A method for predicting user real-time satisfaction based on product KEI index

Abstract

The application relates to the technical field of user experience evaluation and artificial intelligence, and discloses a method for predicting real-time user satisfaction based on product KEI indexes, which comprises the following steps: S1, collecting key experience index (KEI) data in the product and user interaction process, collecting product satisfaction evaluation data, and collecting user feature data; through real-time collection and preprocessing of the KEI indexes, the application adopts an isolated forest algorithm to remove abnormal values, delete repeated data and combine Z-score standardization, so that data noise is effectively reduced; through random forest extraction of key indexes and combination of correlation analysis and clustering analysis, an input feature set of single indexes and combined indexes is constructed, the prediction accuracy of the model is significantly improved, a neural network optimization algorithm based on the principle of star radiation mapping is adopted, global parameter search is carried out through simulation of the nonlinear random propagation of cosmic rays, and the problems of gradient disappearance and local optimization are effectively avoided.

Description

Technical Field

[0001] This invention relates to the field of user experience evaluation and artificial intelligence, specifically a method for predicting real-time user satisfaction based on product KEI metrics. Background Technology

[0002] With the widespread application of smart devices and network services, the interaction between users and products has become increasingly complex and diverse, making user experience a core element of product competitiveness. In fields such as security products and networking products, user satisfaction directly impacts market acceptance and user retention. To accurately assess and improve user satisfaction, companies typically analyze user perceptions and evaluations by collecting key experience indicators (KEIs) during product use. These KEIs include Wi-Fi signal strength, video loading latency, network error rate, and device response time, directly reflecting product performance and user experience.

[0003] Currently, various user satisfaction prediction schemes have been proposed in existing technologies. For example, Chinese invention patent application CN202111324453.6 proposes a user satisfaction prediction method, device, and electronic device. This method acquires feature data and satisfaction tags of multiple historical users within a first time period. The feature data includes network data, user basic data, and network quality data. For each satisfaction tag, it determines the fluctuation degree of the number of historical users before the first time period, thereby determining the preset algorithm type. The prediction model is then trained based on the algorithm type and training data. Finally, the satisfaction prediction result is determined based on the feature data of the target user. This method acquires factors affecting satisfaction prediction from multiple different dimensions, thus improving the model's prediction accuracy to a certain extent.

[0004] For example, Chinese invention patent application CN202410618482.0 proposes a method and apparatus for evaluating comprehensive user satisfaction. This method derives a user network satisfaction index based on implicit and explicit user network satisfaction indices, a user service satisfaction index based on implicit and explicit user service satisfaction indices, and a user service satisfaction index based on implicit and explicit user service satisfaction indices. Finally, it synthesizes these three dimensions of satisfaction indices to obtain a comprehensive user satisfaction score. This method incorporates implicit data into the evaluation system, achieving comprehensive coverage of the dimensions of actual user-encountered problems in the satisfaction score.

[0005] For example, Chinese invention patent application CN202410531047.4 proposes a user satisfaction prediction method and device based on artificial intelligence big data analysis. The method acquires target data associated with vehicles, determines a weighted average value based on a weighted index model, the weighted index model is determined based on historical vehicle data, and then determines user satisfaction based on the weighted average value. The higher the weighted average value, the higher the user satisfaction value, and vice versa.

[0006] However, the aforementioned existing technologies still have the following shortcomings:

[0007] Existing technologies, when collecting user interaction experience data, lack in-depth analysis of the dependencies and correlations between multi-dimensional Key Experience Index (KEI) metrics, and cannot achieve real-time processing of multi-dimensional KEI metrics. The collected data contains a large number of outliers and duplicate data, and the metrics influence each other. Furthermore, the lack of effective standardization processing limits the accuracy of user satisfaction predictions during model training due to limitations in data integrity and quality, resulting in significant data noise.

