Service quality evaluation method and system based on multi-source data fusion

Abstract

The invention relates to the technical field of service quality evaluation, and discloses a service quality evaluation method and system based on multi-source data fusion. The method comprises the following steps: acquiring service data of a plurality of data sources, wherein the service data specifically comprises service request time, service response delay, a service error rate and user feedback information; respectively carrying out time alignment and dimension alignment processing on the service data to form a multi-dimensional data set with a uniform format; performing feature clustering analysis on the basis of the multi-dimensional data set, and extracting target feature data for service quality evaluation and abnormal indication data for dynamic monitoring from the multi-dimensional data set; generating a service quality evaluation result by using the target feature data, wherein the result contains a service grade score and a potential risk index; and identifying an abnormal condition in service data updating according to the abnormal indication data, and dynamically correcting a service quality evaluation result according to the abnormal condition.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of service quality evaluation, in particular to a service quality evaluation method and system based on multi-source data fusion. BACKGROUND

[0002] In the current rapid development of digital services, the scale of various service systems is continuously expanding, and the service forms are increasingly complex, from traditional IT operation services to emerging cloud computing services, Internet of Things services, etc. Users' attention to service quality continues to increase, and service quality evaluation has become an important link to ensure stable operation of services and optimize service experience. However, there are many problems to be solved in the current service quality evaluation process.

[0003] Service data sources show a multi-source characteristic. Different service modules, devices and user interaction links will generate a large amount of data, which are stored in different systems or platforms, for example, service request time data may come from the front-end request log system, service response delay data may be collected by the back-end server performance monitoring tool, service error rate data may be scattered in database error records and application logs, and user feedback information is mostly stored in customer service systems or user evaluation platforms. These scattered data sources make it difficult to form a unified analysis basis for service data. If the evaluation personnel want to obtain comprehensive service data, they often need to switch between multiple systems, which not only increases the complexity of data acquisition, but also may cause the evaluation results to lack timeliness due to the difference in data extraction time.

[0004] The existing technology lacks an effective alignment mechanism for processing multi-source service data. The time granularity of data generated by different data sources is different, some systems record data in seconds, and some systems store data in minutes or even hours. This inconsistency in time dimension will cause deviation in data correlation analysis. At the same time, the definition and statistical dimension of data by each data source are not unified, for example, different monitoring tools may have different calculation ranges for "service response delay", some only calculate the time for the server to process the request, and some include data transmission time. This difference in dimension makes it difficult to directly fuse multi-source data, which further affects the accuracy of the evaluation results.

[0005] Traditional service quality evaluation methods mostly rely on single-dimensional data or static feature analysis, which is difficult to fully reflect the actual running state of the service. Some evaluation methods only evaluate according to technical indicators such as service response delay or service error rate, ignoring the important subjective indicator of user feedback, resulting in the evaluation results being out of touch with the actual experience of users. In addition, the traditional method lacks the ability of dynamic monitoring and abnormal identification of service data. When the service data fluctuates or is abnormal, the evaluation results cannot be found and adjusted in time, so that the evaluation results cannot reflect the changes of service quality in real time, and then affect the timeliness and effectiveness of service optimization decision.

[0006] The evaluation results generated by the existing evaluation methods often only contain simple scores, and lack effective identification and analysis of potential risks of the service. There may be various potential risks in the running process of the service, such as the error rate of a certain service module does not reach the threshold, but shows a rising trend. If such potential risks cannot be identified in time, it may lead to a significant decline in subsequent service quality. The traditional evaluation method cannot extract effective risk indication information from multi-source data, and it is difficult to warn the potential risks of the service, which is not conducive to the stable operation and long-term optimization of the service system. SUMMARY

[0007] The purpose of the present application is to provide a service quality evaluation method and system based on multi-source data fusion to solve the problems raised in the background art.

[0008] To achieve the above purpose, the present application provides a service quality evaluation method based on multi-source data fusion, which comprises:

[0009] Obtaining service data of multiple data sources, the service data including service request time, service response delay, service error rate and user feedback information;

[0010] Performing time alignment and dimension alignment processing on the service data to form a unified multi-dimensional data set;

[0011] Performing feature clustering analysis processing based on the multi-dimensional data set, extracting target feature data for service quality evaluation and abnormal indication data for dynamic monitoring;

[0012] Generating a service quality evaluation result using the target feature data, the service quality evaluation result including a service level score and a potential risk indicator;

[0013] According to the abnormal situation in the service data update identified according to the abnormal indication data, the service quality evaluation result is dynamically corrected based on the abnormal situation.

