Social group service quality evaluation method and system

Abstract

The invention discloses a social group service quality assessment method and system, and the method comprises the steps: collecting the service activity data and to-be-assessed data of a social group, and enabling the service activity data to comprise service process data, resource investment data, effect feedback data and coverage population data; obtaining a growth stage label and a service elasticity coefficient through a time variation linear mixed effect model based on a growth factor fluctuation rate and a feedback continuous index of the service activity data, carrying out matching unbalance degree analysis on resource supply and service demands of a social group, obtaining a growth entropy according to the matching unbalance degree and the growth stage label, and sending the growth entropy to the service activity data; and obtaining a service quality evaluation model according to the growth entropy and the three-node causal ring, and inputting the to-be-evaluated data into the service quality evaluation model to obtain a service quality evaluation result. According to the method, the service quality evaluation model is constructed through the growth factor fluctuation rate and the feedback continuous index of the social group, so that the evaluation precision and the evaluation efficiency of the service quality of the social group can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of group service quality evaluation, and in particular to a social group service quality evaluation method and system. Background Art

[0002] In today's society, social organizations play an increasingly important role in providing a wide range of services. Whether providing community services, educational support, or healthcare, the quality of these services is directly linked to the improvement of social welfare and public satisfaction. Accurately assessing the service quality of social organizations not only helps these organizations optimize their service processes and improve resource efficiency, but also provides a valuable reference for social regulation and public choice.

[0003] When evaluating the service quality of social groups, the existing technology usually has problems such as a single evaluation dimension, a lack of a dynamic monitoring mechanism, and difficulty in accurately identifying key influencing factors of service quality. At the same time, due to the inconsistent development status of social groups, the same evaluation system cannot be effectively applied. To overcome these shortcomings, the present invention proposes a comprehensive social group service quality evaluation method. The present invention collects multi-dimensional data, including information on service process, resource input, effect feedback, and covered population, and introduces the growth factor volatility and feedback continuity index of social groups. It can not only capture the dynamic changes of service activities, but also effectively locate the external factors that have a key impact on service quality, providing support for social groups to improve service quality, optimize resource allocation, and achieve sustainable development. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for evaluating the service quality of a social group.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions: A first aspect of the present invention provides a method for evaluating the service quality of a social group, comprising: Collecting service activity data and data to be evaluated from social groups, including service process data, resource input data, effect feedback data, and data on the population covered; Based on the time series and volatility of the service activity data, the data is divided to obtain the growth factor volatility and the feedback continuity index; Differentially identifying social groups based on the growth factor volatility and feedback continuity index to obtain a dynamic feature set, which is then input into a time-varying linear mixed effects model to obtain growth stage labels and service elasticity coefficients; Based on the growth factor volatility, the service elasticity coefficient matches the resource supply and service demand of the social group, analyzes the matching imbalance degree, obtains the growth entropy according to the matching imbalance degree and the growth stage label, constructs a three-node causal loop according to the growth entropy and the entropy ring rate change, and dynamically updates the causal loop using the elasticity coefficient feedback loop; According to the three-node causal loop and the growth entropy, a service quality evaluation model is obtained, the service quality evaluation model is input into the service quality evaluation model, and a service quality evaluation result is obtained.

[0006] As a further method, the method for obtaining the growth factor volatility comprises: Based on the service activity data, the time sequence is divided through a rolling time window, the social group monthly report period is taken as the window length, the window overlap rate is set to 50%, and the average processing effect before and after the resource input is calculated through double difference according to the service coverage mean in the time window and the resource input data; Based on the resource input data, the subject that inputs resources to the social group is extracted, the survival analysis of the social group is performed according to the type, type proportion, quantity and input duration of the subject, and the resource input persistence of each type of subject is obtained; According to the time window, the trend value of the service activity data is calculated through a three-order polynomial fitting coefficient, the trend value and the residual error of the service activity data are taken as the short-term fluctuation signal, and the Z-score standardization is performed on the time sequence based on the mean and standard deviation of the residual error; Based on the average processing effect, the resource input persistence and the short-term fluctuation signal, the growth factor volatility of the current time window is calculated through a two-dimensional weighted volatility formula, and the two-dimensional weighted volatility formula is: , Wherein is the growth factor energy of the tth time window, is the growth factor volatility of the tth time window, t is the time window index, and is the average processing effect of the tth time window, i is the time window index, is the fluctuation weight of the tth time window, is obtained through grid search optimization according to historical data and subject type proportion in the past year, is the growth factor energy of the t-1th time window, is the resource input persistence of the subject, and s is the type of the subject that inputs resources to the social group, is the short-term fluctuation signal, is the mean of the standardized data of the t-1th time window.

[0007] As a further method, the method for obtaining the feedback continuity index comprises: The effect feedback data and the covered population data are maximum minimum normalized based on a time window of a time sequence, the normalized participation duration proportion mean, the overtime rate and the service penetration rate are extracted according to the time window, and are spliced into a feedback index vector, the local outlier factor score is calculated according to the cosine similarity of the feedback index vectors in adjacent time windows, the time window with the local outlier factor score greater than 3 is determined as an abnormal fluctuation, the cosine similarity of the abnormal window is multiplied by a penalty coefficient 0.8, the smoothed time sequence is obtained according to the penalized time window, and the feedback continuity index is calculated according to the feedback index vector and the moving step of 3 time windows through the weighted moving average.