[0008] Existing neural network algorithms are prone to gradient vanishing, gradient exploding, or getting stuck in local optima when processing high-dimensional and complex data. They lack effective global optimization mechanisms, resulting in unstable performance during model training. This makes it difficult to meet the complex real-time user satisfaction prediction requirements and achieve accurate real-time prediction. Therefore, we propose a method for predicting real-time user satisfaction based on the product KEI index to solve the above-mentioned problems. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention provides a method for predicting real-time user satisfaction based on product KEI indicators. This method solves the problems of existing technologies, such as not considering the correlation between indicators when collecting multi-dimensional KEI data, lacking real-time processing capabilities leading to high data noise, and traditional neural networks being prone to gradient vanishing and local optima.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting real-time user satisfaction based on product KEI indicators, comprising the following steps:

[0011] S1. Collect key experience indicators (KEI) data in real time during the product-user interaction process, collect product satisfaction evaluation data, and collect user characteristic data.

[0012] S2. Preprocessing and feature extraction are performed on the KEI data. The preprocessing includes missing value deletion, outlier detection and correction, duplicate value deletion, and normalization. The feature extraction includes using a random forest model to extract single key KEI indicators, and performing correlation analysis and cluster analysis on the extracted key KEI indicators to identify the combined features of negative experiences and form a KEI combined feature set.

[0013] S3. Using the preprocessed and feature-extracted KEI data, the product satisfaction evaluation data, and the user feature data, train the satisfaction prediction model constructed based on the neural network algorithm; during the training process, use an optimization algorithm based on the principle of interstellar ray mapping to update the parameters of the neural network. The optimization algorithm performs a global search in the parameter space by simulating the nonlinear random propagation process of cosmic rays to find the optimal parameters.

[0014] S4. Using the trained satisfaction prediction model, output the real-time user satisfaction prediction value based on the real-time collected KEI data and user characteristic data.

[0015] S5. Based on the predicted real-time user satisfaction value, perform dynamic early warning and feedback optimization, collect subjective user satisfaction feedback to iteratively optimize the satisfaction prediction model, and convert the prediction results into optimization instructions for product operating parameters and feed them back to the user terminal device to achieve real-time improvement of user experience.

[0016] Beneficial effects

[0017] This invention provides a method for predicting real-time user satisfaction based on product KEI (Key Opinion Quality) indicators. Compared with existing technologies, it has the following advantages:

[0018] This method for predicting real-time user satisfaction based on product KEI metrics accurately identifies key indicators and combinations of indicators affecting user satisfaction through real-time collection and analysis of KEI metrics, ensuring the real-time nature and accuracy of user experience data. The handling of outlier and duplicate data improves data quality, and Z-score standardization allows indicators of different dimensions to be calculated within a unified numerical range, thereby improving the model's predictive accuracy.

[0019] The neural network parameter optimization algorithm, based on the principle of interstellar radiation mapping, effectively avoids the vanishing gradient and local optima problems that occur during model training, ensuring that the neural network can find the optimal solution under different conditions. This method ensures the robustness and stability of the model by dynamically adjusting the diffusion intensity and mutation operation, thereby improving the accuracy of real-time user satisfaction prediction and effectively handling complex network and user interaction environments. Attached Figure Description

[0020] Figure 1 This is a flowchart of a method for predicting real-time user satisfaction based on product KEI index according to the present invention;

[0021] Figure 2 This is a flowchart of the neural network algorithm optimized based on the principle of interstellar radiative mapping in this invention;

[0022] Figure 3 This is a comparison of the loss curves of the optimization algorithm based on the interstellar radiation mapping principle of this invention with the existing SGD optimization algorithm and Adam optimization algorithm;

[0023] Figure 4 This is a comparison of the accuracy curves of the optimization algorithm based on the principle of interstellar radiation mapping in this invention with the existing SGD optimization algorithm and Adam optimization algorithm. Detailed Implementation

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

[0025] like Figure 1-4 As shown:

[0026] A method for predicting real-time user satisfaction based on product KEI metrics includes the following steps:

[0027] S1. Obtain key experience metrics (KEI) data, product satisfaction evaluation data, and user characteristic data during the user's interaction with the product.

[0028] Preferably, KEI data includes at least one of the following: user app opening latency, live video loading latency, cloud playback loading latency, WiFi wall penetration performance, WiFi access waiting time, WiFi dropout rate, WiFi signal strength, and error rate. Product satisfaction evaluation data includes user experience ratings and subjective feedback across dimensions such as overall perception, cost, cost-effectiveness, appearance, hardware, and stability. User characteristic data includes user ordering behavior, user emotions, user complaint behavior, and user product usage frequency.

