshold will be associated with the application version, firmware version, or cloud configuration version to be released. For defects that require collaborative repair involving applications, firmware, cloud, and third-party interfaces, record cross-component dependencies and boundaries of responsibility; Compare defects with the product roadmap. If a defect relates to a pre-defined strategic functional direction, increase the strategic priority weight of that defect.

[0018] Furthermore, the method also includes defect iteration effect verification and model update: Before launching defect repairs or product optimizations, set verification indicators, including the rate of decrease in the number of related negative reviews, the change in the proportion of low-star reviews, the rate of decrease in corresponding abnormal tracking points, the increase in task completion rate, the rate of decrease in customer service tickets, and the increase in user retention. After the new version is released, continuously monitor the changing trends of relevant candidate defect clusters, compare the indicator data of the same time window before and after the release, and determine whether the optimization is effective; If the original candidate defect clusters decrease after repair but new semantic expressions or new sub-problems appear, then the synonym expression mapping relationship and clustering model of the defect topic dictionary are updated. It receives manual confirmation, confirms, merges, splits, or renames the clustering results, and feeds back the manually confirmed classification results as training samples to the clustering model.

[0019] On the other hand, the present invention provides a defect clustering and iterative priority generation system for audio control applications, comprising: The data acquisition module obtains user feedback data from multiple sources; The defect clustering module is used to perform text preprocessing on the user feedback data, generate several problem units, and cluster the problem units based on the semantic vectors and defect topic dictionaries corresponding to the problem units to form candidate defect clusters; The severity assessment module is used to assess the severity of each candidate defect cluster and calculate the defect severity score; The attribution analysis module is used to perform attribution analysis on each of the candidate defect clusters to identify the version correlation and device distribution characteristics of the defects. The priority generation module is used to calculate the iteration priority score of each defect based on the feature information of the candidate defect cluster, and generate a defect task list and product iteration suggestions based on the iteration priority score.

[0020] In addition, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the audio control application defect clustering and iterative priority generation method described in any of the preceding claims.

[0021] Meanwhile, an electronic device is provided, comprising: one or more processors; and a storage device for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the audio control application defect clustering and iterative priority generation method described in any of the preceding claims.

[0022] Compared with the prior art, the beneficial effects of the present invention are: (1) By constructing a multi-source user feedback data collection mechanism, the present invention obtains defect data from multiple channels, which significantly expands the scope of defect discovery and reduces the risk of missing problems; (2) This invention identifies defect topics by using semantic vector clustering and a defect topic dictionary, and aggregates different expressions into the same defect cluster, which effectively improves the recall rate and classification consistency of defect identification; (3) This invention calculates the severity score of defects by comprehensively considering multiple factors, which can accurately distinguish between high-impact defects and low-impact defects, thus improving the accuracy of severity judgment; (4) This invention identifies the version dependency and device distribution characteristics of defects through attribution analysis, providing a precise location direction; (5) This invention calculates the iteration priority score by comprehensively considering the frequency, severity, attribution analysis and embedded data of defects, and automatically generates a list of defect tasks, thereby improving the objectivity and accuracy of priority ranking and shortening the product iteration response cycle. Attached Figure Description

[0023] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a defect clustering and iterative priority generation method for an audio control application according to the present invention; Figure 2 This is a block diagram of a defect clustering and iterative priority generation system for an audio control application according to the present invention. Figure 3 This is a schematic diagram of an embodiment of an electronic device according to the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] The specific embodiments of the present invention will be described below with reference to the accompanying drawings and examples.

[0026] Example 1

[0027] Please see Figure 1 The technical solution for a defect clustering and iterative priority generation method for an audio control application provided in this embodiment includes the following steps: S1: Obtain user feedback data from multiple sources; S2: Perform text preprocessing on the user feedback data to generate several problem units, and cluster the problem units based on the semantic vectors and defect topic dictionaries corresponding to the problem units to form candidate defect clusters; S3: Perform a severity assessment on each candidate defect cluster and calculate the defect severity score; S4: Perform attribution analysis on each candidate defect cluster to identify the version correlation and device distribution characteristics of the defects; S5: Based on the feature information of the candidate defect cluster, the defect severity score, the attribution analysis results, and user behavior tracking data, calculate the iteration priority score for each defect, and generate a defect task list and product iteration suggestions based on the iteration priority score.

[0028] The first step is the collection of user feedback data in step S1. This includes acquiring multi-source user feedback data: S11: Collect low-star ratings, user comments, scores, comment time, comment language, application version, device system version, country / region, and developer response status from multiple app store platforms through app store open interfaces, backend data export, or authorized data synchronization. S12: In addition to app store data, access includes feedback channels from official websites, community forum posts, customer service tickets, email feedback, in-app feedback, return reason text, and questionnaire text; store feedback data from different sources according to a unified data field, which includes source channel identifier, feedback text content, timestamp, user anonymity identifier, and associated device information; S13: Collect data fields specific to the audio control application, including but not limited to device model, user type identifier, multi-device ownership status, audio extension device connection status, audio return channel enabled status, audio calibration function enabled status, local media library usage status, and third-party music service binding status.

[0029] Specifically, when accessing app store review data, the two lowest ratings (e.g., 1 star and 2 stars) are prioritized for analysis, while negative comments from higher ratings (e.g., 4 stars and 5 stars) are retained to supplement problem identification. This tiered data collection strategy ensures the completeness of defect identification.

[0030] In addition to app store data, data from other sources is also integrated, necessitating field standardization. This includes associated device information, associated order information, or associated customer service tracking numbers. All timestamps from all sources are converted to Coordinated Universal Time (UTC) and linked to application and firmware release calendars to establish a mapping between "feedback time" and "version cycle."

