Customer satisfaction feedback collection method and system based on community collaborative filtering

Abstract

The present application relates to the field of business intelligence and data mining technology, specifically to a customer satisfaction feedback collection method and system based on social group collaborative filtering; the method comprises: constructing a customer service experience track by analyzing a service process, and dynamically dividing a service experience social group according to the similarity of behavior, time and abnormal state; generating a stage satisfaction baseline and a reliability index based on the existing feedback data in the social group; combining the social group baseline, the stage reliability and the customer behavior similarity, intelligently screening high-representative customers and determining the best collection time through a collaborative filtering algorithm to achieve accurate reach; for customers who have not fed back, predicting their satisfaction and comparing it with the baseline to determine whether to supplement; the collected feedback is used to update the social group baseline and the analysis model after verification and enhancement, and finally the structured satisfaction conclusion is output through steady-state analysis, sensitivity reconstruction and consistency constraint. The present application effectively improves the feedback collection efficiency, data quality and decision support capability.

Description

Technical Field

[0001] This invention relates to the fields of business intelligence and data mining technology, specifically to a method and system for collecting customer satisfaction feedback based on community collaborative filtering. Background Technology

[0002] With the continuous deepening of customer experience management and the popularization of data-driven decision-making models, enterprises are increasingly demanding real-time, representative, and actionable customer satisfaction feedback. In the context of digitalized service processes and traceable customer behavior, how to efficiently integrate multi-dimensional customer interaction data and extract high-quality experience insights from it has become an important issue.

[0003] Chinese invention patent application CN120525538A discloses an AI-based intelligent customer service interaction method, system, and medium. The method includes: collecting target customer attribute information and behavioral trajectory information, processing this information to obtain a customer profile, extracting personalized needs information of the target customer based on the customer profile, executing intelligent interactive services based on the personalized needs information, monitoring the interactive service process in real time, collecting service status parameters, determining whether to transfer the customer to a human customer service representative based on the service status parameters, monitoring the intelligent interactive service within a preset time period, extracting service indicator data, performing quantitative evaluation based on the service indicator data to obtain a service reliability coefficient, and determining the reliability of the intelligent interactive service and taking corresponding optimization measures, thereby realizing AI-based intelligent customer service interaction technology.

[0004] Collaborative filtering technology, as an effective means of mining group preference patterns, has proven its value in fields such as recommendation, which provides technical inspiration for introducing group similarity analysis into customer satisfaction management. At the same time, the evolution of related technologies such as dynamic community building, time series data analysis, and personalized triggering mechanisms are jointly promoting the development of feedback collection methods towards a more intelligent, accurate, and adaptive direction. Summary of the Invention

[0005] The purpose of this invention is to address the problems existing in the background technology by proposing a method and system for collecting customer satisfaction feedback based on community collaborative filtering.

[0006] The technical solution of this invention: a method for collecting customer satisfaction feedback based on community collaborative filtering, comprising the following specific implementation steps: S1. By analyzing the service process, a customer service experience trajectory is constructed, and service experience communities are dynamically divided based on the similarity of behavioral response, time experience, and abnormal status, generating community service status labels. S2. Based on existing feedback data within the community, perform data collection and preprocessing, stage statistical analysis, baseline construction and dynamic updates to generate a community satisfaction perception baseline that includes stage satisfaction baseline and stage credibility index. S3. Combining the community satisfaction perception baseline, stage credibility and customer behavior trajectory, highly representative customers and key stages are screened by calculating customer contribution and stage triggering factors, and collaborative filtering is used to predict the satisfaction of customers who have not given feedback, and finally the priority of feedback collection objects and triggering strategies are generated. S4. Based on the feedback collection priority and triggering strategy, send personalized feedback requests, receive and verify feedback in real time, and use predicted values ​​and baselines to enhance and complete the credibility data that is missing or has a credibility lower than the set threshold, and update the community satisfaction perception baseline. S5. Perform steady-state analysis, stage sensitivity reconstruction, and community consistency constraint processing on the enhanced community satisfaction data to output a structured comprehensive satisfaction conclusion, and use the conclusion to guide subsequent feedback collection and optimization.

[0007] Preferably, in step S1, constructing the customer service experience trajectory specifically includes: Based on the predefined service process structure of the business system, the service process is broken down into multiple service stages with clear business meanings; Continuously collect customer operational behavior, interaction response, dwell time, and abnormal status information at each service stage; The aforementioned discrete behavioral information is organized and merged according to the order of service stages to form a customer service experience trajectory.

[0008] Preferably, in step S1, dynamically dividing the service experience community specifically includes: Under the premise of ensuring that the service experience trajectories of different customers are in the same service stage, calculate the similarity of customers in behavioral response characteristics, consistency in dwell time, response duration, and consistency in the occurrence of abnormal states; By introducing pre-defined importance weights for service stages, the similarity calculation results of the above three dimensions are weighted and combined to obtain the overall similarity that reflects the consistency of service experience among customers; Based on the overall similarity calculation results, and under the constraint of dynamically adjusted consistency threshold, customers with highly similar experiences are gradually aggregated to form a service experience community.