[0014] Preferably, the time alignment and dimension alignment processing of the service data to form a unified multi-dimensional data set comprises:

[0015] integrating the timing data of the service request time, the service response delay, the service error rate and the user feedback information;

[0016] mapping the integrated timing data to a service instance dimension, a time window dimension and a data source type dimension;

[0017] generating the multidimensional data set for inputting the feature clustering analysis process.

[0018] Preferably, the feature clustering analysis process based on the multidimensional data set comprises:

[0019] converting the multidimensional data set into a feature vector matrix;

[0020] performing a clustering grouping operation on the feature vector matrix to output normal feature clusters and abnormal feature clusters;

[0021] extracting the target feature data from the normal feature clusters, the target feature data comprising service stability indicators and response efficiency indicators;

[0022] extracting the abnormal indication data from the abnormal feature clusters, the abnormal indication data comprising service interruption labels and delay deviation labels.

[0023] Preferably, the generation of the service quality evaluation result using the target feature data comprises:

[0024] calculating a service level score based on the service stability indicators and the response efficiency indicators;

[0025] generating the potential risk indicators in combination with historical evaluation data of the service instance;

[0026] outputting the service quality evaluation result to a service monitoring terminal.

[0027] Preferably, the identification of the abnormal situation in the service data update according to the abnormal indication data comprises:

[0028] monitoring the change trend of the service interruption labels and the delay deviation labels;

[0029] determining the abnormal situation when the service interruption labels or the delay deviation labels exceed a preset threshold;

[0030] inputting the abnormal situation into the process of dynamic correction.

[0031] Preferably, the dynamic correction of the service quality evaluation result based on the abnormal situation comprises:

[0032] A service quality evaluation index is created, which stores a mapping relationship between service instance identifiers and evaluation indicators;

[0033] The evaluation indicators in the service quality evaluation index are updated based on the abnormal situation;

[0034] The service level score and potential risk indicators are recalculated using the updated service quality evaluation index;

[0035] A revised service quality evaluation result is generated.

[0036] Preferably, the creation of the service quality evaluation index comprises:

[0037] An evaluation indicator reference table is constructed, which takes service instance identifiers as keys and associated service level scores and potential risk indicators as values;

[0038] An index update mechanism is configured, which triggers index reconstruction in response to the abnormal situation;

[0039] The service quality evaluation index is used in the process of recalculation.

[0040] Preferably, the method further comprises:

[0041] The multi-source correlation in the abnormal indication data is analyzed to identify cross-data-source linkage abnormalities;

[0042] When the linkage abnormalities are confirmed, an evaluation optimization strategy is executed, which includes adjusting parameters of feature clustering analysis processing;

[0043] The feature clustering analysis processing is re-executed based on the adjusted parameters;

[0044] The re-executed result is input into the update process of the service quality evaluation index.

[0045] Preferably, the execution of the evaluation optimization strategy comprises:

[0046] A service data sampling window is set for iterative training;

[0047] The performance indicators of feature clustering analysis processing are evaluated in the iterative training;

[0048] The granularity of clustering grouping operations is optimized according to the performance indicators;

[0049] The optimized feature clustering analysis processing is output to the re-execution process.

[0050] Preferably, the application further comprises a service quality evaluation system based on multi-source data fusion and big data analysis, which comprises a memory, a processor and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the service quality evaluation method based on multi-source data fusion as described above.

[0051] Compared with the prior art, the application has the following beneficial effects:

[0052] By obtaining service data from multiple data sources, covering multiple types of key data such as service request time, service response delay, service error rate and user feedback information, the limitation of single data source in traditional evaluation methods is broken. Traditional evaluation often relies on one or several types of technical indicators, ignoring feedback information related to user subjective experience, resulting in a one-sided evaluation perspective. However, this method combines objective technical data with subjective user feedback, enabling a more comprehensive capture of various information during service operation, providing a richer and more complete data foundation for subsequent evaluation processes, avoiding evaluation bias caused by data loss or singularity, and making the evaluation results more consistent with the actual operation status and user real experience of the service.

[0053] In the data processing link, the method performs time alignment and dimension alignment processing on the service data to form a unified multi-dimensional data set, effectively solving the core problem in the multi-source data fusion process. Multi-source data often comes from different systems or platforms, with significant differences in time granularity and data definition. For example, the time intervals for recording data by different data sources may vary from seconds to hours, and the statistical ranges for the same indicator may also vary. These differences may lead to data that cannot be directly correlated and analyzed, or even contradictory analysis results. Through time alignment, data of different time granularities can be adjusted to a unified time dimension, ensuring the comparability and relevance of data in the time layer. Through dimension alignment, the definition and statistical standards of data by each data source can be standardized, eliminating inconsistencies in data dimensions, enabling true fusion of multi-source data, and forming a data set with unified structure and logical coherence, laying a solid data foundation for subsequent feature analysis and evaluation work, and reducing evaluation errors caused by data format or standard differences.