[0031] Because audio control applications differ from ordinary utility applications, the system additionally collects fields such as device model, user type identifier (e.g., whether it is a new user), multi-device ownership status, audio extension device connection status (e.g., subwoofer connection status), audio return channel enable status (whether HDMI ARC is used), audio calibration function enabled status (whether RoomFit is enabled), local media library usage status, third-party music service binding status, and whether it is in the first-use bootstrapping phase (OOBE phase).

[0032] For example, a user gives the lowest rating on an app distribution platform and comments: "After the update, my device cannot find my speaker and Music Service A keeps logging out." After collecting this comment, the system records the app version, operating system version, country / region, and comment time, and temporarily stores the text as a negative review for analysis. If the backend can match anonymous event tracking data, it further correlates this with recent device discovery failures and Music Service A authorization failures.

[0033] Next, preprocessing in step S2 is performed. The original feedback text contains a large amount of noise and unstructured expressions, and direct semantic analysis will seriously affect the accuracy of the results. This step transforms the original noisy text into a sequence of clearly structured and semantically simple question units through hierarchical cleaning, language merging, and semantic segmentation. The text preprocessing of the user feedback data includes: S21: Perform a cleaning operation on the feedback text, removing meaningless symbols, repeated characters, and non-problem expressions, while retaining negative words, degree adverbs, function names, and error descriptions. For example, clean "can't connect!!! keeps spinning forever" to "cannot connect keeps spinning"; S22: Identify the language type of the feedback text, perform machine translation on non-analyzed language text to convert it into a unified analysis language, and preserve the original expressions of proprietary technical terms. During the translation process, a pre-set protected vocabulary is used to ensure that proprietary technical terms are not translated, including wireless network communication protocol names, wireless audio transmission protocol names, streaming media service names, network storage protocol names, audio return channel names, audio calibration function names, and parametric equalization adjustment function names, etc. S23: Break down a single feedback text into multiple independent problem units based on sentence boundaries and semantic segments, with each problem unit corresponding to a single defect topic. For example, “App crashes and I cannot connect to my NASlibrary” is broken down into two problem units: “Application crashes” and “Network storage local library cannot be connected”.

[0034] In addition, the system filters units that are too short and do not contain effective keywords, units that contain only emotional expressions without problem descriptions, such as "I hate this", and units that contain only positive expressions without negative words, to ensure that all problem units entering the subsequent clustering process contain valuable defect signals.

[0035] In a certain scenario, a user commented, "The interface is too complex, Room Correction cannot be found, and Bluetooth frequently disconnects." The system breaks this down into three problem units: complex interface, undiscoverable Room Correction entry point, and unstable short-range wireless communication connection. Each problem unit then proceeds to the subsequent clustering steps.

[0036] Based on the problem units generated above, as in step S2, clustering is performed. First, a defect topic dictionary and functional module mapping table for the audio control application needs to be established. The defect topic dictionary pre-contains defect topics and their synonyms within the domain of the audio control application. The defects include, but are not limited to, privacy and access control issues, missing functions, playback control issues, network connection failures, device discovery failures, local media library anomalies, interface availability issues, playback interruptions, third-party music service anomalies, application crashes, firmware update failures, short-range wireless communication pairing anomalies, volume control anomalies, audio return channel anomalies, account login anomalies, and multi-device grouping anomalies. The synonym expressions are mapped to the same defect topic through a synonym expression mapping table. Specifically, the mapping relationship adopts a three-layer structure: an exact matching layer (direct mapping of identical expressions), a fuzzy matching layer (mapping through edit distance for keyword variants), and a semantic matching layer (mapping through word vectors for keywords that are different but semantically similar). For example, "cannotfind device", "device disappeared", "not detected", "device not found", and "device lost" are all mapped to device discovery failure. The method also includes establishing a functional module mapping table, which maps each defect topic to one or more functional modules of an audio control application. For example, the audio return channel corresponds to the input source management, consumer electronics control protocol, TV audio link, and device status feedback modules; the local media library corresponds to the Universal Serial Bus storage, network-attached storage, Digital Life Network Alliance protocol, server message block protocol, indexing service, access control, and media scanning modules.

[0037] In addition, for audio-specific function names (such as "RoomFit (Room Adaptive Calibration)" and "PEQ (Parametric Equalizer)"), the system establishes a detailed synonym mapping table for everyday user expressions to ensure a smooth semantic bridge between technical terms and everyday user expressions.

[0038] Based on this, each problem unit is converted into a semantic vector and clustered. Traditional keyword matching methods cannot solve the problem of synonymous heterogeneity; this step addresses this by combining semantic vectorization and unsupervised clustering. Clustering includes: S24: Each question unit is converted into a semantic vector using a pre-trained natural language processing model. A domain-adaptive fine-tuning strategy is adopted, and the general language model is incrementally trained using user comment corpora of audio products to enable the model to better understand domain-specific expressions. S25: Using the semantic vector as the primary factor, keyword matching, user rating, time window, and functional module mapping as auxiliary factors, and application version and device model as constraint factors, clustering is performed. Problem units with similar semantics and meeting the constraint factors are aggregated into candidate defect clusters. These constraint factors prevent version-specific or device-specific problem units from being aggregated across versions or device models, thus preventing the dilution of defect features corresponding to the problem units. After clustering, the system calculates intra-cluster cohesion and inter-cluster separation for quality assessment. The system adopts an adaptive clustering number determination strategy, performing clustering within a preset range (e.g., 5 to 50) and selecting the cluster number with the optimal silhouette coefficient. S26: Based on the high-frequency words, representative feedback text, and associated functional modules within the candidate defect clusters, automatically generate the cluster name for each candidate defect cluster. In this embodiment, a naming pattern of "qualifying words + core theme" is adopted, such as "Android platform - Device discovery failed".