[0009] Preferably, in step S2, the process of generating the community satisfaction perception baseline specifically includes: Identify and collect customer data that has been provided in the community, and standardize the original feedback values ​​from different sources and formats into values ​​within a preset range; For feedback values ​​that are missing or abnormal, weighted interpolation and adjustment are performed by combining stage weights and abnormal markers; Based on the processed feedback data, the average satisfaction and standard deviation of the community at each service stage were statistically calculated. Analyze the changing trends of satisfaction at each stage over time or as the process progresses, and generate a stage trend index. By combining stage weights, a stage satisfaction baseline vector reflecting the overall community experience pattern is constructed. At the same time, a credibility index is calculated for each stage, which comprehensively reflects the coverage sufficiency and internal consistency of the feedback data in that stage.

[0010] Preferably, in step S3, screening highly representative customers specifically includes: For each customer in the community, calculate the overall similarity between their value and the baseline value of each stage in the community satisfaction perception baseline. By combining credibility metrics at each stage, a customer contribution metric is generated to measure the representativeness of the customer's contribution to the overall satisfaction of the community. All customers in the community are ranked according to the customer contribution index, and customers whose contribution index is higher than the set threshold are selected to form a highly representative customer set, which is then prioritized for feedback collection.

[0011] Preferably, in step S3, the timing for determining the feedback collection specifically includes: For each service stage, a stage triggering factor is calculated based on the stability of the stage satisfaction baseline, the significance of the stage satisfaction change trend, and the level of the stage credibility index. When the triggering factor of a certain service stage exceeds the preset triggering threshold, the stage is determined to be a key stage that needs to collect feedback and is included in the triggering stage set. Feedback collection requests are only sent to the customer within the service phases included in the set of triggering phases.

[0012] Preferably, in step S3, predicting unfeeded customer satisfaction using collaborative filtering specifically includes: For customers who have not yet provided feedback within the community, a neighborhood-based collaborative filtering algorithm is used to predict their satisfaction scores at key service stages based on the similarity of their behavior to that of customers who have provided feedback in their service experience trajectory. The predicted satisfaction score is compared with the community satisfaction baseline for the corresponding service stage, and the difference between the two is calculated. If the difference exceeds a preset difference threshold, it is determined that the experience of the customer who did not provide feedback may deviate significantly from the community baseline, thereby triggering a proactive feedback collection request for that customer.

[0013] Preferably, in step S4, the collaborative enhancement and completion of the feedback data specifically includes: For missing satisfaction data due to lack of customer feedback or stage credibility being lower than the set threshold, the predicted satisfaction value obtained in step S3 and the corresponding stage satisfaction baseline value are used to fill the data through weighted fusion. A confidence index is calculated for the generated satisfaction data to characterize the reliability of the completed data; The actual collected valid feedback data is integrated with the data that has been filled in and completed to form an enhanced community satisfaction matrix.

[0014] Preferably, in step S5, the output of a structured comprehensive satisfaction conclusion specifically includes: Data was extracted from the enhanced community satisfaction matrix, and steady-state analysis was performed on multiple customer feedbacks at each service stage. After removing the influence of extreme values, the steady-state satisfaction value for each stage was obtained. Analyze the historical fluctuation trends and dispersion of satisfaction at each stage, and dynamically determine the weight of each stage in the overall satisfaction assessment through a sensitivity reconstruction mechanism; Calculate the consistency index of the experience distribution within the community, and correct the weighted overall satisfaction. Finally, output a structured conclusion package, which includes at least the overall satisfaction value of the community, a list of key service stages that significantly affect the overall satisfaction, and status labels that reflect the consistency of the experience within the community.

[0015] The technical solution of this invention: A customer satisfaction feedback collection system based on community collaborative filtering, which is used to execute the above-mentioned customer satisfaction feedback collection method based on community collaborative filtering, adopts a cloud-edge collaborative architecture, and includes: The community adaptive building module, deployed on edge nodes, is used to collect customer behavior data in real time, build local service experience trajectories, and form a prototype of a local community. The community baseline and stage credibility generation module, deployed in the cloud, is used to integrate data from multiple edge nodes to generate and maintain a global community satisfaction perception baseline model. The feedback collection object and triggering strategy generation module, deployed at the edge node, is used to generate local feedback collection priorities and triggering instructions based on the baseline model issued by the cloud and the local real-time status. The customer feedback collection and data enhancement module, deployed at edge nodes, is used to perform local feedback request sending and receiving verification, and to perform local enhancement and completion of missing data. The community satisfaction comprehensive analysis and results output module, deployed in the cloud, is used to integrate global enhanced data, perform steady-state analysis, sensitivity reconstruction and consistency verification, and output comprehensive satisfaction conclusions and optimization strategies.

[0016] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects: This invention designs a customer satisfaction feedback collection method and system based on community collaborative filtering. Supported by a cloud-edge collaborative architecture, this method and system effectively improves the intelligence and business adaptability of the feedback collection process. By dynamically constructing customer communities with similar service experience trajectories, the system can accurately identify representative customers and key service stages, achieving low-interference, highly targeted feedback triggering, significantly optimizing customer experience and increasing responsiveness. Simultaneously, by combining collaborative filtering algorithms to predict satisfaction and complete data for customers who haven't provided feedback, the system enhances the coverage and completeness of feedback data, providing a robust foundation for analysis. In the cloud-edge collaborative mode, edge nodes process local behavioral data and collection requests in real time, ensuring low-latency response at the service site. Global data fusion, model training, and strategy optimization are performed in the cloud, achieving closed-loop iteration across regions and the entire process. Finally, through steady-state analysis, sensitivity reconstruction, and consistency constraints, this invention outputs structured satisfaction conclusions, accurately reflecting the experience level and problem stages of each community, providing a reliable basis for service optimization and management decisions. The overall solution achieves a synergistic improvement in feedback collection efficiency, data quality, and analytical depth. Attached Figure Description