[0054] Based on the unified multidimensional data set, feature clustering analysis processing is performed to extract target feature data and abnormal indication data, which fully mines the effective information in the multi-source fusion data. Feature clustering analysis can filter out target features that have a key impact on service quality evaluation from massive multidimensional data according to the internal correlation and feature similarity of the data. These target features can accurately reflect the core influencing factors of service quality, avoiding redundant analysis of irrelevant data in the evaluation process and improving evaluation efficiency. At the same time, the extracted abnormal indication data can capture abnormal fluctuations or potential abnormal trends in service data. Compared with the traditional method that can only identify abnormalities when the data exceeds the threshold, this method can discover abnormal signs in service operation earlier, providing effective data basis for subsequent dynamic monitoring and evaluation correction. The service quality evaluation not only focuses on the current state, but also effectively controls the data change trend.

[0055] The target feature data is used to generate service quality evaluation results containing service level scores and potential risk indicators, which enrich the connotation and application value of the evaluation results. Traditional evaluation results are mostly presented in the form of a single score, which can only roughly reflect the service quality level and cannot provide specific directions for service optimization. The service level score can intuitively display the overall quality level of the current service, helping relevant personnel quickly grasp the service quality profile. The potential risk indicator can identify potential risk points in the service running process from multi-source fusion data, such as the error rate of a certain service module being within the normal range, but combined with the increasing trend of related complaints in user feedback, a potential risk indicator can be formed to indicate that the module may have hidden problems. This evaluation result containing double content not only makes the evaluation conclusion more in-depth, but also provides more specific guidance for service optimization decisions, helping staff to carry out service improvement work targetedly.

[0056] According to the abnormal indication data, the abnormal situation in the service data update is identified, and the service quality evaluation result is dynamically corrected, so that the evaluation result has real-time and accuracy. During service running, data will be updated over time and may appear various abnormal situations, such as sudden increase of service response delay, surge of negative information in user feedback, etc. If the evaluation result cannot be adjusted in time according to these changes, it will lead to the disconnection between the evaluation result and the actual service state, losing the guiding significance. This method monitors the abnormality in data update in real time through abnormal indication data. Once an abnormality is found, the evaluation result is corrected immediately to ensure that the evaluation result can accurately reflect the current service quality status at all times, providing timely and reliable reference for service monitoring, fault troubleshooting and service optimization, helping the service system to maintain stable and efficient operation state and better meet user needs. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1A working principle diagram of the service quality evaluation method based on multi-source data fusion according to the present application;

[0058] Figure 2 A flowchart for time alignment and dimension alignment processing;

[0059] Figure 3 A flowchart for feature clustering analysis processing;

[0060] Figure 4 A flowchart for dynamically correcting service quality evaluation results;

[0061] Figure 5 A flowchart for multi-source correlation analysis and evaluation optimization. DETAILED DESCRIPTION

[0062] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0063] Please refer to Figure 1 The present application provides a service quality evaluation method based on multi-source data fusion, which comprises:

[0064] By obtaining service data of multiple data sources, including service request time, service response delay, service error rate and user feedback information. The service data is processed for time alignment and dimension alignment to form a unified multi-dimensional data set. Based on the multi-dimensional data set, feature clustering analysis processing is performed to extract target feature data for service quality evaluation and abnormal indication data for dynamic monitoring. The target feature data is used to generate service quality evaluation results, including service level score and potential risk indicators. According to the abnormal indication data, abnormal situations in service data update are identified, and the service quality evaluation results are dynamically corrected based on the abnormal situations.

[0065] Embodiment 1: Please refer to Figure 2Time alignment and dimension alignment processing is the core step of building a unified multi-dimensional dataset. The input of this process is raw service data streams from multiple independent systems. For example, service request time data might come from a load balancer's access log, with a record format of timestamp, service instance ID, request ID. Service response latency data might be provided by an application performance monitoring tool, containing request ID, response milliseconds, success status code. Service error rate data might originate from an error tracking platform, aggregating the number of errors per minute for each service instance. User feedback information comes from a client-side experience rating system, sent in the form of asynchronous messages, containing user ID, timestamp, rating value, and optional text comments.