[0008] As a further method, the method for obtaining the dynamic feature set comprises: The growth factor fluctuation rate and the resource allocation concentration degree in a half-year period are extracted based on the time sequence, the growth factor fluctuation rate difference of adjacent time windows is calculated through the exponential weighted moving average, if the resource allocation concentration degree in the time window is greater than 0.7, the growth factor fluctuation rate difference is multiplied by a penalty factor 1.2, if the resource allocation concentration degree is not greater than 0.7, no penalty is given, and the fluctuation rate difference is obtained. The median of the feedback continuity index in the past year is obtained based on the time sequence, the difference between the feedback continuity index in the last half year and the median is calculated, the square root of the absolute value of the difference is taken as the fluctuation amplitude, the trend direction is determined through the sign function according to the sign of the feedback continuity index difference of adjacent time windows in the time sequence, the trend direction is multiplied by the fluctuation amplitude, and the feedback continuity difference is obtained. If the fluctuation rate difference of the same time window is greater than 1.5 and the feedback continuity difference is less than zero, the time window is marked as resource mismatch; if the fluctuation rate difference falls within the interval of 0.5 to 1.5 and the feedback continuity difference is greater than zero, the time window is marked as elastic resource supplement; if the fluctuation rate difference is less than 0.5 and the feedback continuity difference is approximately equal to zero, the time window is marked as steady state operation. The growth factor fluctuation rate, the feedback continuity index, the resource concentration degree and the service penetration rate are spliced into a feature vector according to the time window, the dynamic feature matrix is obtained based on the time sequence and the feature vector, the dynamic feature matrix is bound with the corresponding difference identifier, and the dynamic feature set is obtained.

[0009] As a further method, the method for obtaining the growth stage label and the service elasticity coefficient comprises: The participation duration proportion, the overtime rate, the growth factor fluctuation rate sequence, the feedback continuity index sequence, the difference identifier, the type of social group, the service coverage regional expansion rate and the establishment time limit in the corresponding time sequence are extracted based on the dynamic feature set, and the service quality comprehensive index is calculated according to the participation duration proportion and the overtime rate through the variance weight method. The growth factor volatility, feedback continuous index and difference identification are taken as fixed effect parameters, and the type of social group, service coverage expansion rate and establishment time are taken as random effect parameters of social group. A time-varying linear mixed effect model is constructed according to the service quality comprehensive index, fixed effect parameters and random effect parameters, and the expression is: , wherein is the service quality comprehensive index, is the zero effect benchmark service quality, is the coefficient of B-spline base function, which is estimated by least square method, is the coefficient of B-spline base function, is the coefficient of B-spline base function, is the B-spline base function, and k is the order of base function, is the growth factor volatility, is the corresponding elastic resource supplement, is the random effect of social group z, is the random volatility, and t is the time index; Based on the time-varying linear mixed effect model, the fixed effect parameters and random effect parameters are fitted by Markov chain Monte Carlo algorithm to obtain the time-varying coefficient and random effect. According to the difference identification, the first derivative of time-varying coefficient and the second derivative of time-varying coefficient, the growth phase label is divided according to the time-varying trend. The service elasticity coefficient is calculated according to the time-varying coefficient, and the calculation formula of the service elasticity coefficient is: , wherein is the service elasticity coefficient of current time t, is any historical time point between 0 and t, is the ratio of time resource input change rate to service coverage number change rate, is the influence range of historical elasticity, which is calibrated by the variance contribution of the second derivative in the historical elasticity sliding window, is the time-varying coefficient of time, is the proportion of resource input of a certain type of subject in time t, is the complaint decay rate, is the first derivative of service quality comprehensive index at time t, is the ratio of complaint processing lag time to time window, is the service quality change rate.

[0010] As a further method, the matching imbalance degree analysis method comprises: The product of the service elasticity coefficient and the resource allocation concentration is calculated based on the time sequence, the growth factor volatility is divided by the product, the logarithmic absolute value of the calculation result is taken as the resource supply deviation, the average processing effect is subtracted from the service penetration rate, the absolute value is taken, and the demand coverage gap is obtained. The product of the preset industry weight and the resource supply deviation is added to the product of the preset industry weight and the demand coverage gap to obtain the matching imbalance degree.

[0011] As a further method, the method for obtaining the three-node causal loop comprises: Based on the time sequence, the growth entropy is calculated according to the matching imbalance degree and the growth stage label, and the growth entropy formula is: Among them is the contribution probability of the first growth stage, is the deviation adjustment factor, is the historical average imbalance degree, n is the number of growth stage labels, is the matching imbalance degree of the first growth stage, is the service elasticity coefficient of the social group z, is the feedback continuity index under the first growth stage, is the growth entropy of the social group z; According to the time sequence, the resource supply deviation and the growth factor volatility are taken as the resource supply node, the service elasticity coefficient and the service quality comprehensive index are taken as the service response node, the feedback continuity index and the demand coverage gap are taken as the feedback adjustment node, and the three nodes are connected to establish a three-node causal loop; The average value of the service elasticity coefficient in the past year is taken as the resource service edge weight, the average value of the feedback continuity index is taken as the service feedback edge weight, and the ratio of the current matching imbalance degree to the average value of the matching imbalance degree is taken as the feedback resource edge weight. If the absolute value of the ring-to-ring change rate of the growth entropy in the adjacent period exceeds 10%, the three-node causal loop is updated, the first derivative sign of the time variation coefficient between nodes is determined to determine the causal loop update direction, and the second derivative absolute value of the time variation coefficient is determined to determine the update strength.