[0017] Figure 1 This is a flowchart of a customer satisfaction feedback collection method based on community collaborative filtering proposed in this invention; Figure 2 This is a system architecture diagram of a customer satisfaction feedback collection system based on community collaborative filtering proposed in this invention. Detailed Implementation

[0018] Example 1, as Figure 1 As shown, the present invention proposes a customer satisfaction feedback collection method based on community collaborative filtering, the specific implementation steps of which are as follows: S1. By introducing a customer service experience modeling mechanism with service processes as the core constraint, customer interaction behaviors in specific service scenarios are transformed into alignable and comparable service experience trajectories. Based on this, a customer community structure that can dynamically evolve with service status is constructed, and community service status labels are formed for subsequent satisfaction feedback decisions. This provides a stable, authentic, and business-semantic foundation for the selection of feedback collection objects and collaborative filtering. The specific implementation process is as follows: S11. By analyzing the service process structure, the service process is divided into several service stages with clear business meanings. Within each stage, customer actions, interactions, dwell time, and abnormal states are collected. Discrete behaviors are organized into a service experience trajectory according to the service stage sequence, ensuring that customer behavior is described within a unified service context. This guarantees that subsequent community building has real service comparability. Specifically: When a customer enters a specific service scenario, the service process is first broken down into multiple service stages with clear business meanings according to the predefined service process structure in the business system. Each service stage corresponds to a system state or service node. Without disrupting the normal service process, continuously collect customer interaction behaviors at each service stage and merge them in chronological order to form a customer service experience trajectory. The service experience trajectory of customer i in the current service scenario is defined as follows: ; in, It represents the complete service experience trajectory formed by customer i in a specific service scenario, and is used to describe the customer's behavior and response evolution throughout the service process; This represents the k-th service stage or service node, used to characterize the key state position in the service process. It comes from the service process configuration, business state machine, or system service node definition; K represents the total number of service stages in the current service scenario, the size of which is determined by the complexity of the service process. This indicates that customer i is in the service phase. The behavioral response feature vector is used to comprehensively describe the customer's operation method, interaction depth or problem triggering situation at this stage. It is derived from system operation logs, interaction event records or service behavior records. This indicates that customer i is in the service phase. The dwell time or response time is used to reflect the rhythm of the customer's service experience at this stage, and it is derived from the system timestamp records or the calculation of the event trigger time difference; This indicates that customer i is in the service phase. The status flag indicating whether abnormal, interrupted or unexpected behavior has occurred is used to distinguish between normal and abnormal service experiences. It is derived from the abnormal log, manual intervention record or system error flag. S12. Under the premise of service stage alignment, perform weighted similarity calculation on the behavioral responses, time experience, and abnormal states of different customers within the same service stage. Introduce the importance weight of service stage to highlight the impact of key nodes, thereby forming a similarity metric that reflects the consistency of service experience. This avoids cross-stage or cross-semantic behavioral comparisons and improves the business rationality of community segmentation. Specifically: After completing the service experience trajectory construction, the behaviors of different customers are aligned and compared only under the premise of consistency in service stages. Stage weights are introduced based on the degree of influence of each service stage on the overall experience, and the similarity of customer service experiences is calculated using a weighted average. The similarity of service experience between customer i and customer j is defined as follows: ; in, This represents the similarity of service experience between customer i and customer j in the same service scenario, and is used to measure the degree of consistency of their experience during the service process. Indicates the service stage The importance weight is used to reflect the degree of influence of this stage on the overall service experience, and it is derived from historical service quality analysis, business rule setting or expert experience configuration. This represents the behavioral response similarity function, which measures the similarity of the behavioral characteristics of customer i and customer j at the same service stage. Its input is the behavioral feature vector of the two customers at that stage. This represents a time-experience consistency function, used to measure the similarity of customer response pace or waiting experience within the same service phase; This represents an abnormal state consistency function, used to assess whether customers have experienced the same type of abnormal situation during the service phase, thereby avoiding incorrectly classifying abnormal experiences as similar to normal experiences; It should be noted that the behavioral response similarity function This function is used to measure whether different customers have consistent operating methods and interaction behaviors at the same service stage. Its core idea is to map the multi-dimensional behavioral characteristics generated by customers at this stage to a unified service response semantic space. By comparing the similarity of operation sequence, interaction frequency, function usage combination and problem triggering characteristics, it comprehensively reflects the customer's actual experience of the service at this stage. This function can weaken the interference of individual differences in operating habits on similarity judgment, highlight the commonalities of behavior caused by service process design and service quality itself, thereby ensuring that customer similarity evaluation truly reflects the consistency of service experience. It should be noted that the time experience consistency function This function is used to characterize the similarity of service rhythm and waiting experience among different customers in the same service stage. Its core function is to determine whether customers have experienced similar service smoothness rather than just completing the same operation. By comparing the customer's dwell time, response interval and rhythm changes in the corresponding service stage, this function comprehensively reflects the customer's subjective time perception in the service process, thereby avoiding misjudgment of experience caused by differences in individual operating habits. It should be noted that the consistency function in abnormal states This indicates whether different customers have experienced similar anomalies or deviations from the normal process during the same service phase. Its core function is to distinguish the degree of impact of normal service experience and abnormal service experience on overall similarity. By making consistent judgments on the type of anomaly, duration and recovery method, it avoids incorrectly including experience errors caused by system failure, manual intervention or unexpected interruption in the calculation of normal experience similarity. This ensures that the construction of customer community truly reflects the experience differences of the service itself, rather than the interference of occasional abnormal factors, and improves the stability and interpretability of community division in actual business scenarios. S13. Based on the similarity results of customer service experience, under the constraint of the service consistency threshold, customers with highly similar experiences are gradually aggregated to form a service experience community. The consistency threshold is dynamically adjusted according to the degree of experience differentiation and service status changes within the community, so that the community boundary adapts to the evolution of the service process. Specifically: Based on the similarity results obtained in step S12, multiple service experience communities are gradually built within the same service scenario. Each community uses the average experience trajectory of customers within the current community as a reference center. When the similarity between the experience of the customer to be evaluated and that of the community meets the consistency condition, the customer is included in the community. Defining Customer i and Community Similarity: ; A customer is categorized into the community when the following conditions are met: ; in, Let m represent the m-th service experience community, which is used to represent a set of customers with similar service experience characteristics in the same service scenario; community The number of customers included; Indicates customer i and community The similarity between overall experience features is used to determine whether a customer should be included in the community; community The consistency threshold at time or service stage t is used to limit the minimum requirement for consistent experience within the community. Its initial value comes from historical service stability statistics and can be dynamically adjusted as the service status changes. S14. After the community is built, a corresponding service status label is generated for each service experience community to describe the overall behavioral characteristics, experience dispersion, stage coverage completeness, and stability level of the community. This community status is used as a direct input for the subsequent construction of the satisfaction perception baseline and feedback triggering decisions, so as to achieve a close connection between community results and feedback process. Specifically: After the community is built, a corresponding service status label is generated for each community to characterize the overall service experience of the community at present. This label is represented as follows: ; in, community The service status labeling results are used to comprehensively describe the current overall service experience status of the community. community The average behavioral response characteristics of customers at each service stage are used to characterize the overall behavioral pattern of the community. They are derived from the statistical summary of the behavioral characteristics of each customer in the community. community The degree of dispersion of the service experience within the community is used to reflect whether there is a significant differentiation in the experience within the community. Its value is calculated from the differences in customer experience within the community. community The completeness of the currently covered service stages is used to determine whether the community has experienced a sufficiently complete service process. community Stability metrics are used to reflect the stability of the community structure during the current service phase. They are derived from the frequency of changes in community members and the consistency of experience.