[0029] In one specific implementation, step S1 further includes: defining a calculation method for each KEI metric. For example, video live streaming loading latency is defined as the time interval from when a user requests to watch the live video to when the first frame of the video is successfully rendered and displayed to the user on the client. KEI metric calculation modules are developed for security products and networking products, and metric calculation code is embedded in the products to achieve real-time calculation of experience metrics. Data calculation is completed within each metric calculation cycle, and data is reported to the data platform in real time to complete data collection.

[0030] S2. Preprocess and extract features from KEI data. Preprocessing includes missing value removal, outlier detection and correction, duplicate value removal, and normalization. Feature extraction includes using a random forest model to extract single key KEI indicators, and performing correlation analysis and cluster analysis on the extracted key KEI indicators to identify the combined features of negative experiences and form a KEI combined feature set.

[0031] S2 specifically includes the following sub-steps:

[0032] S21. Check the KEI data for missing values ​​and delete the data records containing missing values ​​directly. In this step, missing values ​​are handled by direct deletion because filling in missing values ​​would affect the accuracy of the model.

[0033] S22. The Isolation Forest algorithm is used to identify outliers in KEI data. System error outliers with temporal clustering characteristics are deleted, while outliers caused by business rule issues are corrected numerically. The Isolation Forest algorithm is an anomaly detection method suitable for high-dimensional data. Its basic principle is to construct multiple isolated trees by randomly selecting features and split points. Outliers, having fewer and more distinct feature values, are more easily isolated and thus identified.

[0034] S23. Define duplicate data from the same time, the same user, and the same metric as duplicate values, and retain only one data record.

[0035] S24. The KEI data after deleting outliers and duplicates is normalized using the Z-score normalization method.

[0036] S25. Calculate the correlation between each Key Evaluation Index (KEI) and user satisfaction using a random forest model, and extract KEI indicators with a correlation higher than a first threshold as single key KEI indicators. The random forest model is an ensemble learning algorithm that effectively assesses the importance of each feature to the target variable by constructing multiple decision trees and combining their prediction results.

[0037] S26. Conduct correlation analysis among individual key performance indicators (KEIs). Combine multiple indicators with correlation coefficients higher than the second threshold. After classifying satisfaction scores, perform cluster analysis on the combinations of indicators within each level to identify negative experience combinations with clustering characteristics in the low-satisfaction user group, forming a KEI combination feature set. For example, experiments found that in the low-satisfaction user group, users with the combination of high video access latency and severe video buffering accounted for 83% of the total number of low-satisfaction users. This combination feature can serve as an important basis for satisfaction prediction.

[0038]

[0039] in, and These are the first two KEI indicators. Each sample value and These are the averages of the two KEI indicators. The sample size. Correlation coefficient. The value range is [−1, 1], and the closer the absolute value is to 1, the stronger the correlation.

[0040] In one specific implementation, the correlation analysis results between KEI indicators are shown in Table 1 below:

[0041] Based on the above analysis, KEI-1 and KEI-2 are strongly correlated with user satisfaction, and are therefore identified as important KEI features. Further analysis of the influence coefficients between key indicators reveals that when two indicators are interdependent and both affect satisfaction results, indicators with stable data collection and high influence coefficients are selected. The impact of combinations of indicators on satisfaction results is also considered.

[0042] S3. Using the preprocessed and feature-extracted KEI data, product satisfaction evaluation data, and user characteristic data, train the satisfaction prediction model built on the neural network algorithm. During the training process, an optimization algorithm based on the principle of interstellar ray mapping is used to update the parameters of the neural network. The optimization algorithm performs a global search in the parameter space by simulating the nonlinear random propagation process of cosmic rays to find the optimal parameters.

[0043] In this invention, a 6-layer fully connected neural network is preferably used for feature extraction and satisfaction prediction. The neural network structure includes an input layer, multiple hidden layers, and an output layer. The number of nodes in the input layer corresponds to the number of input features, and the number of nodes in the output layer corresponds to the number of satisfaction rating categories or regression output values. The number of nodes in the hidden layers is set according to the data scale and complexity.

