MLP-based public service survey satisfaction influence factor analysis method
By using MLP model and one-hot encoding to process data in public service surveys and combining SHAP algorithm for analysis, the problem of limited expression ability of structural equation model when processing nonlinear data is solved, and the accuracy and interpretability of the regression model are improved.
Patent Information
- Application Number
- CN202411668183.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-21
- Publication Date
- 2025-03-25
AI Technical Summary
The structural equation model-based method has limited expression capabilities when processing public service survey data with high nonlinearity, and traditional numerical encoding methods lead to poor accuracy of regression models.
A multi-layer perceptron (MLP) model is used to combine single-hot encoding and SHAP algorithm to construct an analysis method for influencing factors of public service survey satisfaction. The survey data is processed through one-hot encoding, fuse special case samples, and use the MLP model to improve the accuracy of the regression model. The SHAP algorithm is used for model interpretation to generate quantitative results of influencing factors.
It improves the accuracy of the regression model, can effectively process nonlinear data sets, reduces the complexity of the model, and provides explanatory results of the model through the SHAP algorithm.
Smart Images
Figure CN119692839A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of regression models, and in particular to a method for analyzing factors affecting satisfaction in public service surveys based on MLP. Background Art
[0002] The analytical method applied to public services requires first training and obtaining a regression model based on public service satisfaction survey data, and then analyzing and evaluating the impact of survey indicators on satisfaction. The regression model is the core of the method, involving two key steps: data encoding and model training.
[0003] Numerical coding is a common method used in relevant analytical methods for data coding. This is because public service surveys are primarily conducted by inviting respondents to rate specific indicators. Scores are expressed numerically, with the magnitude of the numerical value representing the respondent's rating. Furthermore, to ensure the scientific and valid nature of survey data, values such as "0" or "99" are often used to indicate special cases such as "indicator not applicable" or "missing information," or survey samples with special cases are directly excluded for subsequent analysis. However, when coding special cases, regardless of the specific numerical value used, a certain mathematical relationship will be formed with the numerical value representing the respondent's rating, affecting the accuracy of the analytical results. Furthermore, the special cases in some survey samples themselves have practical significance and therefore need to be incorporated into the relevant analytical process.
[0004] For model training, Structural Equation Modeling (SEM) is often used in relevant analysis methods. This model constructs its linear equation group based on the assumed influence relationship between survey indicators.
[0005] Based on a system of linear equations, the SEM method employs partial least squares or the covariance criterion to calculate the correlation coefficient of the system of equations based on the corresponding survey data. Ultimately, the path coefficient w is used to quantify the influence between the survey indicators. However, the results of structural equation model training are easily affected by the size of the survey dataset. Furthermore, due to its theoretical foundation, the SEM has limited expressive power for datasets with a high degree of nonlinearity. Summary of the Invention
[0006] The purpose of the present invention is to provide a method for analyzing the influencing factors of public service survey satisfaction based on MLP in order to solve the problem that the method based on structural equation model has limited expression ability for data sets with high nonlinearity and does not adopt a reasonable numerical encoding method, resulting in poor accuracy of the regression model.
[0007] The above-mentioned purpose of this application is achieved through the following technical solutions: S1: Construct a survey data set based on survey indicators; S2: Use the one-hot encoding method to binarize the data samples of the survey dataset; S3: Use random sampling method to extract training set from survey data set and validation set ; S4: Construct an MLP model framework based on quality dimension, trust dimension and expectation dimension; S5: Connect the input and output of the three-dimensional MLP model framework twice to construct the overall MLP model of the survey data; S6: Train the overall MLP model using the training set and validate it using the validation set to obtain the regression model for the survey data. S7: Use the SHAP algorithm to perform calculations and analysis on the regression model, generate the Shapley values of each survey indicator for satisfaction, and draw a waterfall chart to complete the analysis of factors affecting public service survey satisfaction.
[0008] Optionally, step S1 includes: The total number of survey indicators is K, and the data sample is ; A set of scores representing service domain indicators, The number of indicators under the quality dimension is ,but ; Represents the service improvement confidence index score, The number of indicators under the trust dimension is ,but ; Represents the expected service quality index score, The number of indicators under the expected dimension is ,but ; Representative public satisfaction index score; 、 as well as are all positive integers; .
[0009] Optionally, step S2 includes: Assume that there are L kinds of ratings in the data sample, so After one-hot encoding, it is converted into .
[0010] In one embodiment of the present application, each indicator score of each data sample is converted into a binary code, and the encoding length L is the total number of possible score situations of the indicator score. Figure 2 , when the score of an indicator is "6", its corresponding one-hot encoding is "00000010000".