[0019] S2. Based on the customer community and its service status generated in step S1, a quantifiable, interpretable, and dynamically adjustable community satisfaction perception baseline is formed through community feedback data collection, phased statistical analysis, baseline construction, and dynamic updates. This baseline reflects the overall community experience model. The specific implementation process is as follows: S21. Collect and preprocess existing customer feedback data within the community, standardizing feedback from different sources and dimensions into numerical vectors, handling missing or outlier data, and introducing stage weight adjustments to ensure that feedback from key stages makes a significant contribution to the baseline, while ensuring that the data is calculable and comparable, providing reliable input for subsequent baseline analysis. Specifically: Identifying communities The customer collection that has provided feedback Extract the feedback vector for each customer i. : ; Standardize different types of feedback into ; Weighted imputation is used for missing or abnormal feedback values: ; Introduce anomaly markers for anomalous data (such as short-term outlier behavior or system log errors). Adjust the weights to reduce their impact on the baseline: ; in, community The set of customers who have already provided effective feedback; This indicates that customer i is in the service phase. The original feedback value; This represents the standardized feedback value, which maps different sources and dimensions to the range [0,1]. This represents the value after imputation of missing feedback or abnormal data; This indicates the adjusted feedback value; Indicates that customer i is in stage The abnormal flags (such as extreme behavior, atypical operation, system error) take values ​​in [0,1]. S22. Based on preprocessing, calculate the average community satisfaction and standard deviation for each service stage, analyze inter-stage trends, generate stage trend indices and weighted trend characteristics, capture the overall community experience pattern and potential improvement or degradation trends, provide stage-level dynamic information for baseline construction and feedback triggering, and realize the perception of key experience changes in the service process, specifically: For the community Each service stage Calculate the average satisfaction level: ; Standard deviation of satisfaction during the calculation phase: ; Satisfaction sequence of stages Conduct trend analysis to identify patterns of change in the experience during a given period (such as continuous decline, rapid increase, or stabilization), and define the period trend index using the following formula: ; Phase trend and phase weight Combined, a weighted trend feature is generated: ; in, community In the stage Average satisfaction rate; Indicates the community stage Standard deviation of satisfaction; Indicator of phase trend; Indicates weighted trend characteristics; This represents a small constant to prevent division by zero in trend calculations; S23. Based on the results of stage statistics and trend analysis, construct a community satisfaction baseline vector and assign a credibility index to each stage to comprehensively reflect the coverage and consistency of feedback within the community. Low-credibility stages are marked for subsequent collaborative inference and supplementation, forming an interpretable and quantifiable baseline. Specifically: Construct a baseline vector for community satisfaction: ; Calculate credibility for each stage: ; Credibility In stages where the baseline is below a set threshold, the annotation is retained but can be supplemented through collaborative inference in subsequent steps to ensure that the overall baseline is stable and interpretable. in, community The baseline vector of satisfaction; Representation phase Baseline credibility; S24. Introduce a dynamic closed-loop mechanism to update the community's average satisfaction, stage standard deviation, trend index, and credibility in real time when new feedback emerges. This ensures the baseline evolves with service stages and new feedback, achieving continuous availability and operability of the community satisfaction baseline. Specifically: When new feedback Upon arrival, update the average satisfaction level during the community phase: ; Simultaneously update the standard deviation. and credibility ; Trends in stages Recalculate and update the weighted trend features. This ensures that the baseline evolves with the service status; in, This indicates newly arrived customer feedback, used to dynamically update the baseline.