[0044] This invention employs a neural network parameter optimization algorithm based on the principle of interstellar radiative mapping, inspired by the propagation patterns of cosmic rays in the Milky Way. Cosmic rays exhibit highly nonlinear and random propagation; utilizing this radiative propagation model to simulate and optimize the parameter update process in a neural network allows for finding the global optimum under different network states, rather than just local optima. Based on the random and nonlinear exploration mechanism, this invention allows the algorithm to perform extensive searches in the parameter space, effectively avoiding the problem of traditional gradient descent methods easily getting trapped in local optima. Furthermore, by simulating the diffusion characteristics of cosmic rays, the algorithm based on the principle of interstellar radiative mapping can adaptively adjust the search intensity and direction during iteration, making the algorithm more flexible and efficient in finding the optimal parameters of the neural network.

[0045] The optimization algorithm in S3, based on the principle of interstellar radiation mapping, updates the parameters of the neural network, specifically including the following sub-steps:

[0046] S31. Initialize the weights of the neural network. and bias And set the initial diffusion vector. The initial diffusion vector represents the initial random propagation state of the cosmic ray.

[0047] The initialization method is represented as follows:

[0048]

[0049]

[0050]

[0051] in, It is a hyperparameter that controls the initial diffusion intensity, and is preferably set to 0.1; This indicates random sampling from a standard normal distribution;

[0052] S32. In each iteration, multiple diffusion points are randomly generated based on the current state of the network parameters. Each diffusion point corresponds to a set of potential neural network parameters, and the parameters are updated as follows:

[0053]

[0054]

[0055] in, and The first Weights and biases for the next iteration τ is the dynamic diffusion factor, and τ is the time decay factor, preferably set to 0.95. This represents the current iteration number. The dynamic diffusion factor helps control the non-linear decrease in the strength of random diffusion with each iteration, ensuring the algorithm gradually stabilizes.

[0056] Dynamic diffusion factor Calculate according to the following formula:

[0057]

[0058] in, The initial diffusion rate, For attenuation control parameters, The nonlinear intensity of the decay;

[0059] 33. Map the updated parameter set back to the parameter space using a mapping function that includes a nonlinear transformation. The mapping process simulates the path changes after cosmic rays interact with interstellar matter, and the mapping result represents the potential update of network weights and biases.

[0060] Mapping function Represented as:

[0061]

[0062]

[0063] in, As the attenuation factor, The first nonlinear mapping parameter, This is the second nonlinear mapping parameter, preferably set to 0.3. It is a hyperbolic tangent function, which increases the nonlinearity of the mapping process and simulates the complex dynamic changes of parameters in high-dimensional space;

[0064] S34. Calculate the loss function value of each mapped parameter set on the training set, and select the parameter set with the optimal loss function value.

[0065] loss function Calculate according to the following formula:

[0066]

[0067] in, For neural network models, For input data, For tags, The number of training samples. It is the L2 norm. The gradient represents the change in the weight parameters of the neural network. Here, ρ is the regularization coefficient, and ρ(⋅) is the nonlinear error function;

[0068] The nonlinear error function ρ(⋅) is defined as:

[0069] when hour, ;

[0070] when hour, ;

[0071] in, = ;

[0072] This nonlinear error function allows for a reduction in the impact of large error values, avoiding excessive influence from outlier data and making the model more robust to data noise.

[0073] S35. Perform a cross operation on the selected sets of multiple excellent parameters to generate a new set of parameter candidates;

[0074] Crossover operations are performed according to the following formula:

[0075]

[0076]

[0077] in, It is a cross factor; and These are two weight parameters in the weight set of the mapped neural network; and These are two bias parameters in the weight set of the mapped neural network; This is the set of weight parameters obtained from the crossover operation; This is the set of bias parameters obtained from the crossover operation;

[0078] Variation intensity Adjust dynamically according to the following formula:

[0079]

[0080] in, It is the initial mutation strength; This represents the total number of iterations, preferably set to 1000.