[0012] As a further method, the method for obtaining the service quality evaluation result comprises: Based on the three-node causal loop, the nodes are mapped to three feature layers of a causal graph neural network, the three-node causal loop edge weight is taken as the causal graph edge weight, the time sequence, the matching imbalance degree and the growth entropy are taken as input data according to the service activity data, the service quality comprehensive index is taken as the training label, the input data is divided into a training set, a validation set and a test set, and the training set and the training label are input into the causal graph neural network.​​ The growth entropy is taken as a regular term of a loss function, the overfitting state of the neural network is adjusted according to the change of the loss function value and the accuracy, the difference between the neural network prediction value and the training label is fed back to each layer of the neural network through back propagation, if the accuracy of the verification set is improved by less than 0.1% for 5 consecutive rounds, the training is ended, the test set is input into the trained causal diagram neural network, the model performance is optimized according to the mean square error and the mean absolute error output by the test set, and a service quality evaluation model is obtained, the time sequence, the matching imbalance degree and the growth entropy of the data to be evaluated are input into the service quality evaluation model, and a service quality evaluation index is obtained.

[0013] The second aspect of the present application provides a social group service quality evaluation system, comprising: A service data acquisition module is configured to acquire service activity data of a social group, wherein the service activity data comprises service process data, resource input data, effect feedback data and covered population data. A service data feature extraction module is configured to divide the service activity data based on time sequence and volatility to obtain growth factor volatility and feedback continuity index. A social group identification module is configured to identify differences in social groups based on the growth factor volatility and the feedback continuity index to obtain a dynamic feature set, input the dynamic feature set into a time variation linear mixed effect model to obtain a growth stage label and a service elasticity coefficient. A social group causal loop analysis module is configured to analyze the matching imbalance degree of resource supply and service demand of a social group based on the growth factor volatility and the service elasticity coefficient, obtain growth entropy based on the matching imbalance degree and the growth stage label, construct a three-node causal loop based on the growth entropy and the entropy ring rate of change, and dynamically update the causal loop using the elasticity coefficient feedback loop. A social group service quality evaluation module is configured to obtain a service quality evaluation model based on the three-node causal loop and the growth entropy, input the service activity data into the service quality evaluation model, and obtain a service quality evaluation result.

[0014] Compared with the prior art, the embodiments of the present application have at least the following advantages or beneficial effects: The application can accurately capture the dynamic changes of service activities by a time-varying linear mixed effect model, identify the development stage of the social group by using the growth stage label and the service elasticity coefficient, quantify the response ability of service activities to external changes, obtain the matching imbalance degree of external resource input and service demand input, and provide certain guidance for the reasonable allocation of resources, construct a service quality evaluation model through a three-node causal loop, realize the quantitative evaluation of the periodical changes of service quality, and provide strong support for the continuous improvement and optimization of the social group, and effectively improve the evaluation accuracy and evaluation efficiency of the service quality of the social group. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 A step flowchart of a social group service quality evaluation method in the embodiments of the application. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the protection scope of the application.