[0066] These data sources naturally differ in collection frequency, time precision, and transmission delay. A load balancer log might produce thousands of records per second, while user feedback might only have a few messages per minute. Application performance monitoring data might be at a granularity of five seconds, while error tracking data is aggregated once per minute. Time alignment processing first needs to establish a unified time axis reference for all data, such as Coordinated Universal Time (UTC). The timestamp in each data entry is extracted and converted to this reference time. For timestamps that are not precise to milliseconds, such as those that are only seconds or minutes, an interpolation method is used to assign them to a finer granularity. For example, the number of errors aggregated per minute is evenly distributed to each second within that minute, but this distribution is marked as an estimated value for subsequent processing to identify.

[0067] After completing the preliminary timestamp standardization, the next step is data integration. Integration is not simply putting all data into a table, but rather associating data based on common keys such as request ID, service instance ID, etc. For example, through the request ID, the load balancer's request record can be associated with the application performance monitoring's latency record. For data without direct common keys, such as user feedback that is not associated with a specific request ID, time proximity matching is used. For example, a user feedback timestamp is T, then all requests sent by the same user within the time window [T-ΔT, T+ΔT] are matched, and the service instances of these requests are associated with the feedback rating.

[0068] After the temporal data integration, the multi-dimensional mapping phase begins, where the service instance dimension is mapped to assign a unique dimension key to each service instance and record its static attributes, such as the cluster it belongs to, version number, deployment region, etc. All data related to this instance, regardless of the source, is associated through this dimension key. The time window dimension maps the continuous time stream into non-overlapping fixed interval windows, such as five-minute windows. Each window has a unique start timestamp as the key. All data records within a window are aggregated into summary data. For example, all request delays of a service instance within a window are aggregated into mean, 95th percentile, maximum, etc. The number of errors is aggregated into a sum, and user ratings are aggregated into a mean. The data source type dimension is mapped to assign a category label to each raw data source, such as "access log", "performance metric", "error metric", "user experience". Each summary data carries its source information to trace the data lineage.

[0069] The final generated multi-dimensional dataset is a hierarchical data structure, with the top level organized by time windows. Each time window object contains entries for all monitored service instances within that window. Each service instance entry contains all its aggregated metric data within that window, with annotations of the original data sources for each metric. This dataset is usually saved in columnar storage format for efficient querying and analysis. For example, the performance metric sequence of a service instance in the past hour can be quickly obtained, as well as the error rate snapshot of all service instances at a certain time point. This structured multi-dimensional dataset provides clear, organized, and semantically rich input for subsequent feature clustering analysis, enabling the clustering algorithm to mine patterns from multiple perspectives such as time, instance, data source, etc.

[0070] The entire processing flow is implemented through a stream processing framework, such as Apache Flink or Spark Streaming. Data flows in real-time, goes through temporal alignment, correlation integration, multi-dimensional mapping, and window aggregation, and the resulting multi-dimensional dataset is updated in real-time and pushed to the feature clustering analysis module. This design enables continuous service quality evaluation based on near real-time data.

[0071] Embodiment 2: Refer to Figure 3 , the multi-dimensional dataset serves as the input source, which has been organized in a three-dimensional structure of service instance, time window, and data source type. For example, in the service monitoring scenario of an e-commerce platform, a five-minute time window may contain data for hundreds of service instances, each corresponding to a set of aggregated metrics: request volume statistics from the load balancer, average and P95 response delays from application performance monitoring, HTTP 500 error counts from the error tracking system, and average rating values from the user feedback system.

[0072] The construction process of the eigenvector matrix first determines the vector dimension, and all the indicators of each service instance within a single time window are flattened into a numerical vector. For example, the vector contains eight dimensions: the time window start timestamp (normalized to the 0-1 range), the service instance ID (hashed into a number), the total number of requests, the average delay, the P95 delay, the number of errors, the average user rating, and the rating sample size. Identical indicators from different data sources retain independent dimensions, such as "delay_load_balancer" and "delay_APM_monitor", which belong to two dimensions. The number of rows in the entire matrix is the product of the number of service instances and the number of time windows, and the number of columns is the total number of determined feature dimensions.

[0073] The clustering grouping operation uses an improved DBSCAN algorithm. The algorithm first calculates the standardized Euclidean distance between vectors, and applies z-score normalization to each dimension to eliminate dimensional differences. In the density parameter setting, the neighborhood radius ε is determined by analyzing the average distance distribution of historical data vectors, and the minimum sample size MinPts is set to 10% of the total number of time windows. In the e-commerce platform case, the clustering result of a certain time period may show that the order service instance forms a high-density core cluster, the payment service instance forms a medium-density cluster, and the inventory service instance forms discrete points due to data fluctuations.