[0081] Repeat steps S32 to S36 until a preset stopping iteration condition is met, and output the optimized neural network parameters. The preset stopping iteration condition is reaching a preset maximum number of iterations; preferably, the preset maximum number of iterations is set to 1000.

[0082] S4. Using the trained satisfaction prediction model, output the real-time user satisfaction prediction value based on the real-time collected KEI data and user characteristic data.

[0083] S5. Based on the real-time user satisfaction prediction value, perform dynamic early warning and feedback optimization, collect users' subjective satisfaction feedback to iteratively optimize the satisfaction prediction model, and convert the prediction results into optimization instructions for product operation parameters and feed them back to the user terminal device to achieve real-time improvement of user experience.

[0084] Preferably, in step S5, the dynamic early warning and feedback optimization includes: setting user characteristic scoring range and product scoring range; when the predicted value of user real-time satisfaction deviates from the user characteristic scoring range or product scoring range, triggering a system early warning and shortening the user's prediction indicator collection cycle; and restoring to normal status after the monitoring indicators stabilize.

[0085] More preferably, step S5 further includes: performing statistical processing on the product satisfaction evaluation data to calculate the normal scoring range of product satisfaction, comparing and verifying the calculated normal scoring range with the actual situation, and adjusting the normal scoring range according to business development and market changes.

[0086] In one specific implementation, after processing KEI data in real time, the system analyzes and optimizes the results in conjunction with key parameters from the product's backend. These optimized parameters are then fed back to the user's terminal device through the backend. The terminal software automatically adjusts the relevant parameters according to preset rules to achieve near real-time performance improvements. For example, the system can optimize key indicators such as WiFi signal strength or video loading latency in real time based on the user's network status. This allows for immediate improvement in user response and experience, ensuring that system adjustments are promptly reflected in the user's usage and enhancing overall user satisfaction and experience.

[0087] Preferably, subjective satisfaction feedback from users is collected through a real-time survey mechanism that evaluates products immediately after use. This subjective satisfaction feedback is then used as validation data to assess and iteratively optimize the prediction accuracy of the satisfaction prediction model.

[0088] In one specific implementation, user feature data is preprocessed before being input into the model, including uniform encoding and numerical conversion of the user feature data; user satisfaction rating is used as the target variable, the preprocessed user feature data is input into the random forest association analysis model, user feature indicators with the greatest impact on user satisfaction rating are extracted, and the extracted user feature indicators are verified and then input into the satisfaction prediction model as user feature data.

[0089] The products are security products or networking products.

[0090] To verify the effectiveness of the technical solution of this invention, comparative experiments were conducted. The comparative experimental data of the optimization algorithm based on the interstellar radiation mapping principle proposed in this invention with existing SGD and Adam optimization algorithms are shown in Table 2 below:

[0091]

[0092] Experimental results show that the algorithm proposed in this invention has significantly higher AUC values ​​than existing technologies on both the training and test sets, indicating that this invention has a clear advantage in prediction accuracy.

[0093] Example 1;

[0094] This invention provides a method for predicting real-time user satisfaction based on product KEI indicators, applicable to security products.

[0095] S1. Real-time acquisition of KEI data.

[0096] The KEI indicator calculation method was developed, specifically including:

[0097] Video live streaming loading latency is the time interval from when a user requests to watch a live video stream to when the first frame of the video is successfully rendered and displayed to the user on the client side.

[0098] WiFi signal strength, the wireless signal strength between the user terminal device and the router, in dBm;

[0099] Video playback success rate, the ratio of the number of successful video playbacks to the number of playback attempts;

[0100] User app opening delay is the time interval from when the user clicks the app icon to when the app's main interface is fully loaded.

[0101] Develop a KEI (Key Experience Index) calculation module for smart camera and networking products, embedding the index calculation code into the products to achieve real-time calculation of experience indicators. Complete data calculation within each index calculation cycle and report to the data platform in real time.

[0102] At the same time, product satisfaction evaluation data is collected, including customer experience ratings and subjective feedback on dimensions such as overall perception, cost, cost-effectiveness, appearance, hardware equipment, and stability of use.

[0103] At the same time, user characteristic data is collected, including user ordering behavior, user emotions, user complaint behavior, and user product usage frequency.