[0017] Referring to Figure 1 The application provides a social group service quality evaluation method, which comprises: Collecting service activity data and to-be-evaluated data of the social group, wherein the service activity data comprises service process data, resource input data, effect feedback data and covered population data; In actual evaluation, a social group is selected for data collection according to a certain city social organization information center, a certain community public service organization is taken as an evaluation object, a public annual report of the social group in 2024 is taken as a service activity data collection interval, and service activity data newly added from January to March in 2025 is taken as to-be-evaluated data, service activity data of January to December in the public annual report of the social group in 2024 is extracted, the service activity data comprises service process data, resource input data, effect feedback data and covered population data, the service process data comprises response time and node completion rate, the resource input data comprises input subject, input event, duration, fund input fluctuation rate, resource allocation concentration degree and establishment time, the effect feedback data comprises participation duration proportion, overtime rate and complaint type distribution, and the covered population data comprises annual service covered regional expansion rate, beneficiary population demand distribution and service penetration rate. Specifically, service process data: monthly response time, 48 hours in January, 42 hours in February, node completion rate, 85% in January, 88% in February; resource input data: monthly input subject, 60% of enterprises and 40% of individuals in January, input event is the Spring Festival care material procurement in January, duration is 30 days, fund input fluctuation rate is 8% in January, resource allocation concentration is 0.65 in January; effect feedback data: participation duration in February accounts for 65%, overtime rate is 10%, complaint type distribution in February accounts for 60% of service efficiency; covered population data: annual service coverage regional expansion rate is 8%, demand distribution of benefited population accounts for 60% of the elderly in April, service penetration rate is 35%; Align all service activity data with the monthly time axis. For non-equidistant records in the data, the service penetration rate is missing in July. Based on the interpolation of 35% in June and 42% in August, the value in July is 38.5%, which is close to the actual value of 38% in July in the historical data. Based on the time sequence and fluctuation rate of the service activity data, the growth factor fluctuation rate and the feedback continuity index are obtained. In actual evaluation, the monthly report period is taken as the window length, the window overlap rate is 50%, that is, the adjacent windows overlap for 15 days, and the 12-month data is divided into 23 time windows, wherein window 1 is from January 1 to January 31, window 2 is from January 16 to February 15. In window 1, the service coverage average is calculated according to the number of days in the time window and the corresponding covered population data. The Spring Festival care material procurement lasts for 30 days, the service coverage average is 580 people in the first 15 days, and the service coverage average is 620 people in the last 15 days. The average treatment effect is 40 per person calculated by the double difference method. The input subject of window 1 is subjected to survival analysis, wherein enterprises account for 60%, last for 30 days, individuals account for 40%, and last for 30 days. The historical average duration of enterprise subjects is 25 days, and the historical average duration of individuals is 20 days, , The weighted average resource input persistence is 1.32. The service activity data of window 1 is fitted with a third-order polynomial trend value, and the residual is the difference between the actual value and the trend value. After Z-score standardization, the short-term fluctuation signal is obtained, wherein the short-term fluctuation signal of the service coverage population is 0.8. According to the average treatment effect, the resource input persistence and the short-term fluctuation signal, the growth factor fluctuation rate of window 1 is calculated by the double-dimensional weighted fluctuation rate formula, which is 0.716. The participation duration proportion mean, overtime rate, and penetration rate in the effect feedback data of the time window in the time sequence are maximum minimum normalized to obtain a time window vector, window 1 [0.84, 0.21, 0.35], window 2 [0.86, 0.18, 0.37], and window 3 [0.90, 0, 0.38]. The cosine similarity of adjacent window vectors is calculated, window 1 to 2 is 0.98, and window 2 to 3 is 0.99. Since the local outlier factor is less than 3, no punishment is needed, and the smoothed sequence maintains the original similarity. The normalized data is weighted and averaged with a moving step of 3 time windows, and the weight is obtained by a linear weighting formula wherein is the weight of the tthtime window, h is the total number of time windows, c is the time window index, the weight is [0.5, 0.333, 0.166], s is the participation duration proportion multiplied by the penetration rate, then multiplied by 1-complaint rate, and then multiplied by the weight to obtain the feedback continuity index is 0.352, is 0.370; According to the growth factor volatility and the feedback continuity index, the social group is identified for difference, a dynamic feature set is obtained, the dynamic feature set is input into a time variation linear mixed effect model to obtain a growth stage label and a service elasticity coefficient. In actual evaluation, the growth factor volatility and resource allocation concentration in the half-year period are extracted based on the time sequence, and the exponential weighted moving average is used, wherein the weight is [0.1, 0.2, 0.3, 0.4] to calculate the growth factor volatility difference of adjacent windows, wherein window 1 to window is 0.682-0.716=0.034, the resource concentration is 0.72, which is greater than 0.7, multiplied by the penalty factor 1.2 to obtain the volatility difference of 0.0408; The median of the feedback continuity index in the past year is 0.89, the absolute value square root of the difference between the median and the median in the last half year and the trend direction is, the fluctuation amplitude of window 2 is 0.892-0.89=0.0447, the trend direction of window 2 to window 1 is 0.892-0.878>0, which is positive, and the feedback continuity difference is 1x0.0447=0.0447. Since the volatility difference of window 2 is less than 0.5, the feedback continuity difference month is equal to 0, which is marked as steady state operation. According to the time window, the growth factor volatility, the feedback continuity index, the resource concentration, and the service penetration rate are spliced into a feature vector [0.682, 0.892, 0.72, 0.38]. Based on the time sequence and the feature vector, a dynamic feature matrix is obtained. The dynamic feature matrix is bound with the corresponding difference identification "steady state operation" to obtain a dynamic feature set. According to the participation duration proportion and the overtime rate, the service quality comprehensive index is calculated by the variance weight method, wherein the participation duration proportion in window 2 is 65%, the overtime rate is 10%, and the service quality comprehensive index is 0.85. The Markov chain Monte Carlo algorithm is used to fit the fixed effect parameters and random effect parameters based on the time-varying linear mixed effect model, where the random effect is ,in is the type of social group, ER is the service coverage area expansion rate, Age is the establishment year, is the random effect coefficient, subject to , is the covariance matrix, estimated by the MCMC algorithm, is the random effect coefficient, is the random effect coefficient, is the random effect coefficient, where the estimation algorithm is ,in are the coefficients of the spline basis function, is the time-varying trend of the j-th type effect, is the smoothing parameter, , represents the initial value of the smoothing parameter, is the historical average of the comprehensive service quality index; The time coefficient of variation for time window 2 is [0.3, 0.5, 0]. The community service category in the random effect is 0.2, the service coverage expansion rate is 5% to 0.05, and the establishment period is 5 years to 0.5. Based on the difference label "steady-state operation", the first-order derivative of the time coefficient of variation is 0.02 greater than 0, and the second-order derivative of the time coefficient of variation is 0.005≈0, the growth stage is divided into "stable growth period" according to the time-varying trend. Since the instantaneous resource elasticity in February is 2, the historical elasticity trend weight is 0.05, the service quality change rate from window 1 to window 2 is 0.05, and the complaint attenuation rate is 0.1. The service elasticity coefficient of time window 2 is 1.456 calculated by the time coefficient of variation. Among them, the difference mark "steady-state operation", the first-order derivative is greater than 0, the second-order derivative is less than or equal to 0.01, the growth stage label is "stable growth period", the difference mark "steady-state operation", the first-order derivative ≈ 0, the second-order derivative ≈ 0, the growth stage label is "maturity period", the difference mark "elastic resource replenishment", the first-order derivative is greater than 0, the second-order derivative is less than or equal to 0.01, the growth stage label is "stable growth period", the difference mark "elastic resource replenishment", the first-order derivative is greater than 0, the second-order derivative is greater than or equal to 0.01, the growth stage label is "high-speed growth period", the difference mark "resource mismatch degradation", the first-order derivative is less than 0, the second-order derivative is less than or equal to 0.01, the growth stage label is "bottleneck period", the difference mark "elastic resource replenishment", the first-order derivative is less than 0, the second-order derivative is greater than or equal to 0.01, the growth stage label is "rapid decline period"; Identify the matching imbalance degree of the resource supply and service demand of the social group by the growth factor volatility, service elasticity coefficient, and obtain the growth entropy according to the matching imbalance degree and the growth stage label, and construct a three-node causal loop according to the growth entropy and the entropy ring rate change, and dynamically update the causal loop using the elasticity coefficient feedback loop; In the actual evaluation, the matching imbalance degree of the resource supply and service demand of window 2 is 2.485 based on the resource supply deviation and the demand coverage gap, the growth entropy is 0.88 according to the matching imbalance degree and the growth stage labels "stable growth period" and "bottleneck period", and the three-node causal loop is constructed according to the growth entropy and the entropy ring rate change, wherein the resource service edge weight is 0.23, the service feedback edge weight is 0.908, and the feedback resource edge weight is 1.092, wherein the growth entropy ring rate change of adjacent windows window 2 and window 3 is 0.12, which exceeds 10%, triggering updating, and since the first order derivative of the time variation coefficient is positive, it is a growth trend, and the updating direction is to strengthen resource supply and improve service response, and the absolute value of the second order derivative is 0.02, so the updating intensity is 0.02. According to the three-node causal loop and the growth entropy, a service quality evaluation model is obtained, and the service quality evaluation result is obtained by inputting the to-be-evaluated data into the service quality evaluation model.