[0074] The identification of normal feature clusters is based on the vector density and size within the cluster. When extracting service stability indicators in the core cluster, the coefficient of variation (standard deviation divided by mean) of all vectors in the cluster in the error count dimension is calculated. This coefficient reflects the relative fluctuation amplitude of the service error rate. The response efficiency indicator is the median value of the P95 delay dimension within the cluster, representing the tail delay level in the typical scenario. The vectors of abnormal feature clusters are usually distributed outside the density threshold. The generation logic of the service interruption flag is as follows: when a vector's value in the error count dimension exceeds three times the standard deviation of that dimension in the normal cluster, the service instance is marked as having a risk of interruption in that time window. The delay deviation flag calculates the percentage deviation of the vector's value in the average delay dimension from the mean value of the normal cluster.

[0075] The calculation of the service level score combines multiple indicators. The service stability indicator weight is set to 0.6, and the response efficiency indicator weight is 0.4. After min-max normalization, the two indicators are weighted and summed to obtain a base score of 0-100. For e-commerce payment services, a user rating adjustment factor is also needed: when the user rating is lower than 3 stars (5-star system), the base score is forced to decrease by 20%. The calculation of the potential risk indicator introduces time series analysis, taking the percentage difference between the current window's service stability indicator value and the 24-hour moving average value, and adding the slope value of the error rate trend line of the service instance in the past 7 days.

[0076] The result output link adopts a hierarchical data structure, and each service instance generates a JSON object containing the following fields: instance ID, time window start and end time, service level score, potential risk level (low / medium / high), detailed indicator values (including stability score, efficiency score), and abnormal marker list (if any). These objects are grouped by time window and pushed to the monitoring terminal through the message queue. The terminal interface can display a heat map of service instances, with color depth representing risk level, and detailed score composition displayed on mouseover.

[0077] The entire processing flow runs in micro-batch mode, triggering a computing task every five minutes to read the latest window data from the multi-dimensional dataset, perform the complete chain of feature vector conversion, cluster analysis, indicator calculation, and result output. Historical cluster center points are cached to accelerate cluster assignment of new data, and when the cluster silhouette coefficient of three consecutive windows is below the threshold, the parameter tuning process is automatically triggered.

[0078] Embodiment 3: refer to Figure 4 This process continuously receives abnormal indication data streams from the feature cluster analysis module, which contain service interruption markers and delay deviation markers, each associated with a service instance identifier, time window stamp, and numerical intensity. The monitoring of service interruption markers focuses on their continuous occurrence patterns. The system maintains a sliding time window, with a default of three consecutive monitoring periods (e.g., 15 minutes). When the cumulative service interruption marker intensity sum of a service instance within the window exceeds the threshold , an abnormal situation is triggered. The threshold is calculated as follows:

[0079] ;

[0080] Where: represents the historical average of service interruption marker intensity for the same period (e.g., 10:00-10:15 on Monday) of the past 7 same working days, is the corresponding historical standard deviation. This dynamic threshold mechanism takes into account the periodicity of service load, avoiding false positives during business peak periods.

[0081] The monitoring of the delay deviation marker analyzes its trend, and the system calculates the first derivative of the marker value for each service instance, i.e. the difference between the current time window and the last time window. When the derivative value is continuously positive and accumulates over a certain range, it indicates that the delay is continuously deteriorating, even if the absolute value has not reached the top, it will be identified as a potential abnormal situation. This trend monitoring helps to find the gradual degradation of performance, rather than just responding to sudden failures. Once the abnormal situation is determined, the system encapsulates it as a structured event object, which contains the abnormal type, the associated service instance identifier, the trigger time, the severity score and the related raw data reference. The event object is immediately sent to the dynamic correction process.

[0082] The core of the dynamic correction process is to maintain a quality of service evaluation index, which is constructed as a nested hash map structure in memory. The outermost key is the service instance identifier, and its value is an embedded dictionary containing the latest evaluation indicators. The dictionary stores the current service level score, potential risk indicator, service stability indicator and response efficiency indicator calculation results snapshot in the form of key-value pairs, as well as the version timestamp corresponding to these snapshots.

[0083] When the abnormal event arrives, the correction process first locates the evaluation indicator dictionary of the instance in the service quality evaluation index according to the service instance identifier in the event. Then, the update logic is executed according to the abnormal type. For service interruption type of abnormality, the system will adjust the service stability indicator value in the instance dictionary according to the severity score of the event. For delay deviation type of abnormality, the response efficiency indicator value is mainly affected. These adjustments are based on a pre-defined rule matrix, which defines the mapping relationship between different abnormal types, severity and the correction amplitude of each evaluation indicator.

