Convolutional neural network-based user satisfaction simulation method and system

Abstract

The invention discloses a user satisfaction simulation method and system based on a convolutional neural network, and the method comprises the steps: collecting behavior data and product test data of a user, and carrying out the preprocessing of the behavior data and the product test data; performing characteristic rule extraction on the behavior data by adopting a convolutional neural network to obtain characteristic data, and performing behavior semantic recognition on the characteristic data to obtain semantic data; associating the product test data with the semantic data to obtain associated data, performing directed weighting according to the associated data to obtain a data weight, and constructing a satisfaction function by adopting the semantic data and the data weight; and constructing a user satisfaction simulation model according to the satisfaction function, optimizing the user satisfaction simulation model, inputting the to-be-simulated data into the user satisfaction simulation model, and outputting a simulation result. The method not only improves the precision of user satisfaction simulation, but also has good interpretability, and is directly applied to a user satisfaction simulation system.

Description

Technical Field

[0001] The present invention relates to the field of simulation, and in particular to a user satisfaction simulation method and system based on a convolutional neural network. Background Art

[0002] In today's digital age, user satisfaction has become one of the key indicators to measure the success of a product or service. In order to more accurately understand and predict user satisfaction with products, enterprises and service providers are increasingly relying on big data analysis and machine learning technologies. Although traditional user satisfaction evaluation methods, such as questionnaires and user interviews, can provide certain feedback, they are often limited by sample size, strong subjectivity, and delayed feedback. Therefore, it is particularly important to develop a simulation method that can automatically, efficiently, and accurately evaluate user satisfaction.

[0003] With the rapid development of artificial intelligence technology, especially the wide application of convolutional neural networks in the field of deep learning in image recognition, natural language processing and other fields, new ideas have been provided for user satisfaction evaluation. With its powerful feature extraction ability, convolutional neural networks can automatically learn and extract key features from massive and complex data, which is crucial for analyzing implicit patterns in user behavior data.

[0004] However, directly applying CNN to user satisfaction evaluation still faces many challenges. Therefore, a new user satisfaction simulation method based on convolutional neural network is proposed to improve the accuracy and real-time performance of satisfaction simulation. Summary of the invention

[0005] The purpose of the present invention is to provide a user satisfaction simulation method based on convolutional neural network.

[0006] To achieve the above object, the present invention is implemented according to the following technical solutions: The present invention comprises the following steps: Collecting user behavior data and product test data, and preprocessing the behavior data and the product test data; Using a convolutional neural network to extract characteristic rules from the behavior data to obtain characteristic data, and performing behavioral semantic recognition on the characteristic data to obtain semantic data; Associating the product test data with the semantic data to obtain associated data, performing directed weighting according to the associated data to obtain data weights, and constructing a satisfaction function using the semantic data and the data weights; The user satisfaction simulation model is constructed according to the satisfaction function, the user satisfaction simulation model is optimized, the data to be simulated is input into the user satisfaction simulation model, and the simulation result is output.

[0007] Furthermore, a method for extracting characteristic rules from the behavior data using a convolutional neural network to obtain characteristic data includes: The image features of the behavior data are extracted using a convolutional neural network to obtain the image features, and the image features are screened. The expression is: , , , The first image feature is , the second image feature is , the binary XOR operation algorithm is , the total number of digits whose statistical operation result is 1 is , the Hamming distance between the first image feature and the second image feature is H, and the minimum value of the Hamming distance is , the remainder is %, and the maximum value of the Hamming distance is , the Hamming distance between the b-th image feature pair is , the automatic change parameters are , set the constants to , ; To satisfy the Hamming distance The image features of the judgment conditions are further screened, and the expression is: , , The window centered on the image feature is G, and the offset coordinates inside the window G are , the first image feature at the s+1th moment is , the first image feature at the sth moment is , the first image feature and The similarity score is U, and the first image feature of the behavior data at the s+1th moment is , the first image feature of the behavior data at the sth moment is , the similarity of the first image feature of the first behavior data and the second behavior data is ; Given a similarity threshold, when the similarity U is less than the similarity threshold, the image feature is output as the key feature; Calculate the degree of change of behavioral data: , The bth key feature at the s+1th moment is , the bth key feature at the sth moment is , the number of key features is , the control coefficient is , the Euclidean norm is , adjust the parameters to , the similarity of the bth key feature between the sth moment and the s+1th moment is , the change degree of behavior data between the sth moment and the s+1th moment is ; The degree of change of the behavior data is classified into groups and the data is exported as characteristic data.