[0039] For example, user comments may contain expressions such as "NAS empty folder", "DLNA cannot browse", "USB library no songs", and "SMB share shows blank". The system uses semantic vector clustering to aggregate these into a defect cluster of "local media library cannot be browsed or content is empty", and further performs sub-clustering according to Network Attached Storage, Digital Life Network Alliance Protocol, Universal Serial Bus Storage, and Server Message Block Protocol.

[0040] Next, step S3 calculates the defect severity score. Defect severity is not solely determined by frequency of occurrence. A defect that occurs 10 times but leads to customer returns may have a far greater business impact than a defect that occurs 100 times but only causes minor inconvenience. This step achieves precise quantification of defect impact through multi-dimensional severity assessment, including: S31: Perform emotion intensity recognition on the problem unit, and identify emotion words, negative expressions, complaint intensity expressions, and return intention expressions in the feedback text. A method combining dictionary matching and semantic analysis is adopted: (1) Use an emotion dictionary to match emotion words, and calculate the basic emotion score based on the polarity and intensity value of the words; (2) Use semantic analysis to identify implicit emotion expressions, such as "after spending $1000 on this", which, although not containing explicit emotion words, implies the gap between high expectations and low satisfaction, corresponding to a higher emotion intensity. In particular, the system assigns the highest emotion intensity weight to return intention expressions (such as "returning", "going back"). S32: Determine the severity of the function based on the degree of impact of the defect on the user's core tasks. Problems that cause the device to be unable to connect to the network, unable to play, application to crash, account to be unable to be authorized, or device to be undiscoverable are of high severity level, while interface or text display problems are of low severity level. Specifically, in this embodiment, the severity of the function is determined based on the degree of impact of the defect on the user's core tasks: (1) Fatal level: causes the device to be unable to complete the first use guide, unable to connect to the network, unable to play, application to crash, account to be unable to be authorized, or device to be undiscoverable; (2) Severe level: specific functions are unavailable but there are alternative paths, playback is interrupted in specific scenarios; (3) General level: operation steps are increased, interface feedback is unclear but does not affect the completion of the function; (4) Minor level: unclear interface text, visual inconsistency, non-core page layout problems; S33: The emotional intensity, functional severity, user rating, whether it leads to a return, whether it affects the first-time user experience, and whether it affects the core playback function of the problem unit in the candidate defect cluster are weighted and fused to obtain the defect severity score. Here, emotional intensity is mapped to an emotional dimension score, functional severity level is mapped to a functional dimension score, and user rating is mapped to a rating dimension score according to a threshold (the lowest rating corresponds to the highest score). The scores of each dimension are normalized and then weighted and summed according to preset weights. Functional severity has a higher weight than emotional intensity because the functional impact is objectively verifiable; whether it affects first-time use has a higher weight than non-first-time use.

[0041] In a certain scenario, both defects appeared 30 times. Defect A, "the button color is not nice," received mostly 3-star and 4-star reviews, while defect B, "network configuration failure causing unusable status," received mostly 1-star reviews and included expressions of wanting to return the product. The system assigned lower scores to each dimension of defect A, while assigning the highest emotional weight, the most critical functional severity weight, and the lowest rating weight to defect B's return intention. Ultimately, defect B's severity score was significantly higher than defect A's.

[0042] After calculating the defect severity score, step S4, attribution analysis, is performed to analyze each candidate defect cluster. The attribution analysis includes: S41: Align the time of negative reviews with the time of application version release, firmware version release, server configuration change, and third-party service interface change to determine whether the defect is related to a specific update. S42: Perform stratified analysis based on equipment model, hardware capabilities, and whether a display screen is provided to identify whether defects occur only in specific equipment models or combinations of equipment. S43: Perform layered analysis by operating system platform to distinguish the unique defects of each system platform; S44: Based on the statistical significance of the change in defect frequency, the consistency between the distribution of defective equipment and the distribution of equipment sales, and the cross-validation results of the embedded data, a confidence level is generated for the attribution analysis.

[0043] First, there's the fine-grained alignment and abrupt change detection of the version timeline. The occurrence time of negative reviews is aligned with the application version, firmware version, server-side configuration, and third-party service interface change times. A time window correlation method is employed, including setting multiple time windows (n days before release, release day, n days after release, and m days after release) based on the version release date, and counting the frequency of the target defect within each window. If the frequency of a defect in the post-release window is significantly higher than in the pre-release window (e.g., the growth rate exceeds a preset threshold), it is marked as highly relevant to the version.

[0044] Secondly, there is the stratified attribution based on device model. This involves stratified analysis based on device model, hardware capabilities, and whether a display screen is provided. A "device model-hardware capability" mapping table is established to identify whether the defect only occurs in a subset of devices with specific hardware capabilities. For example, audio return channel problems only occur on devices with audio return channel input capabilities.

[0045] Secondly, there's the issue of system platform differentiation. This involves a tiered analysis based on operating system platforms. Different platforms have different permission management mechanisms, network access policies, and background operation restrictions. For example, local network permission issues on one mobile operating system platform and Bluetooth scanning permission issues on another mobile operating system platform belong to different platform mechanisms and should not be lumped together.

[0046] In addition, attribution confidence is quantified. A confidence score is generated for each attribution conclusion. Based on the statistical significance of the change in defect frequency, the consistency between the distribution of defective equipment and the distribution of equipment sales (if a certain model has a low sales share but an abnormally high defect share, the confidence is high), and the cross-validation results of the embedded data, the confidence is divided into three levels: high, medium, and low.