[0020] S3. By combining a collaborative filtering method that integrates community baseline, stage credibility, and customer behavior patterns, the most representative customers are intelligently selected as feedback collection targets, and the collection trigger timing is determined to achieve efficient and low-interference customer satisfaction collection. The specific implementation process is as follows: S31. Calculate the similarity between each customer and the community baseline within the community, generate a customer contribution index based on stage credibility, rank all customers, and select a highly representative customer set to ensure priority collection of customers who are most valuable for community satisfaction. Specifically: Calculate customers Similarity to the community baseline: ; For all customers in the community Sorted by value, highly representative customers are listed first; By threshold Select a highly representative set of customers As a priority for data collection; in, Indicates customer i's relationship with the community Representative contribution indicators; community In the stage Average satisfaction baseline; This indicates the threshold for customer contribution, allowing for the selection of highly representative customers. S32. Based on community stage baselines, trend analysis, and stage credibility calculation, identify key stages with significant experience changes or low baseline credibility, generate a set of trigger stages, and dynamically determine the optimal time for feedback collection. Specifically: For each service stage Calculate the triggering factor: ; It should be noted that the higher the stage trigger factor, the more unstable the baseline or the more significant the change in experience during that stage, and customer feedback should be collected first. Based on the set threshold Generate trigger phase set Feedback requests should only be made at these stages to avoid excessive interference; in, This indicates the stage trigger factor, used to determine whether feedback collection is triggered; This represents the trend weighting coefficient, used to adjust the degree of influence of the phase trend on the triggering decision; Indicates the stage trigger threshold; S33. For customers who do not proactively provide feedback, use a neighborhood collaborative filtering method combined with the similarity of customer behavior trajectories to predict their satisfaction scores, and compare them with the stage baseline to determine whether to trigger proactive data collection, achieving efficient coverage and reducing redundant data collection. Specifically: Community All customers and key stages (triggered stages as determined by step S32) The organization is matrix : ; Utilizing neighborhood-based collaborative filtering: ; To increase the reliability of the prediction results, a community phase baseline is introduced. and stage credibility Adjusted forecast: ; Compare the predicted satisfaction level with the community baseline: ,like (The set threshold) indicates that there may be significant differences among customers who have not provided feedback, triggering proactive feedback collection; otherwise, no intervention is required. in, This indicates that customer i who has not responded is in the stage. Predicted satisfaction; This indicates the difference between the predicted value and the community baseline; Indicates the final predicted satisfaction level; S34. Based on customer contribution and collaborative prediction error, generate customer data collection priorities, select data collection targets according to priority, and dynamically adjust the strategy to update the data collection targets and triggering timings with new feedback and stage trend changes, forming a closed-loop feedback data collection control. Specifically: For all potential customers within the community Calculate the final priority: ; For customers Sort the data from highest to lowest, and use the top N data points as the feedback collection targets to generate a collection plan table; Dynamically adjust the data collection triggering strategy: update the prediction based on new feedback during the execution of each phase. and phase trend Adjust the priority list in real time; in, This indicates the priority of collecting final customer feedback; This represents the weighting coefficient, which balances the impact of customer representativeness and prediction bias on data collection priority.

[0021] S4. Based on the feedback collection objects and triggering stages generated in step S3, an efficient, interpretable, and dynamically enhanced customer satisfaction feedback collection system is achieved through personalized request sending, real-time feedback reception, collaborative prediction and completion, and closed-loop baseline updates. The specific implementation process is as follows: S41. Generate personalized feedback request forms based on customer priority and triggering stage, select the most suitable interaction method (such as questionnaire, pop-up window, or voice), and record the request sending status and initial feedback to lay the foundation for efficient feedback collection. Specifically: Based on customer priority and trigger phase set Generate a feedback request form: ; For each request, the interaction method is automatically adjusted (such as online questionnaires, pop-up ratings, voice interaction). : ; Record sending status, response time, and feedback data. ; in, This represents the customer feedback request form, recording customer i's status at stage [of the process]. Feedback requests and trigger time windows ; This represents customer i's preference for or historical response rate weighted score for interaction method m; M represents the set of available feedback interaction methods, such as online questionnaires, pop-up ratings, and voice interaction. S42. Upon receiving customer feedback, perform real-time verification and standardization, including numerical validity, outlier detection, and integrity checks. Simultaneously, assign weights to outlier data to ensure the validity and reliability of the feedback data. Specifically: Feedback data Perform real-time validation, including: numerical validity checks (whether the score is within the valid range), anomaly detection (submissions that are too fast or extreme values ​​that are highly consistent), and completeness checks (whether required fields are missing). Standardize effective feedback: ; Record and label any detected anomalies or incomplete feedback. So that it can be weighted in baseline updates or collaborative inference; in, Indicates that customer i is in stage The actual feedback value; This indicates that the feedback data has been standardized and mapped to the [0,1] interval. This flag indicates an abnormal or incomplete feedback, and its value is in the range [0,1]. This represents the minimum valid original score value allowed in the customer satisfaction feedback rating system. This represents the maximum valid raw score value allowed in the customer satisfaction feedback rating system; S43. For the stage with no feedback or low confidence, use the collaborative prediction results from step S3 and the baseline weighted fill in the missing feedback, calculate the confidence level and generate an enhanced community satisfaction matrix to achieve data completion and multi-dimensional enhancement, specifically: For customers who did not provide feedback or low credibility stage Use the predicted value from step S3 Fill in: ; By weighing the forecasts against the baseline for each stage, stages with higher confidence rely more on the forecasts, while stages with lower confidence rely more on the baseline. Calculate the confidence level for the filled data: Ensure that the supplementary data has interpretable confidence metrics; Update community satisfaction matrix : ; in, This represents the customer satisfaction value after filling in the blanks. This represents the confidence level of the completed data, reflecting the reliability of the predicted data; This represents the enhanced community satisfaction matrix, which includes actual feedback and supplementary data. community Customer i during the service phase Enhanced satisfaction rating; S44. The collected and enhanced data will be used to update the community stage baseline, standard deviation, credibility, and trend, forming a dynamic closed loop. This allows the community satisfaction model to evolve over time, providing a robust foundation for subsequent analysis and prediction. Specifically: Updated community phase baseline: ; Update phase standard deviation With credibility ; Update phase trend And adjust the collection priority list for step S3 based on the new feedback; Generate a complete augmented data matrix and the updated community baseline; in, This represents the updated baseline average satisfaction value for the community phase. This represents the enhanced customer satisfaction value, i.e., the enhanced customer i at stage [i]. The satisfaction score.