[0011] Optionally, step S4 includes: Classify public service survey indicators according to the three dimensions of quality, trust, and expectation, and construct a three-dimensional MLP model framework; The MLP model framework includes: input layer, intermediate layer and output layer, and the nodes in each layer are fully connected; The number of nodes in the input layer is equal to the one-hot encoding length L of the input data, that is, the number of dimension indicators; The number of nodes in the middle layer is half the number of nodes in the input layer but not less than the one-hot encoding length L; The number of nodes in the output layer is the length L of the data's one-hot encoding.
[0012] Optionally, step S5 includes: Based on the China Customer Satisfaction Index model, the input and output of the three-dimensional MLP model are connected twice, as follows: If the output of the MLP model in dimension A is also the input of the MLP model in dimension B, the output of the MLP model in dimension A is first concatenated with the output of the MLP model in dimension B, and the concatenated data is used as the input of the overall MLP model in dimension B. The data then passes through the input layer, the middle layer, and the output layer in sequence to obtain the output of the overall MLP model in dimension B.
[0013] Optionally, step S6 includes: Using the forward calculation and error back propagation mechanism, combined with the training set, the overall MLP model is trained. Each forward calculation and back propagation process is recorded as one training. The specific steps are as follows: The weight matrix in each layer of the overall MLP model and the bias matrix Perform random initialization; The data samples in the training set , input the overall MLP model; According to the forward calculation method shown in formula (1), the overall MLP model is calculated layer by layer to obtain the satisfaction prediction output of the training set ;
[0014] Where, represents the activation function; Encode the layers of the MLP model; Indicates the The input data of the layer; Use the cross entropy function to predict the output based on satisfaction and the true value of satisfaction of the training set , calculation error , and according to formula (2) the error Perform back propagation layer by layer to update the weight matrix and bias matrix;
[0015] Where, represents the learning rate; Indicates the The loss value of the layer; represents the updated weight matrix; represents the updated bias matrix.
[0016] Optionally, step S6 further includes: In each training, before updating the weight matrix and bias matrix, the overall MLP model of the current training round is used to calculate the validation set, obtain the predicted value of the validation set and calculate the accuracy of the predicted value. ; When the accuracy of the verification set Greater than the preset threshold and error When it is less than the preset threshold, the training process of the overall MLP model is terminated.
[0017] An electronic device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device performs an MLP-based public service survey satisfaction influencing factor analysis method.
[0018] A computer-readable storage medium stores instructions. When the instructions are executed, a method for analyzing factors affecting public service survey satisfaction based on MLP is executed.
[0019] The beneficial effects of the technical solution provided by this application are: 1. The public service survey data was processed using one-hot encoding, which merged the score data samples representing satisfaction with the special data samples representing special situations. This not only avoided operational errors caused by the numerical relationship between the score data samples and the special situation samples during the construction of the regression model or the calculation of the correlation coefficient, but also enabled the survey data set used for impact analysis to include special situation samples with practical significance, making the analysis structure more practical.
[0020] 2. The regression model of public service survey data adopts an MLP-based modeling method. On the one hand, it can effectively meet the analysis needs under the highly nonlinear characteristics of the survey data set and improve the accuracy of the regression model. On the other hand, by integrating the inherent relationship between the dimensions of indicators in the public service survey field, it greatly reduces the complexity of the traditional MLP model and improves the simplicity of the regression model.
[0021] 3. Because the survey data uses one-hot encoding, the corresponding MLP model parameters are also matrixed, making it impossible to directly use MLP model parameters as quantitative indicators of impact. The SHAP algorithm effectively addresses this issue and obtains accurate quantitative results. Furthermore, the SHAP algorithm is an interpretable algorithm for machine learning models and offers a better fit than other quantitative algorithms. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] The present application will be further described below with reference to the accompanying drawings and embodiments, in which: Figure 1 It is a step diagram in the embodiment of the present application; Figure 2 Schematic diagram of one-hot encoding in an embodiment of the present application; Figure 3 : is a framework diagram of the MLP model of the quality dimension in the embodiment of the present application; Figure 4 This is the overall MLP model framework diagram in the embodiment of this application; Figure 5 It is a Shapley value waterfall chart in the embodiment of the present application; Figure 6 It is a schematic diagram of the structure of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION
[0023] In order to have a clearer understanding of the technical features, purposes and effects of this application, the specific implementation methods of this application are now described in detail with reference to the accompanying drawings.
[0024] The embodiments of the present application provide a method for analyzing factors affecting satisfaction in a public service survey based on MLP.