[0018] In the actual evaluation, the three-node causal loop is mapped to three feature layers of a causal graph neural network, and the edge weights of the three-node causal loop are used as the edge weights of the causal graph. According to the service activity data, the time sequence, the matching imbalance degree and the growth entropy are used as input data, the service quality comprehensive index is used as a training label, the input data is divided into a training set 70%, a verification set 15% and a test set 15%, and the training set and the training label are input into the causal graph neural network. The growth entropy is used as a regular term of a loss function, the loss function value and the accuracy change are used to adjust the overfitting state of the neural network, the difference between the neural network prediction value and the training label is fed back to each layer of the neural network through back propagation, the weights are adjusted, and the training is ended after the verification set accuracy is improved by less than 0.1% in the 8th round. The test set is input into the trained causal graph neural network, the mean square error 0.021 and the mean absolute error 0.12 output by the test set are used to determine the model performance, and the service quality evaluation model is obtained. The time sequence, the matching imbalance degree and the growth entropy of the to-be-evaluated data from January to March in 2025 are input into the service quality evaluation model, and the service quality evaluation indexes 0.75, 0.8 and 0.85 are obtained.

[0019] In this embodiment, the method for obtaining the growth factor volatility comprises: Based on service activity data, time series is divided by rolling time window, social group monthly report cycle is taken as window length, window overlap rate is set to 50%, according to service coverage mean in time window and resource input data, average treatment effect before and after resource input is calculated by double difference; Based on resource input data, the subject of resource input to social groups is extracted, and survival analysis is performed on social groups according to the type, type proportion, number and input duration of the subject, to obtain the resource input persistence of each type of subject; According to the time window, the trend value of service activity data is calculated by three-order polynomial fitting coefficient, the trend value and the residual of service activity data are taken as short-term fluctuation signal, and Z-score standardization is performed on time series based on the mean and standard deviation of residual; Based on average treatment effect, resource input persistence and short-term fluctuation signal, growth factor volatility of current time window is calculated by two-dimensional weighted volatility formula, two-dimensional weighted volatility formula is: , Wherein is the growth factor energy of the tth time window, is the growth factor volatility of the tth time window, t is the time window index, is the average treatment effect of the tth time window, i is the time window index, is the fluctuation weight of the tth time window, which is obtained by grid search optimization according to historical data and subject type proportion in the past year, is the growth factor energy of the t-1th time window, is the resource input persistence of the subject, s is the type of subject that produces resource input to social groups, is the short-term fluctuation signal, is the mean of standardized data of the t-1th time window.

[0020] In this embodiment, the method for obtaining the feedback continuity index comprises: Based on the time window of time series, the effect feedback data and coverage population data are normalized by maximum and minimum, the normalized participation duration proportion mean, overtime rate and service penetration rate are extracted according to the time window, and are spliced into feedback index vector, the local outlier factor score is calculated according to the cosine similarity of feedback index vector in adjacent time windows, the time window with local outlier factor score greater than 3 is determined as abnormal fluctuation, the cosine similarity of abnormal window is multiplied by penalty coefficient 0.8, the smoothed time series is obtained according to the time window after punishment, the feedback continuity index is calculated by weighted moving average based on feedback index vector according to moving step of 3 time windows.

[0021] In the embodiment, the method for obtaining the dynamic feature set comprises: Based on the time sequence, the growth factor fluctuation rate and the resource allocation concentration in the half-year period are extracted, and the growth factor fluctuation rate difference of adjacent time windows is calculated by an exponential weighted moving average. If the resource allocation concentration in the time window is greater than 0.7, the growth factor fluctuation rate difference is multiplied by a penalty factor 1.2, and if the resource allocation concentration is not greater than 0.7, there is no penalty, and the fluctuation rate difference is obtained. Based on the time sequence, the median of the feedback continuity index in the past year is obtained, the difference between the feedback continuity index in the recent half year and the median is calculated, the absolute value of the difference is square rooted as the fluctuation amplitude, the trend direction is determined by the sign function according to the sign of the feedback continuity index difference of adjacent time windows in the time sequence, and the trend direction is multiplied by the fluctuation amplitude to obtain the feedback continuity difference. If the fluctuation rate difference of the same time window is greater than 1.5 and the feedback continuity difference is less than zero, the time window is marked as resource mismatch; if the fluctuation rate difference falls within the interval of 0.5 to 1.5 and the feedback continuity difference is greater than zero, the time window is marked as elastic resource supplement; and if the fluctuation rate difference is less than 0.5 and the feedback continuity difference is approximately equal to zero, the time window is marked as steady state operation. According to the time window, the growth factor fluctuation rate, the feedback continuity index, the resource concentration, and the service penetration rate are spliced into a feature vector, a dynamic feature matrix is obtained based on the time sequence and the feature vector, the dynamic feature matrix is bound with the corresponding difference identifier, and a dynamic feature set is obtained.