[0047] The attribution confidence score is obtained by weighted fusion of three dimensions, including: S441: Calculate the statistical significance score for time. This calculates the frequency of occurrence of the target defect within a time window following the version release. Frequency of occurrence within the same time window as before release Calculate the ratio of the difference between the two to the benchmark volatility level, and truncate the result to the interval between 0 and 1 to obtain the time statistical significance score. : , This score reflects whether the change in the frequency of defects before and after the release of a version is statistically significant. The higher the score, the stronger the correlation between the defect and the version update. S442: Calculate the equipment distribution deviation score. The system statistically analyzes the target defects in equipment models. The proportion of occurrence above And the sales percentage of this device model among all active devices. Calculate the relative deviation between the two: , The denominator is taken as the larger of the two values ​​to avoid division by zero. This score reflects whether the distribution of defects across different equipment models deviates from the normal sales distribution. The higher the score, the more concentrated the defects are on a specific equipment model, and the higher the confidence level that the defects are attributable to equipment characteristics. S443: Calculate the text-event matching score. The system matches the defect topic described in the negative review text with the abnormal event types recorded in the event tracking data to obtain a matching score. : , in The value ranges from 0 to 1 and is obtained by weighting the text keyword hit rate and the overlap rate of the abnormal event type. This score reflects the consistency between the user's subjective description of the defect and the objective evidence from the tracking points. The higher the score, the higher the reproducibility and credibility of the defect. S444: Calculate the overall confidence score and map it to a confidence level. The system weights and merges the three scores according to preset weights to obtain the overall attribution confidence score. : , in, , and The preset weight coefficients correspond to the weights of three dimensions: time statistical significance, device distribution deviation, and text-tracking point matching degree. The system will consider the overall confidence score. Mapped to three confidence levels: when A confidence level greater than a first preset threshold (e.g., 0.7) is considered high. A confidence level between the second preset threshold and the first preset threshold (e.g., 0.4 to 0.7) is considered medium confidence. A confidence level below the second preset threshold (e.g., 0.4) is considered low. High-confidence attribution conclusions directly trigger task generation, medium-confidence conclusions are marked as "to be verified" and enter the task pool, while low-confidence conclusions are only used as prompts for manual reference.

[0048] Taking the defect cluster "Device Model M - No sound in audio feedback channel" as an example: the feedback frequency was 5 times / day in the 7 days before release, and 12 times / day after release; in this defect cluster, model M accounted for 45% of feedback, and model M's sales accounted for 8%; the cross-validation match between feedback text and event tracking points was 85%. (Weighting is then applied.) , , (Deviation of device distribution is given the highest weight, as device-specific identification is the core innovation of this solution, distinguishing it from existing technologies that "judge solely by the number of comments"). Calculation time statistical significance score. Equipment distribution deviation score Text-tracking matching score Finally, the overall confidence level is calculated. ,final If the cluster is determined to have a high confidence level, the system will automatically mark it as a high-priority review target.

[0049] As can be seen, the above algorithm transforms confidence level from a qualitative description into an automatically executable quantitative judgment method, providing clear technical support for generating confidence level ratings. The three evidence dimensions—time, device, and behavior—are cross-validated from three independent sources, compensating for the limitations of a single criterion. In particular, the device distribution deviation factor effectively identifies hidden defects concentrated in specific models despite a small number of reviews, providing calculable judgment criteria for the original instruction manual examples. Weighting coefficients... , , It supports dynamic configuration based on product stage, maintaining consistency with the weight adjustment mechanism of the aforementioned priority scoring model, and reserves interfaces for future expansion to include evidence dimensions such as geographic distribution and user level. The comprehensive confidence score directly connects to the triggering conditions for manual review, transforming the priority review of low-confidence clustering results from a qualitative description into an executable quantitative decision rule, forming a complete methodological chain from evidence collection to review decision-making. Overall, this algorithm improves the system's automation level, attribution accuracy, and iterative response speed without increasing manual input.

[0050] In addition to calculating the defect severity score mentioned in step S3 and performing the attribution analysis mentioned in step S4, it is also necessary to enhance evidence by correlating user behavior tracking data and system logs. User feedback is a subjective expression and may contain exaggerations or misunderstandings. By cross-validating subjective text with objective tracking data, the system can improve the credibility of defects. Specifically, this includes: Under the premise of meeting privacy compliance requirements, user feedback is matched with application tracking data, device logs and cloud status data to reconstruct the behavioral path sequence within the time window before the user comments, such as what operations the user experienced, which pages the user stayed on abnormally, which function calls were triggered, and whether timeouts or other anomalies were encountered, so as to identify whether any related abnormal events occurred before and after the user feedback. At the same time, not only tracking data is matched, but also device logs and cloud logs are associated to build a chain of evidence. For example, for "cannot play", the following are checked at the same time: (1) whether the playback click in the tracking data is reported; (2) whether the device logs receive the playback instruction; (3) whether the cloud logs have expired authorization or regional restrictions and other rejection responses, so as to locate the problem to the specific technical link. If the defect described in the feedback text is consistent with the abnormal event of the tracking point, the corresponding defect will be marked as a high-confidence defect. Specifically, the degree of matching between the text and the tracking point is classified into the following categories: (1) High consistency (matching degree > 80%), judged as high confidence, and priority score is increased; (2) Partial matching (40%-80%), judged as to be verified, and log analysis or manual review is triggered; (3) Frequent feedback but no abnormality of the tracking point (matching degree < 40%), judged as low confidence, and tracking point coverage check is triggered at the same time. If a defective topic receives frequent feedback (e.g., more than 100 reports) but the tracking data does not capture the corresponding anomaly, a missing tracking message will be generated.

[0051] After implementing the above solution, if a group of users comments that "the wireless audio transmission protocol connection is not working," the system's tracking data reveals that these users clicked multiple times on the connection page but received no playback status callback, and the server logs show that the authorization token refresh failed. The system then elevates this defect from a common complaint to a reproducible service authorization issue.

[0052] Then, step S5 involves iterative priority scoring for each defect. The impact of a defect depends not only on the absolute number of comments but also on the size and structure of the user group represented by those comments. Specifically, this includes: S51: Statistically analyze the frequency and proportion of each candidate defect cluster within a specified time window, compare the frequency changes between the current period and the previous period, and identify problems where the growth trend exceeds a preset threshold. S52: Estimate the user scale affected by each defect by combining active user count, device sales and feature utilization rate; S53: Based on the feature information of the candidate defect cluster, the defect severity score, the attribution analysis results and user behavior tracking data, set multiple priority scoring factors, obtain the iteration priority score through weighted calculation, and dynamically adjust the weight configuration of each factor according to the product stage.