[0025] Please refer to Figure 1 , Figure 1 This is a step diagram of a method for analyzing factors affecting public service satisfaction based on MLP in an embodiment of the present application, including: S1: Construct a survey data set based on survey indicators; S2: Use the one-hot encoding method to binarize the data samples of the survey dataset; S3: Use random sampling method to extract training set from survey data set and validation set ; The present application provides an embodiment as follows, wherein the random sampling method includes: Latin hypercube sampling and Monte Carlo sampling, etc.
[0026] S4: Construct an MLP model framework based on quality dimension, trust dimension and expectation dimension; S5: Connect the input and output of the three-dimensional MLP model framework twice to construct the overall MLP model of the survey data; S6: Train the overall MLP model using the training set and validate it using the validation set to obtain the regression model for the survey data. S7: Use the SHAP algorithm to perform calculations and analysis on the regression model, generate the Shapley values of each survey indicator for satisfaction, and draw a waterfall chart to complete the analysis of factors affecting public service survey satisfaction.
[0027] This application provides an embodiment as follows, using the SHAP (SHapley Additive exPlanations) algorithm. The numerical value corresponding to the Shapley value expresses the influence of the corresponding indicator on satisfaction, and the positive and negative signs of the values express the direction of the influence of the corresponding indicator on satisfaction. The Shapley value waterfall chart of the top indicators of the influence of the 15 indicators in the five service areas and the public trust and public expectation indicators on satisfaction in the example is shown in Figure 1. Figure 5 .
[0028] This application provides an embodiment as follows: To address the "black box" characteristics of machine learning models and enhance the interpretability of the models, researchers proposed the SHAP (Shapley Additive exPlanations) algorithm. This algorithm is based on game theory and helps researchers better understand the data transmission process of the model by quantifying the specific contribution of model input to model output.
[0029] Game theory is a fair distribution method that aims to quantify the contribution of participants to the total benefits in a cooperative game. The quantitative value of the participant's contribution is also called the Shapley value.
[0030] The calculation process of Shapley value is as follows
[0031] in, represents the overall MLP model, S represents the subset of the survey data set, N represents the input survey data set, Indicates the data samples.
[0032] After obtaining the Shapley value of each model input, the corresponding overall MLP model output can be expressed as follows:
[0033] in, Represents the output benchmark (usually the average of the training sample set).
[0034] The Shapley value consists of two parts: the sign and the value. The sign represents the direction of contribution to the output, and the value represents the degree of contribution to the output.
[0035] Step S1 includes: The total number of survey indicators is K, and the data sample is ; A set of scores representing service domain indicators, The number of indicators under the quality dimension is ,but ; Represents the service improvement confidence index score, The number of indicators under the trust dimension is ,but ; Represents the expected service quality index score, The number of indicators under the expected dimension is ,but ; Representative public satisfaction index score; 、 as well as are all positive integers; .
[0036] Step S2 includes: Assume that there are L kinds of ratings in the data sample, so After one-hot encoding, it is converted into .
[0037] In one embodiment of the present application, each indicator score of each data sample is converted into a binary code, and the encoding length L is the total number of possible score situations of the indicator score. Figure 2 , when the score of an indicator is "6", its corresponding one-hot encoding is "00000010000".
[0038] Step S4 includes: Classify public service survey indicators according to the three dimensions of quality, trust, and expectation, and construct a three-dimensional MLP model framework; The MLP model framework includes: input layer, intermediate layer and output layer, and the nodes in each layer are fully connected; The number of nodes in the input layer is equal to the one-hot encoding length L of the input data, that is, the number of dimension indicators; The number of nodes in the middle layer is half the number of nodes in the input layer but not less than the one-hot encoding length L; The number of nodes in the output layer is the length L of the data's one-hot encoding.
[0039] The present application provides an embodiment as follows: the trust dimension includes one indicator of confidence in service improvement, the expectation dimension includes one indicator of expected service quality, and the remaining survey indicators of public services are all classified into the quality dimension.
[0040] The present application provides an embodiment as follows. The public service survey indicators include relevant indicators in service areas such as public education, public employment, medical services, social security, public utilities, ecological environment, public transportation, public safety, public culture, public sports, elderly care services, government services, and indicators reflecting public trust, public expectations, and public satisfaction. The indicator score is an integer greater than 0, and the larger the value, the higher the satisfaction. In addition, 0 represents the special case where the respondent cannot evaluate the indicator. A specific example is used here to explain the relevant steps in detail.
[0041] This application provides an example as follows: In a survey, data was collected and influencing factors were analyzed for 15 indicators in five public service areas, three indicators in each area, and three indicators: confidence in service improvement, expected service quality, and service satisfaction. The indicator scores range from 1 to 10, with 0 indicating special circumstances.