[0084] After completing the update of the evaluation indicator dictionary, the system recalculates the service level score and potential risk indicator using the updated indicator values. The algorithm used for recalculation is consistent with the initial evaluation module, but its input data has reflected the latest abnormal state. For example, a service instance with delay deviation caused by network jitter, its response efficiency indicator is adjusted, and the recalculated service level score will decrease accordingly, and the potential risk indicator level may be raised from "low" to "medium".

[0085] The corrected service quality evaluation result is generated. The result contains not only the updated score and indicators, but also a correction log recording the reason for this correction (i.e. the ID of the triggered abnormal event), the correction time, and the change of the indicator value before and after the correction, thereby ensuring the transparency and traceability of the evaluation process. The corrected result is updated in real time to the service quality evaluation index, and at the same time pushed to all service monitoring terminals subscribed to the state of the service instance, so that the operation and maintenance personnel can immediately perceive the dynamic change of the service state. The whole monitoring and correction cycle runs in an asynchronous and event-driven manner, ensuring that the system can respond to abnormal situations in near real time.

[0086] Embodiment 4: refer to Figure 5 The construction of the evaluation indicator reference table is the core foundation, which exists in a separate table space in the relational database, and its physical structure is designed as a row storage mode to support high-frequency point queries. Each record corresponds to an evaluation state snapshot of a service instance at a specific time point. The table contains the following key fields: primary key ID, service instance identifier, timestamp, service level score value, potential risk level code, service stability indicator value, response efficiency indicator value, data version number, and record creation time. The service instance identifier and the timestamp form a joint unique index, ensuring that there is only one valid record for the same instance at the same time. The potential risk level code uses enumeration type, for example, using numbers 1, 2, and 3 to represent low, medium, and high risk respectively. The data version number is used to implement the optimistic locking mechanism to ensure data consistency during concurrent updates.

[0087] The implementation of the index update mechanism relies on an event listener architecture, which subscribes to a topic named "abnormal event" in the message middleware. Whenever the dynamic correction module publishes an abnormal situation event, the listener triggers the index reconstruction process. The process first parses the abnormal event details from the transaction log, extracting key elements including the list of affected service instance identifiers, the time window of the abnormal event occurrence, the abnormal type and severity coefficient. Subsequently, the process generates a batch of structured query language statements based on these elements. The operation objects of these statements are all records in the evaluation indicator reference table whose timestamp is within the abnormal time window and whose service instance identifier is in the list of affected instances. The update logic is based on a pre-defined rule set, for example, for the "service interruption" type of abnormality, the service stability indicator value of the related record is multiplied by a decay factor based on the severity coefficient. All update operations are completed in a database transaction, ensuring that either all succeed or all rollback. After the update is completed, the data version number is incremented and a new timestamp is recorded.

[0088] The identification of a correlated anomaly is a multi-stage analysis process. The system maintains a sliding time window and continuously collects all the labeled anomaly indication data within this window. These data not only contain the anomaly labels themselves, but also retain their complete data source context information. The analysis engine first groups these anomaly events by data source type, such as infrastructure monitoring sources, application performance monitoring sources, and end-user experience monitoring sources. Then, the engine calculates the correlation between anomaly events occurring in different data source groups. The calculation is not based on simple occurrence time matching, but uses an overlap analysis based on event duration. For example, a "CPU usage surge" event from infrastructure monitoring and a "request timeout rate increase" event from application performance monitoring are considered to have a potential correlation if their high-duration windows significantly overlap.

[0089] Table 1: Correlated anomaly correlation rule table.

[0090]

[0091] When the identified correlation strength exceeds the preset confidence threshold and matches a rule pattern defined in the matching table, the correlated anomaly is confirmed. After confirmation, the system generates a correlated anomaly event, which includes all the associated original anomaly event identifiers, the list of involved data source types, and a comprehensive severity score.

[0092] The execution of the evaluation optimization strategy is a key step in dealing with correlated anomalies. The first action of this strategy is to adjust the parameters of the feature clustering analysis process. Specific adjustments include modifying the neighborhood search radius of the clustering algorithm and the minimum sample size for defining core points. In the context of correlated anomalies, since multiple data sources simultaneously exhibit anomalies, the overall density of data distribution and the definition of outliers may change, so the definition conditions of core points need to be relaxed to avoid mistakenly dividing normal service instances into outliers. The specific values of parameter adjustment are recommended by an analysis model based on historical correlated anomaly cases.