[0008] Furthermore, the method of performing behavioral semantic recognition on the characteristic data to obtain semantic data includes: The characteristic data is input into a behavior semantic recognition model, which includes a non-local feature extraction module and an attention mechanism module; The non-local feature extraction module reorganizes the semantic information matrix obtained by fusion in the dynamic range by reshaping, and obtains behavioral information by enabling local semantic information to interact with global semantic information through matrix multiplication; Calculate the similarity between behavior information and feature data: , The query vector is Q, the key vector is K, the value vector is V, and the dimension of K is , transposed to T, semantic information is Z, and the weight matrix of the query vector is , the weight matrix of the key vector is , the weight matrix of the value vector is , the control coefficient is , the characteristic data is u, the behavior information is f, and the bias is y; The attention mechanism module uses the graph convolutional network of information interaction to model the behavior information in time and space, and introduces adaptive parameters, which are expressed as: , The adaptive parameters are , the channel dimension is D, and the constants are p and v respectively; The channel weights of feature information are adjusted according to the importance of the graph convolutional network through adaptive parameters, and the effective behavior information of feature data is extracted through aggregation to calculate the importance of feature data: , , The ath characteristic data is , characteristic data The importance of , the amount of behavior information is , the cth behavior information is , the jth behavior information is , the number of characteristic data is M, characteristic data China Bank Information and behavioral information The similarity is , the norm is ; Clustering is performed according to the similarity between behavioral information to obtain cluster data, and the cluster data is output as semantic data.

[0009] Furthermore, the method of constructing a satisfaction function using the semantic data and the data weights includes: Classify semantic data through natural language semantic analysis of Naive Bayes and convert semantic data into attitude score data; The satisfaction function is constructed based on the attitude score data and data weights. The expression is: , The sth data weight is , the cth attitude score data of the sth semantic data is , the amount of semantic data is , the amount of semantic data is , the satisfaction function is .

[0010] Furthermore, the method for constructing the user satisfaction simulation model according to the satisfaction function includes: The objective function of the user satisfaction simulation model is constructed by adding the satisfaction function and the loss function; The user satisfaction simulation model includes image feature extraction algorithm, feature comparison algorithm, convolutional neural network, and particle swarm optimization algorithm; The image feature extraction algorithm extracts features from the input image data to obtain user features; The feature comparison algorithm compares user features with behavioral semantic features to obtain comparison data; The convolutional neural network mines correlations based on the comparative data, combines the user's historical behavior and preferences based on the correlation, and uses the target data to output the user's simulated satisfaction; The particle swarm optimization algorithm continuously iterates through the variance of user simulation satisfaction and user actual satisfaction until the variance is lower than 0.13, then stops iterating.

[0011] Furthermore, the method for optimizing the user satisfaction simulation model includes: The intelligent group optimization algorithm is introduced, the satisfaction error is used as the fitness function, and the minimum satisfaction error is used as the search target; The chaotic mapping of particle population is expressed as: , The i-th original sequence is , the i+1th chaotic mapping sequence is , the control parameters are , a random number from 0 to 1 is , the convergence parameter is ; Take the particle position with the smallest fitness as the best position and calculate the particle position: , The initial optimal position of the particle in the tth iteration is , the convergence parameter is , the position of the wth particle in the t+1th iteration is , a random number between -1 and 1 is , the position of the wth particle in the tth iteration is ; Use the inhibition factor to update the particle position and obtain the inhibition position. The expression is: , , The suppression position of the wth particle in the t+1th iteration is The inhibitory factor is , the maximum number of iterations is , the current iteration number is t, and the random numbers from 0 to 1 are , , the position of the random particle at the tth iteration is ; The dynamic selection probability is used to update the position of the particle to obtain the dynamic position, which is expressed as: , , The adaptive weight of the tth iteration is , the suppression position of the wth particle in the tth iteration is , the current number of iterations is t, and the maximum number of iterations is , the normal distribution random number is , the best position of the tth iteration is , the dynamic position of the wth particle in the t+1th iteration is , the dynamic selection probability is ; Iterate continuously until the satisfaction error is minimized, otherwise adjust the particle population and update the dynamic selection probability.