[0042] This application provides an embodiment as follows: all indicators in the service field are classified into the quality dimension, and other indicators are classified into the trust and expectation dimensions according to the connotation of the indicators. At the same time, the trust dimension and the expectation dimension each contain only one indicator, namely, confidence in service improvement and expected service quality. Therefore, the number of input layer nodes of the two dimensions is 11 (the number of dimension indicators is 1*11), the number of middle layer nodes is 11, and the number of output layer nodes is 11. The quality dimension in this example contains a total of 15 indicators in 5 service fields, so the MLP model framework under this dimension is referenced. Figure 3 The number of input layer nodes is 165 (the number of dimension indicators is 15*11), the number of middle layer nodes is 82, and the number of output layer nodes is 11.
[0043] Step S5 includes: Based on the China Customer Satisfaction Index model, the input and output of the three-dimensional MLP model are connected twice, as follows: When the output of the MLP model in dimension A is simultaneously the input of the MLP model in dimension B, first concatenate the output of the MLP model in dimension A with the output of the MLP model in dimension B, and then use the concatenated data as the input of dimension B of the overall MLP model. Then, successively pass through the input layer, the intermediate layer, and the output layer to obtain the output of dimension B of the overall MLP model.
[0044] Specifically, as Figure 4 shown, in the overall MLP model, when the output of the MLP model in the trust dimension is simultaneously the input of the MLP model in the expectation dimension (i.e., the expected dimension), first concatenate the output of the MLP model in the trust dimension (Y_trust) with the output of the MLP model in the expectation dimension (Y_exp), and then use the concatenated data as the input of the expectation dimension of the overall MLP model. Then, successively pass through the input layer, the intermediate layer, and the output layer to obtain the output of the expectation dimension of the overall MLP model. The connections between the above layers are fully connected. In this way, the overall model of the survey data is constructed.
[0045] The present application provides an embodiment as follows. Take the satisfaction as the output of the overall MLP (Multilayer Perceptron) model, and the framework structure refers to Figure 4 .
[0046] Step S6 includes: Using the forward calculation and error backpropagation mechanism, combined with the training set, train the overall MLP model. Each forward calculation and backpropagation process is recorded as one training. The specific steps are as follows: Randomly initialize the weight matrix and the bias matrix in each layer of the overall MLP model; Input the data samples in the training set into the overall MLP model; Perform layer-by-layer calculation on the overall MLP model according to the forward calculation method shown in Equation (1) to obtain the satisfaction prediction output of the training set ;
[0047] In the formula, represents the activation function; is the layer encoding of the MLP model; represents the input data of the th layer; Adopt the cross-entropy function, according to the satisfaction prediction output and the true value of the satisfaction in the training set , calculate the error Perform back propagation layer by layer to update the weight matrix and bias matrix;
[0048] Where, represents the learning rate; Indicates the The loss value of the layer; represents the updated weight matrix; represents the updated bias matrix.
[0049] Step S6 further includes: In each training, before updating the weight matrix and bias matrix, the overall MLP model of the current training round is used to calculate the validation set, obtain the predicted value of the validation set and calculate the accuracy of the predicted value. ; When the accuracy of the verification set Greater than the preset threshold and error When it is less than the preset threshold, the training process of the overall MLP model is terminated.
[0050] The present application provides an embodiment as follows: In each training, before updating the weight matrix and bias matrix, the current MLP model is used to calculate the model validation set, obtain the predicted value of the model validation set and calculate its prediction accuracy. When the accuracy of the model validation set is And the model training set error When , the training process of the MLP model is terminated.
[0051] This application also discloses an electronic device. Figure 6 , Figure 6 Schematic diagram of the structure of an electronic device disclosed in an embodiment of the present application. The electronic device 500 may include: at least one processor 501, at least one network interface 504, a user interface 503, a memory 505, and at least one communication bus 502.
[0052] The communication bus 502 is used to implement the connection and communication between these components.
[0053] The user interface 503 may include a display screen, and the optional user interface 503 may also include a standard wired interface or a wireless interface.
[0054] The network interface 504 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).
[0055] The present application also discloses a computer-readable storage medium storing a plurality of instructions suitable for loading by a processor to execute the above-mentioned method for analyzing factors affecting public service survey satisfaction based on MLP.
[0056] The above are merely exemplary embodiments of the present disclosure and are not intended to limit the scope of the present disclosure. In other words, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure.
[0057] This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not described herein. The description and examples are to be considered as exemplary only, and the scope and spirit of the present disclosure are to be defined by the claims.