[0093] Based on the adjusted parameters, the system will re-execute the feature clustering analysis process for the current time window and the data of several affected historical time windows. The scope of this re-computation is local, limited to the service instances and data time range directly or indirectly affected by the correlated anomaly event, in order to control the computational overhead. The re-execution process is the same as the initial clustering analysis, but uses new parameter settings, thus producing a set of corrected feature cluster division results.

[0094] The result of re-execution is encapsulated as a data update event and input to the update process of the quality of service evaluation index. In addition to updating the specific values in the evaluation index reference table as in the processing of ordinary exceptions, there is an additional operation: in the extension metadata field of the related record, it is recorded that this update is caused by the recalculation triggered by the linkage exception, and is associated with the corresponding linkage exception event ID. This design makes the evolution history of the evaluation result have complete traceability, facilitating subsequent root cause analysis or auditing. The entire process from event listening to index updating forms a closed-loop automated feedback regulation system, which can cope with complex cross-data source exception scenarios.

[0095] In embodiment 5, a hierarchical random sampling strategy is used to construct the service data sampling window. The system first extracts all records in the last 72 hours from the multi-dimensional data set as a candidate pool. The candidate pool is layered according to the business domain to which the service instance belongs, for example, divided into order processing domain, payment clearing domain, inventory management domain, etc. Within each business domain, the time stamps are arranged in chronological order. Then, the system randomly selects 5% of the service instances from each business domain and extracts the complete time window data of these instances in the last 72 hours. This sampling method covers different business characteristics, preserves time continuity, and controls the size of the training data.

[0096] The iterative training process runs in a separate computing sandbox environment, and each iteration includes three main stages: parameter combination generation, clustering execution, and effect evaluation. In the parameter combination generation stage, a grid search strategy is used, and the system predefines a set of parameters to be optimized and their value ranges. The core parameters include the neighborhood radius of the clustering algorithm, the minimum sample size, the distance metric method, etc. The search range of the neighborhood radius is dynamically determined based on the statistical quantile of the distance between vectors in the historical clustering results, and the search range of the minimum sample size is linked with the percentage interval of the number of service instances in the sampled data set. A set of parameter combinations is selected for testing in each iteration.

[0097] The clustering execution stage converts the sampled window data into a feature vector matrix and runs the clustering algorithm using the current tested parameter combination. The algorithm outputs the cluster labels for each vector and the core point labels. The effect evaluation stage calculates two core performance indicators: the silhouette coefficient reflects the separation and compactness of the clusters, and the value closer to 1 indicates better clustering effect; the intra-class distance ratio calculates the ratio of the average intra-class distance to the global average distance, and a value lower than 1 indicates that the intra-class structure is compact. The system records the performance indicator values corresponding to each parameter combination. The parameter optimization decision is based on the evaluation results of multiple iterations. The system maintains a parameter performance history table, which accumulatively records the parameter combination and its performance indicators of each iteration test. When the predetermined number of iterations is completed or the time limit is reached, the optimization engine analyzes the history table. The analysis process adopts the Pareto optimization principle to filter out candidate combinations that are not fully surpassed by other parameter combinations in both the silhouette coefficient and the intra-class distance ratio dimensions. If there are multiple candidate combinations, the combination with the lowest historical usage frequency is selected to promote exploration diversity.

[0098] The optimized parameter combination is output to the configuration center of the feature clustering analysis process, which uses version management, generating a new configuration version number for each update and preserving historical versions. The system automatically pushes the new parameters to the online feature clustering analysis module, which executes the clustering grouping operation using the new parameters in the next processing period. To verify the generalization ability of the new parameters, the system performs shadow evaluation on the online clustering results in the next three time windows. Shadow evaluation uses the same performance indicators, but the data comes from the complete data set of real-time processing rather than sampled data. If the shadow evaluation indicators are better than the performance of the old parameters, the new parameters are officially marked as stable versions; otherwise, the rollback mechanism is triggered and the optimization process is restarted.

[0099] The metadata generated during the entire iterative training process is recorded in its entirety, including the composition details of the sampling window, the sequence of tested parameter combinations, the performance indicator values of each iteration, the basis for the final optimization decision, and the online verification results. These records are used for audit tracking and provide a reference basis for adjusting the parameter range in subsequent optimization cycles. This closed-loop optimization mechanism enables the feature clustering analysis to continuously adapt to changes in service data distribution, maintaining the accuracy of service quality evaluation.

[0100] It should be noted that, in the present text, relational terms such as first and second are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or device.