[0022] S5. Based on the enhanced community satisfaction data generated in step S4, through steady-state analysis, stage sensitivity reconstruction, community consistency constraints, and structured output of results, discrete and dynamic customer feedback is transformed into stable, interpretable, and reusable comprehensive satisfaction conclusions. The analysis results are then used in reverse to optimize the system operation, achieving closed-loop optimization of customer satisfaction assessment and feedback collection strategies. The specific implementation process includes: S51. A steady-state analysis is performed on the enhanced community stage satisfaction data. By suppressing anomalies and extracting central tendency from multiple customer feedback within a stage, steady-state satisfaction results that accurately reflect the overall experience level of the community are obtained. Simultaneously, stage reliability information is retained to lay a reliable foundation for subsequent analysis. Specifically: With community In units, from the enhancement matrix Extracting the stage vector: ; Steady-state constraint aggregation is performed on the vectors of each stage, using the "median interval truncation mean" method: first, extreme values ​​within the upper and lower abnormal intervals are removed, and then the steady-state satisfaction value is calculated. ; Steady-state satisfaction and stage credibility Joint annotation is performed to form stage steady-state satisfaction pairs: ; in, This indicates that the m-th community is in stage m. The satisfaction vector; This indicates that the m-th community is in stage m. The steady-state satisfaction value; This indicates that the m-th community is in stage m. The credibility of the stage; S52. Based on steady-state satisfaction, a stage-sensitivity reconstruction mechanism is introduced. This mechanism dynamically adjusts the weight of different service stages in overall satisfaction, considering both stage fluctuation trends and dispersion. This ensures that key stages with frequent experience changes or concentrated problems naturally stand out in the comprehensive analysis, avoiding distorted judgments caused by fixed weights. Specifically: Based on historical fluctuation trends and stage dispersion Sensitivity factor during calculation stage: ; Normalize the stage sensitivity to obtain the set of reconstruction weights. ; Mapping steady-state satisfaction to the service target space generates a stage contribution vector: ; in, Indicates the stage sensitivity factor; This represents the stage sensitivity mapping function; This indicates that the m-th community is in stage m. Reconstruction weights; This represents the contribution value to stage satisfaction. It should be noted that the stage sensitivity mapping function This function is used to convert the satisfaction trends and dispersion of the community at each service stage into a unified stage sensitivity value. It automatically increases the weight of important stages based on the fluctuation range of the stage experience and the differences in customer feedback, ensuring that key stages are fully reflected in the comprehensive satisfaction analysis, while suppressing the excessive influence of stable stages on the overall results. This allows the stage weights to adaptively reflect the criticality and differences of the real service experience, improving the accuracy and interpretability of the analysis results. S53. By introducing community consistency constraints, the stage contribution results are corrected as a whole, taking into account the concentration of experience distribution within the community, preventing a few extreme feedbacks or local fluctuations from dominating the final conclusion, thus obtaining a comprehensive community satisfaction assessment result that balances overall level and internal stability, specifically: Calculate the overall community consistency index: ; Constructing an overall satisfaction score: ; If the consistency is lower than the preset threshold, the community is marked as a "differentiated experience community" and a structured prompt is output instead of a single numerical conclusion. in, This represents the consistency index of the m-th community; This represents the preset maximum dispersion reference value; This represents the final overall satisfaction value of the m-th community; S54. The final satisfaction analysis results are structured and output to form an information set including comprehensive score, key impact stages, consistency status, and improvement directions. This set is then synchronously provided to management decision-making, business optimization, and system adaptive scheduling modules as an important input for the next round of feedback collection and community modeling, realizing a continuously evolving satisfaction feedback mechanism. Specifically: Generate a results package for each community: {Overall satisfaction score} Key impact stages set, community consistency state label}; Map the results to a multi-level output format: Management side: Trend changes, risk community alerts; Business side: Suggestions for improvement at each stage; System side: Serves as input for the priority of data collection in the next round of step S3.