Claims
1. A method for analyzing factors affecting public service survey satisfaction based on MLP, characterized in that: The method comprises the following steps: S1: Construct a survey data set based on survey indicators; S2: Binarize the data samples of the survey dataset using the one-hot encoding method; S3: Use random sampling method to extract training set from survey data set and validation set ; S4: Construct an MLP model framework based on quality dimension, trust dimension and expectation dimension; S5: connect the input and output of the three-dimensional MLP model framework twice to construct the overall MLP model of the survey data; S6: Train the overall MLP model through the training set and verify it through the validation set to obtain the regression model of the survey data; S7: Use the SHAP algorithm to calculate and analyze the regression model, generate the Shapley value of each survey indicator for satisfaction and draw a waterfall chart to complete the analysis of factors affecting public service survey satisfaction.
2. The method for analyzing factors affecting public service survey satisfaction based on MLP as claimed in claim 1, characterized in that: Step S1 includes: The total number of survey indicators is K, and the data sample is ; A set of scores representing service domain indicators, The number of indicators under the quality dimension is ,but ; Represents the service improvement confidence index score, The number of indicators under the trust dimension is ,but ; Represents the expected service quality index score, The number of indicators under the expected dimension is ,but ; Representative public satisfaction index score; , as well as All are positive integers; .
3. The method for analyzing factors affecting public service survey satisfaction based on MLP as claimed in claim 2, characterized in that: Step S2 includes: Assume that there are L kinds of ratings in the data sample, so After one-hot encoding, it is converted into .
4. The method for analyzing factors affecting public service survey satisfaction based on MLP as claimed in claim 1, characterized in that: Step S4 includes: The survey indicators of public services are classified into three dimensions: quality, trust and expectation, and a three-dimensional MLP model framework is constructed; The MLP model framework includes: input layer, intermediate layer and output layer, and the nodes in each layer are fully connected; The number of nodes in the input layer is equal to the one-hot encoding length L of the input data, that is, the number of dimension indicators; The number of nodes in the middle layer is half of the number of nodes in the input layer but not less than the one-hot encoding length L; The number of nodes in the output layer is the length L of the data's one-hot encoding.
5. The method for analyzing factors affecting public service survey satisfaction based on MLP as claimed in claim 1, characterized in that: Step S5 includes: Based on the China Customer Satisfaction Index model, the input and output of the three-dimensional MLP model are connected twice, as follows: If the output of the MLP model in dimension A is also the input of the MLP model in dimension B, the output of the MLP model in dimension A is first concatenated with the output of the MLP model in dimension B, and the concatenated data is used as the input of the overall MLP model in dimension B. The data is then passed through the input layer, the middle layer, and the output layer in sequence to obtain the output of the overall MLP model in dimension B.
6. The method for analyzing factors affecting public service survey satisfaction based on MLP as claimed in claim 1, characterized in that: Step S6 includes: Using the forward calculation and error back propagation mechanism, combined with the training set, the overall MLP model is trained. Each forward calculation and back propagation process is recorded as one training. The specific steps are as follows: For the weight matrix in each layer of the overall MLP model and the bias matrix Perform random initialization; The data samples in the training set , input the overall MLP model; According to the forward calculation method shown in formula (1), the overall MLP model is calculated layer by layer to obtain the satisfaction prediction output of the training set ; In the formula, represents the activation function; Encode the layers of the MLP model; Indicates The input data of the layer; Use the cross entropy function to predict the output based on satisfaction And the satisfaction value of the training set , calculation error , and according to formula (2) the error Perform back propagation layer by layer to update the weight matrix and bias matrix; In the formula, represents the learning rate; Indicates The loss value of the layer; represents the updated weight matrix; represents the updated bias matrix.
7. The method for analyzing factors affecting public service survey satisfaction based on MLP as claimed in claim 6, characterized in that: Step S6 also includes: In each training, before updating the weight matrix and bias matrix, the overall MLP model of the current training round is used to calculate the validation set, obtain the predicted value of the validation set and calculate the accuracy of the predicted value. ; When the accuracy of the validation set Greater than the preset threshold and error When it is less than the preset threshold, the training process of the overall MLP model is terminated.
8. An electronic device, characterized in that: The electronic device comprises a processor (501), a memory (505), a user interface (503) and a network interface (504), wherein the memory (505) is used to store instructions, the user interface (503) and the network interface (504) are used to communicate with other devices, and the processor (501) is used to execute the instructions stored in the memory (505) so that the electronic device executes the method according to any one of claims 1 to 7.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed by a computer, the method according to any one of claims 1 to 7 is executed.