[0012] The second aspect is the user satisfaction simulation system based on convolutional neural network, including: Data collection module: used to collect user behavior data and product test data, and pre-process the behavior data and the product test data; Feature extraction module: used to extract characteristic rules from the behavior data using a convolutional neural network to obtain characteristic data, and to perform behavioral semantic recognition on the characteristic data to obtain semantic data; Weighted construction module: used for associating the product test data with the semantic data to obtain associated data, performing directed weighting according to the associated data to obtain data weights, and constructing a satisfaction function using the semantic data and the data weights; Model building module: used to build the user satisfaction simulation model according to the satisfaction function, optimize the user satisfaction simulation model, input the data to be simulated into the user satisfaction simulation model, and output the simulation results.

[0013] The beneficial effects of the present invention are: The present invention is a user satisfaction simulation method and system based on convolutional neural network. Compared with the prior art, the present invention has the following technical effects: The present invention can improve the accuracy of user satisfaction simulation based on convolutional neural networks through preprocessing, characteristic law extraction, behavior semantic recognition, data association, directed weighting, construction of satisfaction function, model building and model optimization steps, thereby improving the precision of user satisfaction simulation based on convolutional neural networks, optimizing user satisfaction simulation based on convolutional neural networks, greatly saving resources, and improving work efficiency, and can realize intelligent simulation of user satisfaction, and perform characteristic law extraction and behavior semantic recognition on user satisfaction simulation based on convolutional neural networks in real time, which is of great significance to user satisfaction simulation based on convolutional neural networks, can adapt to user satisfaction simulation based on convolutional neural networks of different standards and different user satisfaction simulation needs based on convolutional neural networks, and has a certain universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a flow chart of the steps of the user satisfaction simulation method based on convolutional neural network of the present invention. DETAILED DESCRIPTION

[0015] The present invention is further described below by means of specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0016] The user satisfaction simulation method and system based on convolutional neural network of the present invention include the following steps: like Figure 1 As shown, in this embodiment, the following steps are included: Collecting user behavior data and product test data, and preprocessing the behavior data and the product test data; In the actual evaluation, the behavioral data of user A using the XXX online education system to live broadcast courses and the product test data of the XXX online education system are collected; Course ID4001, duration 30min, difficulty level medium, teacher rating 4.82; Using a convolutional neural network to extract characteristic rules from the behavior data to obtain characteristic data, and performing behavioral semantic recognition on the characteristic data to obtain semantic data; In the actual evaluation, the characteristic data of user A are disappearance for 1 minute, mouth corners raised 2 times, mouth opened 3 times, and concentration time of 26 minutes; The semantic data are that the course has some difficulties, the course feels good overall, and the teacher’s explanation is good; Associating the product test data with the semantic data to obtain associated data, performing directed weighting according to the associated data to obtain data weights, and constructing a satisfaction function using the semantic data and the data weights; In the actual evaluation, the weighted association rule mining algorithm is used to perform directionally weighted weighting on the semantic data to obtain the primary data weight, and the product test data is used to correct the primary data weight to obtain the data weight; The data weights are: the course has some difficulties 0.258, the course feels good overall 0.896, and the teacher's explanation is good 0.831; The user satisfaction simulation model is constructed according to the satisfaction function, the user satisfaction simulation model is optimized, the data to be simulated is input into the user satisfaction simulation model, and the simulation result is output.

[0017] In this embodiment, the method of extracting characteristic rules from the behavior data using a convolutional neural network to obtain characteristic data includes: The image features of the behavior data are extracted using a convolutional neural network to obtain the image features, and the image features are screened. The expression is: , , , The first image feature is , the second image feature is , the binary XOR operation algorithm is , the total number of digits whose statistical operation result is 1 is , the Hamming distance between the first image feature and the second image feature is H, and the minimum value of the Hamming distance is , the remainder is %, and the maximum value of the Hamming distance is , the Hamming distance between the b-th image feature pair is , the automatic change parameters are , set the constants to , ; To satisfy the Hamming distance The image features of the judgment conditions are further screened, and the expression is: , , The window centered on the image feature is G, and the offset coordinates inside the window G are , the first image feature at the s+1th moment is , the first image feature at the sth moment is , the first image feature and The similarity score is U, and the first image feature of the behavior data at the s+1th moment is , the first image feature of the behavior data at the sth moment is , the similarity of the first image feature of the first behavior data and the second behavior data is ; Given a similarity threshold, when the similarity U is less than the similarity threshold, the image feature is output as the key feature; Calculate the degree of change of behavioral data: , The bth key feature at the s+1th moment is , the bth key feature at the sth moment is , the number of key features is , the control coefficient is , the Euclidean norm is , adjust the parameters to , the similarity of the bth key feature between the sth moment and the s+1th moment is , the change degree of behavior data between the sth moment and the s+1th moment is ; The degree of change of the behavior data is classified into groups and the data is exported as characteristic data.