[0101] While embodiments of the application have been shown and described, it is to be understood that the application is not limited to the details of the embodiments described, since numerous changes, modifications, substitutions and variations can be made thereto without departing from the spirit and scope of the application as defined by the appended claims and their equivalents.

Claims

1. A service quality evaluation method based on multi-source data fusion, characterized in that: include: Acquire service data from multiple data sources, including service request time, service response delay, service error rate, and user feedback information; Performing time alignment and dimension alignment processing on the service data to form a unified multi-dimensional data set; Performing feature clustering analysis based on the multidimensional dataset to extract target feature data for service quality assessment and abnormality indicator data for dynamic monitoring; Generating a service quality assessment result using the target feature data, the service quality assessment result including a service level score and a potential risk indicator; An abnormality in the service data update is identified according to the abnormality indication data, and the service quality evaluation result is dynamically corrected based on the abnormality.

2. The service quality evaluation method based on multi-source data fusion according to claim 1 is characterized in that: The performing time alignment and dimension alignment processing on the service data to form a unified multi-dimensional dataset includes: Integrate the time series data of the service request time, service response delay, service error rate and user feedback information; Map the integrated time series data to the service instance dimension, time window dimension, and data source type dimension; The multidimensional dataset is generated, and the multidimensional dataset is used to input the feature clustering analysis process.

3. The service quality evaluation method based on multi-source data fusion according to claim 1 is characterized in that: The performing feature clustering analysis based on the multidimensional data set includes: converting the multidimensional dataset into a matrix of eigenvectors; Performing a clustering and grouping operation on the feature vector matrix to output normal feature clusters and abnormal feature clusters; Extracting the target feature data from the normal feature cluster, the target feature data including a service stability index and a response efficiency index; The abnormality indication data is extracted from the abnormality feature cluster, where the abnormality indication data includes a service interruption flag and a delay deviation flag.

4. The service quality evaluation method based on multi-source data fusion according to claim 3 is characterized in that: Generating a service quality evaluation result by using the target feature data includes: Calculate a service level score based on the service stability index and the response efficiency index; Generating the potential risk indicator in combination with historical assessment data of the service instance; Output the service quality evaluation result to the service monitoring terminal.

5. The service quality evaluation method based on multi-source data fusion according to claim 3 is characterized in that: The identifying an abnormality in the service data update according to the abnormality indication data includes: monitoring the changing trends of the service interruption mark and the delay deviation mark; When the service interruption mark or the delay deviation mark exceeds a preset threshold, determining the abnormal situation; The abnormal situation is input into the dynamic correction process.

6. The service quality evaluation method based on multi-source data fusion according to claim 5 is characterized in that: The dynamically revising the service quality evaluation result based on the abnormal situation includes: Creating a service quality evaluation index, wherein the service quality evaluation index stores a mapping relationship between a service instance identifier and an evaluation indicator; Updating the evaluation index in the service quality evaluation index based on the abnormal situation; Recalculating the service level score and potential risk indicator using the updated service quality assessment index; Generate a revised service quality evaluation result.

7. The service quality evaluation method based on multi-source data fusion according to claim 6 is characterized in that: The creating of the service quality evaluation index includes: Constructing an evaluation indicator reference table, wherein the evaluation indicator reference table uses the service instance identifier as a key and the associated service level score and potential risk indicator as a value; configuring an index update mechanism, wherein the index update mechanism triggers index reconstruction in response to the abnormal situation; The service quality evaluation index is used in the recalculation process.

8. The service quality evaluation method based on multi-source data fusion according to claim 7 is characterized in that: The method further comprises: Analyzing multi-source correlations in the anomaly indication data to identify linkage anomalies across data sources; When the linkage anomaly is confirmed, executing an evaluation optimization strategy, the evaluation optimization strategy includes adjusting parameters of a feature cluster analysis process; Re-execute the feature cluster analysis process based on the adjusted parameters; The re-execution result is input into the updating process of the service quality evaluation index.

9. The service quality evaluation method based on multi-source data fusion according to claim 8 is characterized in that: The execution evaluation optimization strategy includes: Set the service data sampling window for iterative training; Evaluating a performance indicator of the feature clustering analysis process during the iterative training; Optimizing the granularity of clustering grouping operations according to the performance indicators; Outputting the optimized feature clustering analysis process to the re-execution process.

10. A service quality evaluation system based on multi-source data fusion and big data analysis, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When executing the computer program, the processor implements the steps of the service quality assessment method based on multi-source data fusion as described in any one of claims 1 to 9 above.

Images