[0053] In this embodiment, the frequency and percentage of occurrence of each defect cluster within a specified time window, version, region, and platform are statistically analyzed. To avoid bias caused by differences in channel activity, normalization is performed: the frequency of defect occurrences in each channel is divided by the total number of feedbacks in that channel during the same period to obtain the relative percentage, and then a weighted average is calculated according to channel weight to obtain the comprehensive impact score.

[0054] Based on this, the frequency changes of the current period and the previous period are compared. A sliding window comparison method is used, dividing the time axis into continuous windows, calculating the frequency of each defect within each window, and comparing the change rates of adjacent windows. When a defect shows positive growth across multiple consecutive windows and the cumulative growth rate exceeds a preset threshold (e.g., 50%), it is marked as a "rapidly rising problem". If a single window experiences a sudden increase (e.g., more than 3 times the growth of the previous window) and coincides with a version release time, it is marked as strongly associated with a version change.

[0055] The impact on user scale is estimated by combining the number of active users, feature usage rate, and comment coverage. The base value is the number of unique users mentioning the defect in comments, divided by the usage rate of the feature module affected by the defect (obtained from event tracking), and then multiplied by the total number of active users in that user group. That is, the impact on user scale is approximately equal to the ratio of the number of unique users mentioning the defect to the usage rate of that feature module. For example, if 100 unique users commented on music service A, and event tracking shows that only 5% of active users used the service, then the estimated impact on user scale is 100 ÷ 5% = 2000 people, which is more objective than simply relying on the number of comments.

[0056] Based on the feature information of candidate defect clusters, the iterative priority score of each defect is calculated comprehensively. Priority determination is the step most susceptible to subjective experience. Step S53 achieves data-driven decision-making through a quantifiable multi-factor scoring model.

[0057] In this embodiment, multiple scoring factors include: frequency of occurrence (normalized comprehensive impact score); growth rate; severity score; emotional intensity score; degree of impact on core functionality (0-1 continuous value); degree of impact on first-time use (Boolean value); degree of impact on high-value users; degree of impact on multi-device systems; return relevance; repair difficulty; credibility of reproducibility evidence; and strategic feature importance. Differentiated weights are set according to different product stages. During the new product launch stage (the first quarter after release), the weights of factors affecting first-time use and device discovery are increased to 1.5 times the normal weights, as the first experience determines user retention. During the feature expansion stage, the weights of defects related to new features are increased. During the maturity optimization stage, the weights of experience and interface-related factors are relatively increased. Weights are stored in configuration files and can be adjusted by the product manager.

[0058] Next, after normalizing each factor value to a uniform range, a weighted sum is calculated to obtain the comprehensive priority score, expressed as: , in, These are the normalized values ​​for each rating factor. For the corresponding weights, and .each The normalization method is as follows: for continuous factors (such as frequency and growth rate), minimum-maximum normalization is used; for categorical factors (such as whether it affects the first use), binary processing of 0 or 1 is used.

[0059] Defects are mapped to different levels based on the comprehensive score: scores above the first threshold (e.g., 0.8) are first priority (blocking core use or causing large-scale crashes, requiring a response within 24 hours); scores between the first and second thresholds (e.g., 0.6-0.8) are second priority (high-frequency, serious issues, to be resolved in the current iteration); scores between the second and third thresholds (e.g., 0.4-0.6) are third priority (there is an alternative path, to be addressed in subsequent iterations); and scores below the third threshold are fourth priority (low-frequency, minor issues, to be addressed depending on resource availability).

[0060] Secondly, the generation of the defect task list and product iteration suggestions in step S5 includes: S54: Generate defect task cards for defects whose iteration priority scores are higher than a preset threshold. The defect task card includes the defect name (derived from the cluster name), representative feedback text (comments with moderate sentiment and clear description within the cluster), scope of impact (estimated by combining frequency and user scale), associated versions (suspicious versions identified by attribution analysis), associated devices (a list of high-risk models), suspected causes (a list of initial hypotheses with confidence ranking), evidence data (representative comments, anomaly statistics of data points, log summaries), priority level, suggested responsible person (automatically matched according to functional modules), and verification indicators (automatically generated according to the defect type). S55: Generate product optimization suggestions based on the defect type. The product optimization suggestions include: permission usage descriptions and step-by-step authorization suggestions for permission and privacy defects; fault diagnosis and compatibility prompts for network connectivity defects; hierarchical simplification and operation guidance suggestions for interface usability defects; and authorization status detection and re-authentication path suggestions for third-party service defects. S56: Automatically generate test scenarios based on candidate defect clusters. These test scenarios cover specific network environments, specific music services, specific operating system versions, specific input source combinations, and specific multi-device combinations. The system extracts key variables (network environment, music service, operating system version, input source type, device combination) from the defect clusters and uses a combination testing method to generate test cases covering different combinations.

[0061] In one embodiment, after the defect cluster "local media library content is empty" generates a task card, the system suggests testing four types of entry points: Universal Serial Bus storage, Network Attached Storage, Digital Life Network Alliance Protocol, and Server Message Block Protocol, covering scenarios such as permission denial, indexing failure, format not supported, and authentication failure.

[0062] In addition, the method also includes: Defects with iteration priority scores higher than a preset threshold will be associated with the application version, firmware version, or cloud configuration version to be released. For defects that require collaborative repair involving applications, firmware, cloud, and third-party interfaces, record cross-component dependencies and boundaries of responsibility; Compare defects with the product roadmap. If a defect relates to a pre-defined strategic functional direction, increase the strategic priority weight of that defect.