[0022] In the embodiment, the method for obtaining the growth stage label and the service elasticity coefficient comprises: Based on the dynamic feature set, the participation duration proportion, the timeout rate, the growth factor fluctuation rate sequence, the feedback continuity index sequence, the difference identifier, the type of social group, the service coverage regional expansion rate, and the establishment time in the corresponding time sequence are extracted, and the service quality comprehensive index is calculated by the variance weight method according to the participation duration proportion and the timeout rate. The growth factor fluctuation rate, the feedback continuity index, and the difference identifier are taken as fixed effect parameters, the type of social group, the service coverage regional expansion rate, and the establishment time are taken as social group random effect parameters, a time-varying linear mixed effect model is constructed according to the service quality comprehensive index, the fixed effect parameters, and the random effect parameters, and the expression is: , Wherein is the service quality comprehensive index, is the zero effect benchmark service quality, is the coefficient of the B-spline basis function, which is estimated by the least square method, is the coefficient of the B-spline basis function, are the coefficients of the B-spline basis function, is the B-spline basis function, k is the order of the basis function, is the growth factor volatility, is the feedback continuous index, is the difference identifier, j is a dummy variable, when j=1, it corresponds to resource mismatch, when j=2, it corresponds to elastic resource supplementation, is the random effect of social group z, is random fluctuation, t is the time index; Based on the time-varying linear mixed-effect model, the Markov chain Monte Carlo algorithm is used to fit the fixed effect parameters and random effect parameters to obtain the time variation coefficient and random effect. Based on the difference mark, the first-order derivative of the time variation coefficient, and the second-order derivative of the time variation coefficient, the growth stage labels are divided according to the time-varying trend. The service elasticity coefficient is calculated based on the time variation coefficient. The service elasticity coefficient calculation formula is: , in is the service elasticity coefficient at the current time t, is any historical time point between 0 and t, for The ratio of the change rate of time resource investment to the change rate of the number of people covered by the service, The influence range of historical elasticity is calibrated by the variance contribution of the second-order derivative within the historical elastic sliding window. for The temporal coefficient of variation of time, for The proportion of resource input of a certain type of subject in time, is the complaint attenuation rate, is the first-order derivative of the comprehensive index of service quality at time t, is the ratio of complaint processing delay time to time window, is the service quality change rate.

[0023] In this embodiment, the matching imbalance analysis method includes: Based on the time series, the product of the service elasticity coefficient and the resource allocation concentration is calculated, the growth factor volatility is divided by the product, and the logarithm absolute value of the calculation result is used as the resource supply deviation. The square root of the absolute value of the difference between the average treatment effect and the service penetration rate is taken to obtain the demand coverage gap. The matching imbalance is obtained by adding the product of the preset industry weight and the resource supply deviation plus the product of the preset industry weight and the demand coverage gap.

[0024] In this embodiment, the method for obtaining the three-node causal loop includes: The growth entropy is calculated based on the time series according to the matching imbalance and growth stage label. The growth entropy formula is: , in For the The contribution probability of the class growth stage, is the deviation adjustment factor, , is the historical average imbalance, n is the number of labels in the growth stage, For the The matching imbalance degree of the class growth stage, is the service elasticity coefficient of social group z, For the Feedback continuity index under the growth stage, is the growth entropy of social group z; According to the time series, resource supply deviation and growth factor volatility are used as resource supply nodes, service elasticity coefficient and service quality comprehensive index are used as service response nodes, feedback continuity index and demand coverage gap are used as feedback adjustment nodes, and the three nodes are connected to establish a three-node causal loop; The average service elasticity coefficient in the past year is used as the resource service edge weight, the average feedback continuity index is used as the service feedback edge weight, and the ratio of the current matching imbalance to the average matching imbalance is used as the feedback resource edge weight. If the absolute value of the month-on-month change rate of growth entropy in adjacent time periods exceeds 10%, the three-node causal loop update is triggered. The causal loop update direction is determined according to the sign of the first-order derivative of the time variation coefficient between nodes, and the update intensity is determined according to the absolute value of the second-order derivative of the time variation coefficient.

[0025] In this embodiment, the method for obtaining the service quality evaluation result includes: Based on the three-node causal loop, the nodes are mapped into the three feature layers of the causal graph neural network. The edge weights of the three-node causal loop are used as the edge weights of the causal graph. Based on the service activity data, the time series, matching imbalance, and growth entropy are used as input data. The comprehensive service quality index is used as the training label. The input data is divided into training set, validation set, and test set. The training set and training labels are used to input the causal graph neural network. The growth entropy is used as the regularization term of the loss function, and the overfitting state of the neural network is adjusted according to the changes in the loss function value and accuracy. The difference between the neural network prediction value and the training label is fed back to each layer of the neural network through back propagation. If the accuracy of the validation set is improved by less than 0.1% for five consecutive rounds, the training is terminated. The test set is input into the trained causal graph neural network, and the model performance is optimized according to the mean square error and mean absolute error output by the test set to obtain a service quality evaluation model. The time series sequence, matching imbalance and growth entropy of the data to be evaluated are input into the service quality evaluation model to obtain the service quality evaluation index.

[0026] The second aspect of the present invention further provides a social group service quality evaluation system, comprising: Service data collection module: used to collect service activity data of social groups, including service process data, resource input data, effect feedback data and coverage population data; Service data feature extraction module: used to divide the service activity data based on the time series and volatility, and obtain the growth factor volatility and feedback continuity index; Social group identification module: used to differentially identify social groups based on the growth factor volatility and feedback continuity index, obtain a dynamic feature set, use the dynamic feature set to input the time-varying linear mixed effect model, and obtain growth stage labels and service elasticity coefficients; Social Group Causal Loop Analysis Module: This module is used to analyze the matching imbalance between resource supply and service demand of social groups based on the growth factor volatility and the service elasticity coefficient, obtain growth entropy based on the matching imbalance and the growth stage label, construct a three-node causal loop based on the growth entropy and the entropy ring change rate, and dynamically update the causal loop using the elasticity coefficient feedback loop; Social group service quality evaluation module: used to obtain a service quality evaluation model based on the three-node causal loop and growth entropy, input the service activity data into the service quality evaluation model, and obtain a service quality evaluation result.