[0018] In this embodiment, the method of performing behavioral semantic recognition on the characteristic data to obtain semantic data includes: The characteristic data is input into a behavior semantic recognition model, which includes a non-local feature extraction module and an attention mechanism module; The non-local feature extraction module reorganizes the semantic information matrix obtained by fusion in the dynamic range by reshaping, and obtains behavioral information by enabling local semantic information to interact with global semantic information through matrix multiplication; Calculate the similarity between behavior information and feature data: , The query vector is Q, the key vector is K, the value vector is V, and the dimension of K is , transposed to T, semantic information is Z, and the weight matrix of the query vector is , the weight matrix of the key vector is , the weight matrix of the value vector is , the control coefficient is , the characteristic data is u, the behavior information is f, and the bias is y; The attention mechanism module uses the graph convolutional network of information interaction to model the behavior information in time and space, and introduces adaptive parameters, which are expressed as: , The adaptive parameters are , the channel dimension is D, and the constants are p and v respectively; The channel weights of feature information are adjusted according to the importance of the graph convolutional network through adaptive parameters, and the effective behavior information of feature data is extracted through aggregation to calculate the importance of feature data: , , The ath characteristic data is , characteristic data The importance of , the amount of behavior information is , the cth behavior information is , the jth behavior information is , the number of characteristic data is M, characteristic data China Bank Information and behavioral information The similarity is , the norm is ; Clustering is performed according to the similarity between behavioral information to obtain cluster data, and the cluster data is output as semantic data.

[0019] In this embodiment, the method of constructing a satisfaction function using the semantic data and the data weight includes: Classify semantic data through natural language semantic analysis of Naive Bayes and convert semantic data into attitude score data; The satisfaction function is constructed based on the attitude score data and data weights. The expression is: , The sth data weight is , the cth attitude score data of the sth semantic data is , the amount of semantic data is , the amount of semantic data is , the satisfaction function is .

[0020] In this embodiment, the method for constructing the user satisfaction simulation model according to the satisfaction function includes: The objective function of the user satisfaction simulation model is constructed by adding the satisfaction function and the loss function; The user satisfaction simulation model includes image feature extraction algorithm, feature comparison algorithm, convolutional neural network, and particle swarm optimization algorithm; The image feature extraction algorithm extracts features from the input image data to obtain user features; The feature comparison algorithm compares user features with behavioral semantic features to obtain comparison data; The convolutional neural network mines correlations based on the comparative data, combines the user's historical behavior and preferences based on the correlation, and uses the target data to output the user's simulated satisfaction; The particle swarm optimization algorithm continuously iterates through the variance of user simulation satisfaction and user actual satisfaction until the variance is lower than 0.13, then stops iterating.

[0021] In this embodiment, the method for optimizing the user satisfaction simulation model includes: The intelligent group optimization algorithm is introduced, the satisfaction error is used as the fitness function, and the minimum satisfaction error is used as the search target; The chaotic mapping of particle population is expressed as: , The i-th original sequence is , the i+1th chaotic mapping sequence is , the control parameters are , a random number from 0 to 1 is , the convergence parameter is ; Take the particle position with the smallest fitness as the best position and calculate the particle position: , The initial optimal position of the particle in the tth iteration is , the convergence parameter is , the position of the wth particle in the t+1th iteration is , a random number between -1 and 1 is , the position of the wth particle in the tth iteration is ; Use the inhibition factor to update the particle position and obtain the inhibition position. The expression is: , , The suppression position of the wth particle in the t+1th iteration is The inhibitory factor is , the maximum number of iterations is , the current iteration number is t, and the random numbers from 0 to 1 are , , the position of the random particle at the tth iteration is ; The dynamic selection probability is used to update the position of the particle to obtain the dynamic position, which is expressed as: , , The adaptive weight of the tth iteration is , the suppression position of the wth particle in the tth iteration is , the current number of iterations is t, and the maximum number of iterations is , the normal distribution random number is , the best position of the tth iteration is , the dynamic position of the wth particle in the t+1th iteration is , the dynamic selection probability is ; Iterate continuously until the satisfaction error is minimized, otherwise adjust the particle population and update the dynamic selection probability.