[0104] S2. Preprocess and extract features from the collected KEI data.

[0105] Check for missing values ​​in the KEI data and delete the data records containing missing values ​​directly.

[0106] Outliers are identified using the Isolation Forest algorithm. Isolation Forest constructs multiple isolated trees by randomly selecting features and split points. For the video live streaming loading latency data in this embodiment, the normal data range is (0, 2123) ms, the normal data range for Playback_card_play_lag is (0, 1171) ms, and the normal data range for Playback_cloud_play_lag is (0, 835) ms. System error-type outliers with time-clustering characteristics are deleted, and outliers caused by business rule issues are numerically corrected.

[0107] Duplicate data from the same time, the same user, and the same metric are defined as duplicate values, and only one data record is retained.

[0108] The KEI data after removing outliers and duplicates was normalized using the Z-score normalization method to ensure that all features have the same numerical range.

[0109] A random forest model is used to calculate the correlation between each Key Evaluation Index (KEI) and user satisfaction. KEI indicators with a correlation higher than a first threshold are extracted as single key KEI indicators. In this embodiment, WiFi signal strength and video live streaming loading latency are identified as the two indicators with the highest correlation to user satisfaction.

[0110] Correlation analysis was performed on individual key KEI indicators, and multiple indicators with correlation coefficients higher than the second threshold were grouped together. The results of the correlation analysis are shown in Table 1.

[0111] Based on the analysis results, KEI-1 and KEI-2 are strongly correlated with user satisfaction, and therefore KEI-1 and KEI-2 are identified as important KEI features.

[0112] After classifying satisfaction ratings, cluster analysis was performed on the combinations of indicators within each level to identify unpleasant experience combinations that clustered in the low-satisfaction user group. Experiments revealed that the combination of high video access latency and severe video buffering accounted for 83% of all low-satisfaction users, making this combination an important basis for satisfaction prediction.

[0113] S3. Construct and train a satisfaction prediction model based on a neural network algorithm.

[0114] A 6-layer fully connected neural network was used for feature extraction and satisfaction prediction. Processed KEI data, user feature data, and product satisfaction evaluation data were input into the model for training.

[0115] During training, an optimization algorithm based on the principle of interstellar radiation mapping is used to update the parameters of the neural network. The specific process is as follows:

[0116] Initialize the weights and biases of the neural network, and set the initial diffusion vector.

[0117] In each iteration, according to the formula Update the weights, and update the biases in the same way;

[0118] Dynamic diffusion factor according to calculate;

[0119] The updated parameters are passed through the mapping function. Map back to parameter space;

[0120] Calculate the loss function value; the loss function uses... The nonlinear error function when At that time ;

[0121] when At that time ;

[0122] Perform a cross operation on multiple selected sets of excellent parameters;

[0123] Perform mutation operation on the crossover parameter candidate set The mutation intensity is dynamically adjusted with each iteration, and the iteration is repeated until the maximum number of iterations of 1000 is reached, and the optimized neural network parameters are output.

[0124] S4. Use the trained satisfaction prediction model for real-time prediction.

[0125] The system receives product and network experience metrics in real time during the perception period, combines them with user characteristic data, inputs them into a trained model, and obtains real-time user satisfaction predictions.

[0126] S5, Dynamic Early Warning and Feedback Optimization.

[0127] Set user characteristic rating ranges and product rating ranges. In this embodiment, the normal rating range for home security product satisfaction is [3.5, 5.0].

[0128] When a predicted value deviates from the user characteristic rating range or product rating range, a system alert is triggered. Simultaneously, the prediction period for that user is shortened, and once the monitoring indicator stabilizes, the user configuration is restored to normal.

[0129] We collect customer satisfaction feedback through a real-time post-use evaluation mechanism and use it to develop a satisfaction prediction model.

[0130] After processing KEI data in real time, the system analyzes and optimizes the results in conjunction with key parameters from the product's backend. The optimized parameters are then fed back to the user's terminal device, where the terminal software automatically adjusts the relevant parameters according to preset rules. For example, when the system predicts a potential decline in user satisfaction, it automatically optimizes WiFi signal strength or reduces video loading latency.