[0023] Example 2, as Figure 2 As shown, the present invention proposes a customer satisfaction feedback collection system based on community collaborative filtering. It adopts a cloud-edge collaborative architecture, including a central platform deployed in the cloud and edge processing units deployed at edge nodes in various service locations. These units work together to execute the customer satisfaction feedback collection method based on community collaborative filtering proposed in Embodiment 1. Specifically, the system includes: a community adaptive construction module, a community baseline and stage credibility generation module, a feedback collection object and triggering strategy generation module, a customer feedback collection and data enhancement module, and a community satisfaction comprehensive analysis and result output module.

[0024] The community adaptive construction module is deployed on edge nodes and is responsible for collecting customer behavior data in the service area in real time, constructing local service experience trajectories, calculating the initial similarity between customers, forming a prototype of local community, and synchronizing the processed trajectory features and local community summary information to the cloud. The community baseline and stage credibility generation module is deployed in the cloud. It receives summary information from various edge nodes, performs cross-regional and full-volume data fusion and in-depth analysis, generates and maintains a globally unified community satisfaction baseline, stage credibility and trend model, and distributes the updated baseline model to relevant edge nodes. The feedback collection object and triggering strategy generation module is deployed on edge nodes. Based on the baseline model issued by the cloud and the local real-time service status, it calculates the customer contribution and stage triggering factors locally, and performs collaborative filtering prediction to dynamically generate a personalized feedback collection priority list and triggering instructions for the service area. The customer feedback collection and data enhancement module is deployed on edge nodes to perform local personalized feedback request sending, real-time reception and verification; for missing data, it uses the baseline distributed from the cloud and the local prediction results to enhance and complete the data, forming local enhanced data, and synchronizes the key data to the cloud. The community satisfaction comprehensive analysis and results output module is deployed in the cloud. It integrates the enhanced data uploaded by all edge nodes, performs global steady-state analysis, sensitivity reconstruction and consistency verification, generates the final comprehensive satisfaction conclusion and global optimization strategy, and distributes the updated strategy model to the edge nodes to guide the next round of edge side data collection.

[0025] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for collecting customer satisfaction feedback based on community collaborative filtering, characterized in that, The specific implementation steps include the following: S1. By analyzing the service process, a customer service experience trajectory is constructed, and service experience communities are dynamically divided based on the similarity of behavioral response, time experience, and abnormal status, generating community service status labels. The dynamic division of service experience communities specifically includes: Under the premise of ensuring that the service experience trajectories of different customers are in the same service stage, calculate the similarity between customers in terms of behavioral response characteristics, time experience, and occurrence of abnormal states; By introducing pre-defined importance weights for service stages, the similarity calculation results of the three dimensions of behavioral response characteristics, time experience, and occurrence of abnormal states are weighted and integrated to obtain the overall similarity that reflects the consistency of service experience among customers. Based on the overall similarity calculation results, and under the constraint of dynamically adjusted consistency threshold, customers with highly similar experiences are gradually aggregated to form a service experience community; S2. Based on existing feedback data within the community, perform data collection and preprocessing, stage statistical analysis, baseline construction and dynamic updating to generate a community satisfaction perception baseline that includes a stage satisfaction baseline vector and a stage credibility index. The process of generating the community satisfaction perception baseline specifically includes: Identify the set of customers who have provided feedback within the community, extract the original feedback values ​​of each customer at each service stage to form a feedback vector; standardize the feedback values ​​from different sources and dimensions; For feedback values ​​that are missing or deemed abnormal, weighted imputation is performed using standardized feedback values ​​from other customers in the same service stage within the community to fill in the missing or abnormal data. For data generated due to short-term out-of-group behavior or system recording errors, anomaly markers are introduced to adjust the weight of such data, reducing the contribution of abnormal data to subsequent baseline calculations, and obtaining adjusted feedback values. Based on the adjusted feedback values, the average satisfaction of the community at each service stage is statistically calculated; at the same time, the standard deviation of satisfaction at each service stage is calculated, and trend analysis is performed on the average satisfaction sequence of each service stage to identify the change pattern of stage experience. A stage trend index is generated by comparing the average satisfaction change between adjacent stages; the stage trend index of each service stage is combined with the preset stage importance weight to generate a weighted trend feature. Based on the average satisfaction level of each stage, a stage satisfaction baseline vector is constructed. A credibility index is calculated for each stage by combining the number of customers who have provided feedback, the total number of community members, and the stage standard deviation. The baseline vector and the credibility index are used together as the community satisfaction perception baseline. Stages with credibility levels below the threshold are marked. S3. Combining the community satisfaction perception baseline, stage credibility and customer behavior trajectory, highly representative customers and key stages are screened by calculating customer contribution and stage triggering factors, and collaborative filtering is used to predict the satisfaction of customers who have not given feedback, and finally the priority of feedback collection objects and triggering strategies are generated. The selection of highly representative clients specifically includes: For each customer in the community, the actual feedback value of the customer at each service stage is compared with the baseline value of the corresponding stage in the community satisfaction baseline vector. The degree of difference is weighted using the credibility index of each service stage, so that the stage with higher credibility has a larger proportion in the similarity calculation. The weighted results of all stages are summed and divided by the sum of the credibility index of each stage for normalization, so as to obtain the contribution index of the customer to the overall satisfaction of the community, i.e., the customer contribution. All customers in the community are sorted from highest to lowest according to their contribution index. Customers with higher contribution indexes indicate that their service experience is more representative of the overall experience of the community. Customers with contribution indexes above a preset threshold are selected to form a highly representative customer set, and customers in this set are given priority for subsequent feedback collection. S4. Based on the feedback collection priority and triggering strategy, send personalized feedback requests, receive and verify feedback in real time, and use predicted values ​​and baselines to enhance and complete the credibility data that is missing or has a credibility lower than the set threshold, and update the community satisfaction perception baseline. S5. Perform steady-state analysis on the enhanced community satisfaction data to extract steady-state satisfaction values ​​for each stage. Based on the steady-state satisfaction, dynamically adjust the influence weight of each stage in the overall satisfaction through stage sensitivity reconstruction. Introduce community consistency constraints to correct the stage contribution results as a whole, output a structured comprehensive satisfaction conclusion, and use the conclusion to guide subsequent feedback collection and optimization.