[0027] The above content is merely an example and explanation of the structure of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the scope of protection of the present invention.

Claims

1. A method for evaluating the service quality of a social group, characterized in that: The following steps are involved: Collecting service activity data and data to be evaluated from social groups, including service process data, resource input data, effect feedback data, and data on the population covered; Based on the time series and volatility of the service activity data, the data is divided to obtain the growth factor volatility and the feedback continuity index; Differentially identifying social groups based on the growth factor volatility and feedback continuity index to obtain a dynamic feature set, which is then input into a time-varying linear mixed effects model to obtain growth stage labels and service elasticity coefficients; Based on the growth factor volatility and service elasticity coefficient, the matching imbalance degree of the resource supply and service demand of the social group is analyzed, the growth entropy is obtained according to the matching imbalance degree and the growth stage label, a three-node causal loop is constructed according to the growth entropy and the entropy ring change rate, and the causal loop is dynamically updated using the elasticity coefficient feedback loop; A service quality evaluation model is obtained according to the three-node causal loop and the growth entropy, and the data to be evaluated is input into the service quality evaluation model to obtain a service quality evaluation result.

2. A social group service quality assessment method according to claim 1, characterized in that: The method for obtaining the growth factor fluctuation rate comprises: Based on the service activity data, the time series is divided into rolling time windows, the monthly reporting period of social groups is used as the window length, and the window overlap rate is set to 50%. The average treatment effect before and after resource investment is calculated by double difference based on the mean service coverage and resource investment data in the time window; Based on the resource input data, we extract the subjects that have invested resources in social groups, conduct survival analysis on social groups based on the type, proportion, number and duration of the subject, and obtain the sustainability of resource input by each type of subject; The trend value of service activity data is calculated using the third-order polynomial fitting coefficient according to the time window. The residual between the trend value and the service activity data is used as a short-term fluctuation signal. Z-score normalization is performed on the time series based on the mean and standard deviation of the residual. Based on the average treatment effect, resource input persistence and short-term fluctuation signals, the growth factor volatility of the current time window is calculated using the two-dimensional weighted volatility formula. The two-dimensional weighted volatility formula is: ， in is the growth factor energy in the t-th time window, is the growth factor volatility of the t-th time window, t is the time window index, and The average treatment effect of the t-th time window, i is the time window index, is the volatility weight of the t-th time window, which is obtained through grid search optimization based on historical data and the proportion of subject types in the past year. is the growth factor energy in the t-1th time window, is the sustainability of the subject's resource input, s is the type of subject that generates resource input to social groups, It is a short-term fluctuation signal. is the mean of the standardized data in the t-1th time window.

3. A social group service quality assessment method according to claim 2, characterized in that: The method for obtaining the feedback continuity index includes: Based on the time window of the time series, the effect feedback data and the coverage population data are normalized to the maximum and minimum values. According to the time window, the normalized mean participation time proportion, timeout rate, and service penetration rate are extracted and spliced ​​into a feedback index vector. The local outlier factor score is calculated based on the cosine similarity of the feedback index vectors in adjacent time windows. The time window with a local outlier factor score greater than 3 is determined to be an abnormal fluctuation. The cosine similarity of the abnormal window is multiplied by a penalty coefficient of 0.

8. The smoothed time series is obtained according to the time window after the penalty. According to the smoothed time series, with 3 time windows as the moving step, the feedback continuity index is calculated by weighted moving average based on the feedback index vector according to the moving step.

4. A social group service quality assessment method according to claim 1, characterized in that: The method for obtaining the dynamic feature set includes: Based on the time series, the volatility of the growth factor and the concentration of resource allocation within a six-month period are extracted. The difference in the volatility of the growth factor in adjacent time windows is calculated by exponentially weighted moving average. If the concentration of resource allocation in the time window is greater than 0.7, the difference in the volatility of the growth factor is multiplied by a penalty factor of 1.

2. If the concentration of resource allocation is not greater than 0.7, no penalty is applied to obtain the volatility difference. Based on the time series, the median of the feedback continuity index over the past year is obtained. The difference between the feedback continuity index and the median over the past six months is calculated. The square root of the absolute value of the difference is taken as the fluctuation amplitude. The trend direction is determined by the sign function based on the positive and negative signs of the feedback continuity index differences between adjacent time windows in the time series. The trend direction is multiplied by the fluctuation amplitude to obtain the feedback continuity difference. If the volatility difference in the same time window is greater than 1.5 and the feedback continuity difference is less than zero, the time window is marked as resource mismatch; if the volatility difference falls between 0.5 and 1.5 and the feedback continuity difference is greater than zero, the time window is marked as elastic resource replenishment; if the volatility difference is less than 0.5 and the feedback continuity difference is approximately equal to zero, the time window is marked as steady-state operation; According to the time window, the growth factor volatility, feedback continuity index, resource concentration, and service penetration are spliced ​​into feature vectors. Based on the time series and feature vectors, a dynamic feature matrix is ​​obtained. The dynamic feature matrix is ​​bound to the corresponding difference identifier to obtain a dynamic feature set.