[0022] The second aspect is the user satisfaction simulation system based on convolutional neural network, including: Data collection module: used to collect user behavior data and product test data, and pre-process the behavior data and the product test data; Feature extraction module: used to extract characteristic rules from the behavior data using a convolutional neural network to obtain characteristic data, and to perform behavioral semantic recognition on the characteristic data to obtain semantic data; Weighted construction module: used for associating the product test data with the semantic data to obtain associated data, performing directed weighting according to the associated data to obtain data weights, and constructing a satisfaction function using the semantic data and the data weights; Model building module: used to build the user satisfaction simulation model according to the satisfaction function, optimize the user satisfaction simulation model, input the data to be simulated into the user satisfaction simulation model, and output the simulation results.

[0023] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A user satisfaction simulation method based on convolutional neural network, characterized in that: The following steps are involved: Collecting user behavior data and product test data, and preprocessing the behavior data and the product test data; Using a convolutional neural network to extract characteristic rules from the behavior data to obtain characteristic data, and performing behavioral semantic recognition on the characteristic data to obtain semantic data; Associating the product test data with the semantic data to obtain associated data, performing directed weighting according to the associated data to obtain data weights, and constructing a satisfaction function using the semantic data and the data weights; The user satisfaction simulation model is constructed according to the satisfaction function, the user satisfaction simulation model is optimized, the data to be simulated is input into the user satisfaction simulation model, and the simulation result is output.

2. The user satisfaction simulation method based on convolutional neural network according to claim 1 is characterized in that: The method of extracting characteristic rules from the behavior data using a convolutional neural network to obtain characteristic data includes: The image features of the behavior data are extracted using a convolutional neural network to obtain the image features, and the image features are screened. The expression is: ， ， ， The first image feature is , the second image feature is , the binary XOR operation algorithm is , the total number of digits whose statistical operation result is 1 is , the Hamming distance between the first image feature and the second image feature is H, and the minimum value of the Hamming distance is , the remainder is %, and the maximum value of the Hamming distance is , the Hamming distance between the b-th image feature pair is , the automatic change parameters are , set the constants to , ; To satisfy the Hamming distance The image features of the judgment conditions are further screened, and the expression is: ， ， The window centered on the image feature is G, and the offset coordinates inside the window G are , the first image feature at the s+1th moment is , the first image feature at the sth moment is , the first image feature and The similarity score is U, and the first image feature of the behavior data at the s+1th moment is , the first image feature of the behavior data at the sth moment is , the similarity of the first image feature of the first behavior data and the second behavior data is ; Given a similarity threshold, when the similarity U is less than the similarity threshold, the image feature is output as the key feature; Calculate the degree of change of behavioral data: ， The bth key feature at the s+1th moment is , the bth key feature at the sth moment is , the number of key features is , the control coefficient is , the Euclidean norm is , adjust the parameters to , the similarity of the bth key feature between the sth moment and the s+1th moment is , the change degree of behavior data between the sth moment and the s+1th moment is ; The variability of the behavioral data is classified into groups and the data is exported as characteristic data.

3. The user satisfaction simulation method based on convolutional neural network according to claim 1 is characterized in that: The method of performing behavioral semantic recognition on the characteristic data to obtain semantic data includes: The characteristic data is input into a behavior semantic recognition model, which includes a non-local feature extraction module and an attention mechanism module; The non-local feature extraction module reorganizes the semantic information matrix obtained by fusion in the dynamic range by reshaping, and obtains behavioral information by enabling local semantic information to interact with global semantic information through matrix multiplication; Calculate the similarity between behavior information and feature data: ， The query vector is Q, the key vector is K, the value vector is V, and the dimension of K is , transposed to T, semantic information is Z, and the weight matrix of the query vector is , the weight matrix of the key vector is , the weight matrix of the value vector is , the control coefficient is , the characteristic data is u, the behavior information is f, and the bias is y; The attention mechanism module uses the graph convolutional network of information interaction to model the behavior information in time and space, and introduces adaptive parameters, which are expressed as: ， The adaptive parameters are , the channel dimension is D, and the constants are p and v respectively; The channel weights of feature information are adjusted according to the importance of the graph convolutional network through adaptive parameters, and the effective behavior information of feature data is extracted through aggregation to calculate the importance of feature data: ， ， The ath characteristic data is , characteristic data The importance of , the amount of behavior information is , the cth behavior information is , the jth behavior information is , the number of characteristic data is M, characteristic data China Bank Information and behavioral information The similarity is , the norm is ; Clustering is performed according to the similarity between behavioral information to obtain cluster data, and the cluster data is output as semantic data.