[0063] This method links defects to product planning. Defect remediation depends not only on the defect's impact but also on the product's strategic direction. Low-frequency defects that affect core differentiating selling points should have higher priority than high-frequency but marginal defects. First and second priority defects are automatically linked to the current development cycle version; third priority defects are added to the subsequent version's backlog based on remediation difficulty and resource availability; fourth priority defects are added to the long-term observation list. The "Associated Functional Modules" and "Historical Remediation Plans" fields in the defect topic dictionary determine whether cross-component remediation is involved, presented in the "Dependencies" field in the task card for easy coordination of cross-team resources. Simultaneously, defects are compared with the product roadmap; if they affect a preset strategic functional direction, their strategic weight is increased. A set of functional modules is associated with each strategic direction. The system calculates the correlation between the modules affected by the defect and each strategic direction; defects with a strategic weight factor exceeding a threshold are set to greater than 1.

[0064] The method of this invention also includes defect iteration effect verification and model updating. The system monitors the change trend after defect repair, forming a complete closed loop, specifically including: Before launching defect repairs or product optimizations, set verification indicators, including the rate of decrease in the number of related negative reviews, the change in the proportion of low-star reviews, the rate of decrease in corresponding abnormal tracking points, the increase in task completion rate, the rate of decrease in customer service tickets, and the increase in user retention. After the new version is released, continuously monitor the changing trends of relevant candidate defect clusters, compare the indicator data of the same time window before and after the release, and determine whether the optimization is effective; If the original candidate defect clusters decrease after repair but new semantic expressions or new sub-problems appear, then the synonym expression mapping relationship and clustering model of the defect topic dictionary are updated. It receives manual confirmation, confirms, merges, splits, or renames the clustering results, and feeds back the manually confirmed classification results as training samples to the clustering model.

[0065] Ideally, during validation, we set up multi-dimensional validation metrics, including performance metrics and regression metrics. Performance metrics include: the rate of change in the number of related negative reviews, the change in the percentage of the lowest rating, the change in the occurrence rate of abnormal event tracking (such as permission denial rate), and the change in task completion rate (such as the initial guidance completion rate). Regression metrics include: the rate of change in negative reviews of adjacent modules, the change in the overall average rating, and the change in crash rate. Baseline data is collected before launch, and comparative data is collected within the same time window after launch. Among them, in the short term (3 days after release), the focus is on regression metrics; in the medium term (7 days), the focus is on performance metrics, calculating the rate of change and conducting statistical significance tests; in the long term (30 days), the focus is on trend persistence and the emergence of new sub-problems.

[0066] If the original defects are significantly reduced, the repair plan, associated versions, and verification results will be entered into the historical repair record database. If new expressions or sub-issues appear, they will be included in the synonym expression mapping table or added as a sub-category. If the overall negative review rate of a module decreases significantly, its baseline quality score will be increased, thereby relatively improving the severity rating of similar issues in that module. In addition, if the initial repair effect is not as expected (e.g., the decrease in negative reviews is less than a preset threshold, such as 30%), the system will automatically re-cluster the remaining related negative reviews, generate a new task card, and increase its priority level.

[0067] For example, after adding permission instructions, negative reviews related to "privacy concern" decreased, but "local network permission is still confusing" remained. The system clustered the remaining comments into a "insufficient explanation of local network permissions" defect cluster, prompting the product team to add more illustrated explanations to the key steps of the initial onboarding.

[0068] In addition, the system still provides a manual verification mechanism to ensure the accuracy of the analysis while maintaining automation efficiency, specifically including: Prioritize reviewing the clustering results of high-priority defect clusters (first and second priorities) and low-confidence labels, and provide a batch operation interface to control the review workload to 5%-10% of the total clustering results; The manual correction operation is converted into a training signal, including: Confirmation: Enhance clustering confidence; Merging: Establish equivalence relations, adjust similarity thresholds, or supplement synonym mappings; Splitting: Label distinguishing features and adjust clustering sensitivity; Renaming: Supplement the naming rule keyword library to continuously reduce subsequent clustering error rates; The product view focuses on priorities and user scenarios; the R&D view focuses on reproduction paths and logs; the testing view focuses on test scenarios and execution status; the customer service view focuses on original comments and scripts; and the operations view focuses on trends and reputation comparisons.

[0069] In optimizing user feedback analysis according to this invention, anonymization and desensitization processes are used to protect the privacy and ensure compliance of user data from various sources, specifically including: Data de-identification. User accounts are generated using one-way hashing to produce irreversible anonymous identifiers; email addresses, order numbers, and device serial numbers are de-identified; custom fields such as home network names are identified and replaced with de-identified tags. Minimize data collection and mask content. Do not collect music playback content, Wi-Fi passwords, or precise IP addresses. Personal information in comment text is automatically identified and masked during database entry. Access control and permissions. Access is controlled by role: customer service staff can view comments and task statuses, developers can view technical logs, and product staff can view clusters and priority summaries.

[0070] In summary, this invention integrates multi-source data, including app store data, official feedback, and customer service tickets, to clean and segment negative review texts, perform semantic vectorization clustering, and identify defect topics. It then calculates a defect severity score by combining emotional intensity, functional severity, and user ratings. Based on this, the invention performs attribution analysis on defect clusters with application version, firmware version, and device model. Combining frequency of occurrence, growth trends, user behavior tracking evidence, and product strategy direction, it generates iteration priorities through a multi-factor weighted scoring model, automatically outputting defect task cards and product optimization suggestions. This invention effectively improves defect identification efficiency and priority judgment accuracy, shortens product iteration response cycles, and enhances the user experience of audio products.