5. A social group service quality assessment method according to claim 1, characterized in that: The method for obtaining the growth stage label and the service elasticity coefficient includes: Based on the dynamic feature set, the proportion of participation time, overtime rate, growth factor volatility series, feedback continuous index series, difference identifier, type of social group, service coverage area expansion rate and establishment years in the corresponding time series are extracted. The comprehensive service quality index is calculated based on the proportion of participation time and overtime rate using the variance weighting method. Taking the growth factor volatility, feedback continuity index and difference marker as fixed effect parameters, and the type of social group, service coverage area expansion rate and establishment years as social group random effect parameters, a time-varying linear mixed effect model is constructed based on the service quality comprehensive index, fixed effect parameters and random effect parameters. The expression is: ， in is the comprehensive index of service quality, To achieve zero-effect benchmark service quality, are the coefficients of the B-spline basis function, estimated by the least squares method, are the coefficients of the B-spline basis function, are the coefficients of the B-spline basis function, is the B-spline basis function, k is the order of the basis function, is the growth factor volatility, Corresponding to flexible resource replenishment, is the random effect of social group z, is random fluctuation, t is the time index; Based on the time-varying linear mixed-effect model, the Markov chain Monte Carlo algorithm is used to fit the fixed effect parameters and random effect parameters to obtain the time variation coefficient and random effect. Based on the difference mark, the first-order derivative of the time variation coefficient, and the second-order derivative of the time variation coefficient, the growth stage labels are divided according to the time-varying trend. The service elasticity coefficient is calculated based on the time variation coefficient. The service elasticity coefficient calculation formula is: ， in is the service elasticity coefficient at the current time t, is any historical time point between 0 and t, for The ratio of the change rate of time resource investment to the change rate of the number of people covered by the service, The influence range of historical elasticity is calibrated by the variance contribution of the second-order derivative within the historical elastic sliding window. for The temporal coefficient of variation of time, for The proportion of resource input of a certain type of subject in time, is the complaint attenuation rate, is the first-order derivative of the comprehensive index of service quality at time t, is the ratio of complaint processing delay time to time window, is the service quality change rate.

6. A social group service quality assessment method according to claim 1, characterized in that: The matching imbalance analysis method includes: Based on the time series, the product of the service elasticity coefficient and the resource allocation concentration is calculated, the growth factor volatility is divided by the product, and the logarithm absolute value of the calculation result is used as the resource supply deviation. The square root of the absolute value of the difference between the average treatment effect and the service penetration rate is taken to obtain the demand coverage gap. The matching imbalance is obtained by adding the product of the preset industry weight and the resource supply deviation plus the product of the preset industry weight and the demand coverage gap.

7. A social group service quality assessment method according to claim 1, characterized in that: The method for obtaining the three-node causal loop includes: The growth entropy is calculated based on the time series according to the matching imbalance and growth stage label. The growth entropy formula is: ， in For the The contribution probability of the class growth stage, is the deviation adjustment factor, , is the historical average imbalance, n is the number of labels in the growth stage, For the The matching imbalance degree of the class growth stage, is the service elasticity coefficient of social group z, For the Feedback continuity index under the growth stage, is the growth entropy of social group z; According to the time series, resource supply deviation and growth factor volatility are used as resource supply nodes, service elasticity coefficient and service quality comprehensive index are used as service response nodes, feedback continuity index and demand coverage gap are used as feedback adjustment nodes, and the three nodes are connected to establish a three-node causal loop; The average service elasticity coefficient in the past year is used as the resource service edge weight, the average feedback continuity index is used as the service feedback edge weight, and the ratio of the current matching imbalance to the average matching imbalance is used as the feedback resource edge weight. If the absolute value of the month-on-month change rate of growth entropy in adjacent time periods exceeds 10%, the three-node causal loop update is triggered. The causal loop update direction is determined according to the sign of the first-order derivative of the time variation coefficient between nodes, and the update intensity is determined according to the absolute value of the second-order derivative of the time variation coefficient.

8. A social group service quality assessment method according to claim 1, characterized in that: The method for obtaining the service quality evaluation result includes: Based on the three-node causal loop, the nodes are mapped into the three feature layers of the causal graph neural network. The edge weights of the three-node causal loop are used as the edge weights of the causal graph. Based on the service activity data, the time series, matching imbalance, and growth entropy are used as input data. The comprehensive service quality index is used as the training label. The input data is divided into training set, validation set, and test set. The training set and training labels are used to input the causal graph neural network. The growth entropy is used as the regularization term of the loss function, and the overfitting state of the neural network is adjusted according to the changes in the loss function value and accuracy. The difference between the neural network prediction value and the training label is fed back to each layer of the neural network through back propagation. If the accuracy of the validation set is improved by less than 0.1% for five consecutive rounds, the training is terminated. The test set is input into the trained causal graph neural network, and the model performance is optimized according to the mean square error and mean absolute error output by the test set to obtain a service quality evaluation model. The time series sequence, matching imbalance and growth entropy of the data to be evaluated are input into the service quality evaluation model to obtain the service quality evaluation index.

9. A social group service quality evaluation system for executing a social group service quality evaluation method according to any one of claims 1 to 8, characterized in that: The system comprises: Service data collection module: used to collect service activity data of social groups, including service process data, resource input data, effect feedback data and coverage population data; Service data feature extraction module: used to divide the service activity data based on the time series and volatility to obtain the growth factor volatility and feedback continuity index; Social group identification module: used to differentially identify social groups based on the growth factor volatility and feedback continuity index, obtain a dynamic feature set, use the dynamic feature set to input the time-varying linear mixed effect model, and obtain growth stage labels and service elasticity coefficients; Social Group Causal Loop Analysis Module: This module analyzes the matching imbalance between resource supply and service demand of social groups based on the growth factor volatility and the service elasticity coefficient, obtains growth entropy based on the matching imbalance and the growth stage label, constructs a three-node causal loop based on the growth entropy and the entropy ring change rate, and dynamically updates the causal loop using the elasticity coefficient feedback loop. Social group service quality evaluation module: used to obtain a service quality evaluation model based on the three-node causal loop and growth entropy, input the service activity data into the service quality evaluation model, and obtain a service quality evaluation result.

Images