4. The user satisfaction simulation method based on convolutional neural network according to claim 1 is characterized in that: The method of constructing a satisfaction function using the semantic data and the data weight comprises: Classify semantic data through natural language semantic analysis of Naive Bayes and convert semantic data into attitude score data; The satisfaction function is constructed based on the attitude score data and data weights. The expression is: ， The sth data weight is , the cth attitude score data of the sth semantic data is , the amount of semantic data is , the amount of semantic data is , the satisfaction function is .

5. The user satisfaction simulation method based on convolutional neural network according to claim 1 is characterized in that: The method for constructing the user satisfaction simulation model according to the satisfaction function comprises: The objective function of the user satisfaction simulation model is constructed by adding the satisfaction function and the loss function; The user satisfaction simulation model includes image feature extraction algorithm, feature comparison algorithm, convolutional neural network, and particle swarm optimization algorithm; The image feature extraction algorithm extracts features from the input image data to obtain user features; The feature comparison algorithm compares user features with behavioral semantic features to obtain comparison data; The convolutional neural network mines correlations based on the comparative data, combines the user's historical behavior and preferences based on the correlation, and uses the target data to output the user's simulated satisfaction; The particle swarm optimization algorithm continuously iterates through the variance of user simulation satisfaction and user actual satisfaction until the variance is lower than 0.13, then stops iterating.

6. The user satisfaction simulation method based on convolutional neural network according to claim 1 is characterized in that: The method for optimizing the user satisfaction simulation model comprises: The intelligent group optimization algorithm is introduced, the satisfaction error is used as the fitness function, and the minimum satisfaction error is used as the search target; The chaotic mapping of particle population is expressed as: ， The i-th original sequence is , the i+1th chaotic mapping sequence is , the control parameters are , a random number from 0 to 1 is , the convergence parameter is ; Take the particle position with the smallest fitness as the best position and calculate the particle position: ， The initial optimal position of the particle in the tth iteration is , the convergence parameter is , the position of the wth particle in the t+1th iteration is , a random number between -1 and 1 is , the position of the wth particle in the tth iteration is ; Use the inhibition factor to update the particle position and obtain the inhibition position. The expression is: ， ， The suppression position of the wth particle in the t+1th iteration is The inhibitory factor is , the maximum number of iterations is , the current iteration number is t, and the random numbers from 0 to 1 are , , the position of the random particle at the tth iteration is ; The dynamic selection probability is used to update the position of the particle to obtain the dynamic position, which is expressed as: ， ， The adaptive weight of the tth iteration is , the suppression position of the wth particle in the tth iteration is , the current number of iterations is t, and the maximum number of iterations is , the normal distribution random number is , the best position of the tth iteration is , the dynamic position of the wth particle in the t+1th iteration is , the dynamic selection probability is ; Iterate continuously until the satisfaction error is minimized, otherwise adjust the particle population and update the dynamic selection probability.

7. A user satisfaction simulation system based on convolutional neural network, used to execute the method according to any one of claims 1 to 6, characterized in that: include: Data collection module: used to collect user behavior data and product test data, and pre-process the behavior data and the product test data; Feature extraction module: used to extract characteristic rules from the behavior data using a convolutional neural network to obtain characteristic data, and to perform behavioral semantic recognition on the characteristic data to obtain semantic data; Weighted construction module: used for associating the product test data with the semantic data to obtain associated data, performing directed weighting according to the associated data to obtain data weights, and constructing a satisfaction function using the semantic data and the data weights; Model building module: used to build the user satisfaction simulation model according to the satisfaction function, optimize the user satisfaction simulation model, input the data to be simulated into the user satisfaction simulation model, and output the simulation results.

Images