2. The customer satisfaction feedback collection method based on community collaborative filtering according to claim 1, characterized in that, In step S1, constructing the customer service experience trajectory specifically includes: Based on the predefined service process structure of the business system, the service process is broken down into multiple service stages with clear business meanings; Continuously collect customer operational behavior, interaction response, dwell time, and abnormal status information at each service stage; The customer's operational behavior, interactive response, dwell time, and abnormal status information in each service stage are organized and merged according to the order of the service stages to form the customer's service experience trajectory.

3. The customer satisfaction feedback collection method based on community collaborative filtering according to claim 2, characterized in that, In step S3, the specific timing for determining the feedback collection includes: For each service stage, a stage triggering factor is calculated based on the stability of the stage satisfaction baseline, the significance of the stage satisfaction change trend, and the level of the stage credibility index. When the triggering factor of a certain service stage exceeds the preset triggering threshold, the stage is determined to be a key stage that needs to collect feedback and is included in the triggering stage set. Feedback collection requests are only sent to the customer within the service phases included in the set of triggering phases.

4. The customer satisfaction feedback collection method based on community collaborative filtering according to claim 3, characterized in that, In step S3, predicting unreported customer satisfaction using collaborative filtering specifically includes: For customers who have not yet provided feedback within the community, a neighborhood-based collaborative filtering algorithm is used to predict their satisfaction scores at key service stages based on the similarity of their behavior to that of customers who have provided feedback in their service experience trajectory. The predicted satisfaction score is compared with the community satisfaction baseline for the corresponding service stage, and the difference between the two is calculated. If the difference exceeds a preset difference threshold, it is determined that the experience of the customer who did not provide feedback may deviate significantly from the community baseline, thereby triggering a proactive feedback collection request for that customer.

5. The customer satisfaction feedback collection method based on community collaborative filtering according to claim 4, characterized in that, Step S4, specifically the collaborative enhancement and completion of the feedback data, includes: For missing satisfaction data due to lack of customer feedback or stage credibility being lower than the set threshold, the predicted satisfaction value obtained in step S3 and the corresponding stage satisfaction baseline value are used to fill the data through weighted fusion. A confidence index is calculated for the generated satisfaction data to characterize the reliability of the completed data; The actual collected valid feedback data is integrated with the data that has been filled in and completed to form an enhanced community satisfaction matrix.

6. The customer satisfaction feedback collection method based on community collaborative filtering according to claim 5, characterized in that, In step S5, the output of the structured overall satisfaction conclusion specifically includes: Steady-state analysis is performed on the enhanced community satisfaction data, namely: extracting the satisfaction vector corresponding to each service stage from the enhancement matrix on a community-by-community basis; using the median interval truncation mean method for each stage vector, first removing extreme values ​​located in the upper and lower abnormal intervals, and then calculating the mean of the remaining data as the steady-state satisfaction value for that stage; and jointly labeling the steady-state satisfaction value with the stage credibility to form a stage steady-state satisfaction pair. The stage sensitivity factor is calculated based on the historical fluctuation trend and dispersion of each service stage. The stage sensitivity factor is normalized to obtain the reconstruction weight of each stage. The steady-state satisfaction is weighted using the reconstruction weight to generate the satisfaction contribution value of each stage. Calculate the overall community consistency index, and use the cumulative result of the stage satisfaction contribution value to adjust the consistency index to obtain the community's comprehensive satisfaction value; when the consistency index is lower than the preset threshold, the community is marked as a differentiated experience community. Generate a result package for each community, including an overall satisfaction score, a set of key impact stages, and a community consistency status label. Map this result package to trend change and risk community alerts for management, stage improvement suggestions for business, and collection priority inputs for the system.

7. A customer satisfaction feedback collection system based on community collaborative filtering, used to execute the customer satisfaction feedback collection method based on community collaborative filtering as described in any one of claims 1 to 6, characterized in that, The cloud-edge collaborative architecture is adopted, including: The community adaptive building module, deployed on edge nodes, is used to collect customer behavior data in real time, build local service experience trajectories, and form a prototype of a local community. The community baseline and stage credibility generation module, deployed in the cloud, is used to integrate data from multiple edge nodes to generate and maintain a global community satisfaction perception baseline model. The feedback collection object and triggering strategy generation module, deployed at the edge node, is used to generate local feedback collection priorities and triggering instructions based on the baseline model issued by the cloud and the local real-time status. The customer feedback collection and data enhancement module, deployed at edge nodes, is used to perform local feedback request sending and receiving verification, and to perform local enhancement and completion of missing data. The community satisfaction comprehensive analysis and results output module, deployed in the cloud, is used to integrate global enhanced data, perform steady-state analysis, sensitivity reconstruction and consistency verification, and output comprehensive satisfaction conclusions and optimization strategies.

Images