[0131] This solution, through real-time collection and analysis of KEI metrics, can accurately identify key indicators and combined indicators affecting user satisfaction, ensuring the real-time nature and accuracy of user experience data. The handling of outlier and duplicate data improves data quality, and Z-score standardization allows indicators of different dimensions to be calculated within a unified numerical range, thereby improving the model's predictive accuracy.

[0132] The neural network parameter optimization algorithm, based on the principle of interstellar radiation mapping, effectively avoids the vanishing gradient and local optima problems that occur during model training, ensuring that the neural network can find the optimal solution under different conditions. This method ensures the robustness and stability of the model by dynamically adjusting the diffusion intensity and mutation operation, thereby improving the accuracy of real-time user satisfaction prediction and effectively handling complex network and user interaction environments.

[0133] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting real-time user satisfaction based on product KEI indicators, characterized in that: Includes the following steps: S1. Collect key experience indicators (KEI) data in real time during the product-user interaction process, collect product satisfaction evaluation data, and collect user characteristic data. S2. Preprocessing and feature extraction are performed on the KEI data. The preprocessing includes missing value deletion, outlier detection and correction, duplicate value deletion, and normalization. The feature extraction includes using a random forest model to extract single key KEI indicators, and performing correlation analysis and cluster analysis on the extracted key KEI indicators to identify the combined features of negative experiences and form a KEI combined feature set. S3. Using the preprocessed and feature-extracted KEI data, the product satisfaction evaluation data, and the user feature data, train the satisfaction prediction model constructed based on the neural network algorithm; during the training process, use an optimization algorithm based on the principle of interstellar ray mapping to update the parameters of the neural network. The optimization algorithm performs a global search in the parameter space by simulating the nonlinear random propagation process of cosmic rays to find the optimal parameters. S4. Using the trained satisfaction prediction model, output the real-time user satisfaction prediction value based on the real-time collected KEI data and user characteristic data. S5. Based on the predicted real-time user satisfaction value, perform dynamic early warning and feedback optimization, collect subjective satisfaction feedback from users to iteratively optimize the satisfaction prediction model, and convert the prediction results into optimization instructions for product operating parameters and feed them back to the user terminal device to achieve real-time improvement of user experience.

2. The method for predicting real-time user satisfaction based on product KEI index according to claim 1, characterized in that: In S1, the KEI data includes at least one of the following: user app opening latency, live video loading latency, cloud playback loading latency, WiFi wall penetration performance, WiFi access waiting time, WiFi disconnection rate, WiFi signal strength, and error rate. The product satisfaction evaluation data includes user experience scores and subjective feedback in dimensions such as overall perception, cost, cost-effectiveness, appearance, hardware, and stability of use; the user characteristic data includes user ordering behavior, user emotions, user complaint behavior, and user product usage frequency.

3. The method for predicting real-time user satisfaction based on product KEI index according to claim 1, characterized in that: Specifically, S2 includes: S21. Check the missing values ​​in the KEI data and directly delete the data records containing missing values; S22. Use the isolated forest algorithm to identify outliers in the KEI data, perform deletion operations on system error outliers with time clustering characteristics, and perform numerical correction operations on outliers caused by business rule issues. S23. Define duplicate data from the same time, the same user, and the same metric as duplicate values, and retain only one data record. S24. The KEI data after deleting outliers and duplicates is normalized using the Z-score normalization method. S25. Use the random forest model to calculate the correlation between each KEI indicator and user satisfaction, and extract the KEI indicators with a correlation higher than the first threshold as single key KEI indicators. S26. Perform correlation analysis on the single key KEI indicators, combine multiple indicators with correlation coefficients higher than the second threshold, and perform cluster analysis on the indicator combinations within each level after classifying the satisfaction scores to identify the negative experience combinations with clustering characteristics in the low satisfaction user group, thus forming the KEI combination feature set.