[0071] Based on the above methods, this invention provides a defect clustering and iterative priority generation system for audio control applications, such as... Figure 2 As shown, it includes: Data acquisition module 10 acquires user feedback data from multiple sources; The defect clustering module 20 is used to perform text preprocessing on the user feedback data, generate several problem units, and cluster the problem units based on the semantic vectors and defect topic dictionaries corresponding to the problem units to form candidate defect clusters; The severity assessment module 30 is used to assess the severity of each candidate defect cluster and calculate the defect severity score; The attribution analysis module 40 is used to perform attribution analysis on each of the candidate defect clusters to identify the version correlation and device distribution characteristics of the defects. The priority generation module 50 is used to calculate the iteration priority score of each defect based on the feature information of the candidate defect cluster, and generate a defect task list and product iteration suggestions based on the iteration priority score.

[0072] It should be noted that the steps in the audio control application defect clustering and iteration priority generation method provided in this embodiment can be implemented based on the corresponding modules in the audio control application defect clustering and iteration priority generation system. Those skilled in the art can refer to the technical solution of the system to implement the steps of the method. That is, the embodiments in the system can be understood as preferred examples of implementing the method, and will not be elaborated here.

[0073] This embodiment also provides an electronic device, such as... Figure 3 As shown, the electronic device includes a processor 14 and a memory 13. The memory 13 stores machine-executable instructions that can be executed by the processor 14. The processor 14 executes the machine-executable instructions to implement the above-described audio control application defect clustering and iterative priority generation method.

[0074] Furthermore, Figure 3 The electronic device shown also includes a bus 12 and a communication interface 11, with the processor 14, the communication interface 11 and the memory 13 connected via the bus 12.

[0075] The memory 13 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage device. Communication between this system network element and at least one other network element is achieved through at least one communication interface 11 (which can be wired or wireless), such as the Internet, wide area network, local area network, metropolitan area network, etc. The bus 12 may be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 3 The symbol is represented by a single double-headed arrow, but this does not mean that there is only one bus or one type of bus.

[0076] Processor 14 may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by the integrated logic circuitry in the hardware of processor 14 or by instructions in software form. Processor 14 can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in this embodiment. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this embodiment can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory 13. The processor 14 reads the information in memory 13 and, in conjunction with its hardware, completes the steps of the audio control application defect clustering and iterative priority generation method.

[0077] This disclosure also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, storing a computer program that, when run on a computer, causes the computer to perform the steps of an audio control application defect clustering and iterative priority generation method.

[0078] Finally, it should be noted that the above description is only a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be pointed out that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

[0079] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for defect clustering and iterative priority generation in an audio control application, characterized in that, Includes the following steps: Obtain user feedback data from multiple sources; The user feedback data is preprocessed to generate several problem units. Based on the semantic vectors and defect topic dictionaries corresponding to the problem units, the problem units are clustered to form candidate defect clusters. For each of the candidate defect clusters, a severity assessment is performed, and a defect severity score is calculated; Attribution analysis is performed on each of the candidate defect clusters to identify the version correlation and device distribution characteristics of the defects; Based on the feature information of the candidate defect cluster, the defect severity score, the attribution analysis results, and user behavior tracking data, an iteration priority score is calculated for each defect, and a defect task list and product iteration suggestions are generated based on the iteration priority score.

2. The method for defect clustering and iterative priority generation in audio control applications according to claim 1, characterized in that, The acquisition of multi-source user feedback data includes: By using app store open interfaces, exporting data from the backend, or synchronizing authorized data, we collect low-star ratings, user comments, scores, comment times, comment languages, application versions, device system versions, countries / regions, and developer response status from multiple app store platforms. In addition to app store data, the system accesses feedback channels including official websites, community forum posts, customer service tickets, email feedback, in-app feedback, return reason texts, and questionnaire texts. Feedback data from different sources are stored using a unified data field, which includes source channel identifier, feedback text content, timestamp, user anonymity identifier, and associated device information. Collect data fields specific to the audio control application, including but not limited to device model, user type identifier, multiple device ownership status, audio extension device connection status, audio return channel enabled status, audio calibration function enabled status, local media library usage status, and third-party music service binding status.

3. The method for defect clustering and iterative priority generation in audio control applications according to claim 1, characterized in that, Text preprocessing of the user feedback data includes: Perform a cleaning operation on the feedback text to remove meaningless symbols, duplicate characters and non-problem expressions, and retain negative words, degree adverbs, function names and error descriptions; The language type of the feedback text is identified, and machine translation is performed on the non-analytical language text to convert it into the unified analysis language, while preserving the original expression of the technical terms; A single feedback text is split into multiple independent question units based on sentence boundaries and semantic segments, with each question unit corresponding to a single defect topic.

4. The method for defect clustering and iterative priority generation in audio control applications according to claim 1, characterized in that, The defect topic dictionary pre-loads defect topics and their synonyms for the audio control application's domain, wherein... The defects include issues related to permissions and privacy, missing functions, playback control, network connection failure, device discovery failure, local media library anomalies, interface availability, playback interruptions, third-party music service anomalies, application crashes, firmware update failures, short-range wireless communication pairing anomalies, volume control anomalies, audio return channel anomalies, account login anomalies, and multi-device grouping anomalies. The synonym expressions are mapped to the same defect topic through a synonym expression mapping table; The method also includes establishing a functional module mapping table, which maps each defect topic to one or more functional modules of an audio control application.

5. The method for defect clustering and iterative priority generation in audio control applications according to claim 4, characterized in that, Clustering the problem units includes: Each of the problem units is converted into a semantic vector using a natural language processing model; Clustering is performed using the semantic vector as the dominant factor, keyword matching, user rating, time window and functional module mapping as auxiliary factors, and application version and device model as constraint factors. Problem units with similar semantics and that meet the constraint factor conditions are aggregated into candidate defect clusters. The constraint factors are used to prevent problem units with version specificity or device specificity from being aggregated across versions or device models, thereby preventing the defect features corresponding to the problem units from being diluted. The cluster name of the candidate defect cluster is automatically generated based on the high-frequency words, representative feedback text, and associated functional modules within the candidate defect cluster.