4. The method for predicting real-time user satisfaction based on product KEI index according to claim 1, characterized in that: In step S3, the optimization algorithm based on the principle of interstellar radiation mapping updates the parameters of the neural network, specifically including: S31. Initialize the weights of the neural network. and bias And set the initial diffusion vector. ; S32. In each iteration, multiple diffusion points are randomly generated based on the current state of the network parameters. Each diffusion point corresponds to a set of potential neural network parameters, and the parameters are updated as follows: in, and The first Weights and biases for the next iteration The dynamic diffusion factor, The time decay factor, This represents the current iteration number; S33. Map the updated parameter set back to the parameter space using a mapping function, wherein the mapping function includes a nonlinear transformation; S34. Calculate the loss function value of each mapped parameter set on the training set, and select the parameter set with the optimal loss function value. S35. Perform a cross operation on the selected sets of multiple excellent parameters to generate a new set of parameter candidates; S36. Perform mutation operation on the crossover candidate set of parameters; S37. Repeat steps S32 to S36 until the preset stopping iteration condition is met, and output the optimized neural network parameters.

5. The method for predicting real-time user satisfaction based on product KEI index according to claim 4, characterized in that: In step S32, the dynamic diffusion factor Calculate according to the following formula: in, The initial diffusion rate, For attenuation control parameters, The nonlinear intensity of the decay.

6. The method for predicting real-time user satisfaction based on product KEI index according to claim 4, characterized in that: In step S33, the mapping function Represented as: in, As the attenuation factor, The first nonlinear mapping parameter, For the second nonlinear mapping parameter, It is the hyperbolic tangent function.

7. The method for predicting real-time user satisfaction based on product KEI index according to claim 4, characterized in that: In step S34, the loss function Calculate according to the following formula: in, For neural network models, For input data, For tags, The number of training samples. It is the L2 norm. The gradient represents the change in the weight parameters of the neural network. Here, ρ is the regularization coefficient, and ρ(⋅) is the nonlinear error function; The nonlinear error function ρ(⋅) is defined as: when hour, ; when hour, ; in, = .

8. The method for predicting real-time user satisfaction based on product KEI index according to claim 4, characterized in that: In S35, the crossover operation is performed according to the following formula: in, It is a cross factor; and These are two weight parameters in the weight set of the mapped neural network; and These are two bias parameters in the weight set of the mapped neural network; This is the set of weight parameters obtained from the crossover operation; This is the set of bias parameters obtained from the crossover operation; In step S36, the mutation operation is performed according to the following formula: in, It is the intensity of variation; This is the set of weight parameters obtained from the mutation operation; This is the set of bias parameters obtained from the mutation operation; The intensity of variation Adjust dynamically according to the following formula: in, It is the initial mutation strength; This represents the total number of iterations.

9. The method for predicting real-time user satisfaction based on product KEI index according to claim 4, characterized in that: In S5, the dynamic early warning and feedback optimization includes: setting user characteristic scoring range and product scoring range; when the real-time user satisfaction prediction value deviates from the user characteristic scoring range or the product scoring range, the system will trigger an early warning and shorten the user's prediction indicator collection cycle; and the system will return to normal after the monitoring indicators stabilize. The product satisfaction evaluation data is statistically processed to calculate the normal scoring range for product satisfaction. The calculated normal scoring range is compared and verified with the actual situation, and the normal scoring range is adjusted according to business development and market changes.

10. The method for predicting real-time user satisfaction based on product KEI index according to claim 1, characterized in that: The user feature data collected in S1 is preprocessed before being input into the model. This preprocessing includes unified encoding and numerical conversion of the user feature data. Using user satisfaction scores as the target variable, the preprocessed user feature data is input into a random forest association analysis model to extract the user feature indicators that have the greatest impact on user satisfaction scores. After verifying the extracted user feature indicators, they are input into the satisfaction prediction model as user feature data. A real-time survey mechanism for post-use evaluation is used to collect users' subjective satisfaction feedback. This subjective satisfaction feedback is used as verification data to evaluate and iteratively optimize the prediction accuracy of the satisfaction prediction model. The prediction results output by the satisfaction prediction model are analyzed in conjunction with key parameters from the product's backend to generate optimized parameters. These optimized parameters are then fed back to the user's terminal device through the backend, enabling the terminal device to automatically adjust its operating parameters according to preset rules. The product is a security product or a networking product.

Images