6. The method for defect clustering and iterative priority generation in audio control applications according to claim 1, characterized in that, The calculated defect severity score includes: The problem unit is subjected to emotion intensity recognition, which identifies emotion words, negative expressions, expressions of complaint intensity, and expressions of return intention in the feedback text; The severity of a feature is determined based on how much it affects the user’s core tasks. Issues that cause devices to be unable to connect to the network, play content, crash applications, become unauthorized, or become undiscoverable are classified as high severity, while issues with the interface or text display are classified as low severity. The severity score of the defect is obtained by weighting and fusing the emotional intensity, functional severity, user rating, whether it leads to a return, whether it affects the first-time user experience, and whether it affects the core playback function of the problem unit in the candidate defect cluster.

7. The method for defect clustering and iterative priority generation in audio control applications according to claim 1, characterized in that, The attribution analysis includes: Align the time of negative reviews with the time of application version release, firmware version release, server configuration change, and third-party service interface change to determine whether the defect is related to a specific update. Stratified analysis is conducted based on equipment model, hardware capabilities, and whether a display screen is provided to identify whether defects occur only in specific equipment models or combinations of equipment. Perform layered analysis based on operating system platforms to identify the unique defects of each system platform; Based on the statistical significance of the defect frequency variation, the consistency between the defective equipment distribution and the equipment sales distribution, and the cross-validation results of the embedded data, a confidence level is generated for the attribution analysis.

8. The method for defect clustering and iterative priority generation in audio control applications according to claim 1, characterized in that, The method further includes: While meeting privacy compliance requirements, user feedback is matched with application tracking data, device logs, and cloud status data to identify whether any abnormal events occurred before or after the user feedback. If the defect topic described in the feedback text is consistent with the abnormal event of the tracking point, the corresponding defect will be marked as a high-confidence defect. If a defective topic receives frequent feedback but the tracking data does not capture the corresponding anomaly, a missing tracking information message will be generated.

9. The method for defect clustering and iterative priority generation in audio control applications according to claim 1, characterized in that, The calculation of the iteration priority score for each defect includes: The frequency and proportion of each candidate defect cluster are statistically analyzed within a specified time window, and the frequency changes between the current period and the previous period are compared to identify problems whose growth trend exceeds a preset threshold. Estimate the user scale affected by each defect by combining active user count, device sales, and feature utilization rate; Based on the feature information of the candidate defect cluster, the defect severity score, the attribution analysis results, and user behavior tracking data, multiple priority scoring factors are set, and the iteration priority score is obtained through weighted calculation. The weight configuration of each factor is dynamically adjusted according to the product stage.

10. The method for defect clustering and iterative priority generation in audio control applications according to claim 1, characterized in that, The generated defect task list and product iteration suggestions include: For defects whose iteration priority score is higher than a preset threshold, a defect task card is generated. The defect task card includes the defect name, representative feedback text, scope of impact, associated version, associated device, suspected cause, evidence data, priority level and verification indicators. Based on the defect type, product optimization suggestions are generated. These suggestions include: explanations of permission usage and step-by-step authorization suggestions for privacy-related defects; fault diagnosis and compatibility tips for network connectivity-related defects; hierarchical simplification and operation guidance suggestions for interface usability-related defects; and authorization status detection and re-authentication path suggestions for third-party service-related defects. Test scenarios are automatically generated based on candidate defect clusters. These test scenarios cover specific network environments, specific music services, specific operating system versions, specific combinations of input sources, and specific combinations of multiple devices.

11. The method for defect clustering and iterative priority generation in audio control applications according to claim 1, characterized in that, The method further includes: Defects with iteration priority scores higher than a preset threshold will be associated with the application version, firmware version, or cloud configuration version to be released. For defects that require collaborative repair involving applications, firmware, cloud, and third-party interfaces, record cross-component dependencies and boundaries of responsibility; Compare defects with the product roadmap. If a defect relates to a pre-defined strategic functional direction, increase the strategic priority weight of that defect.

12. The method for defect clustering and iterative priority generation in audio control applications according to claim 1, characterized in that, The method also includes defect iteration effect verification and model update: Before launching defect repairs or product optimizations, set verification indicators, including the rate of decrease in the number of related negative reviews, the change in the proportion of low-star reviews, the rate of decrease in corresponding abnormal tracking points, the increase in task completion rate, the rate of decrease in customer service tickets, and the increase in user retention. After the new version is released, continuously monitor the changing trends of relevant candidate defect clusters, compare the indicator data of the same time window before and after the release, and determine whether the optimization is effective; If the original candidate defect clusters decrease after repair but new semantic expressions or new sub-problems appear, then the synonym expression mapping relationship and clustering model of the defect topic dictionary are updated. It receives manual confirmation, confirms, merges, splits, or renames the clustering results, and feeds back the manually confirmed classification results as training samples to the clustering model.

13. A defect clustering and iterative priority generation system for audio control applications, characterized in that, include: The data acquisition module obtains user feedback data from multiple sources; The defect clustering module is used to perform text preprocessing on the user feedback data, generate several problem units, and cluster the problem units based on the semantic vectors and defect topic dictionaries corresponding to the problem units to form candidate defect clusters; The severity assessment module is used to assess the severity of each candidate defect cluster and calculate the defect severity score; The attribution analysis module is used to perform attribution analysis on each of the candidate defect clusters to identify the version correlation and device distribution characteristics of the defects. The priority generation module is used to calculate the iteration priority score of each defect based on the feature information of the candidate defect cluster, and generate a defect task list and product iteration suggestions based on the iteration priority score.

14. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the audio control application defect clustering and iterative priority generation method as described in any one of claims 1-12.

15. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the audio control application defect clustering and iterative priority generation method as described in any one of claims 1-12.

Images