Artificial Intelligence-Based Personalized E-Government Service Recommendation Method and System

Abstract

The present invention discloses a personalized government service recommendation method and system based on artificial intelligence. The method includes data collection, data preprocessing, user classification model construction, service recommendation model construction, and personalized service recommendation. The present invention relates to the technical field of intelligent government services, specifically referring to a personalized government service recommendation method and system based on artificial intelligence. The present invention obtains original personalized recommendation data through data collection; adopts data preprocessing methods such as data cleaning, data encoding, data normalization, and dataset segmentation; uses a clustering model based on an optimization algorithm for user classification, which can effectively group complex user data and improve the interpretability and structuring degree of the data; and uses an improved transformer model combined with a bidirectional gated recurrent unit for government service recommendation, which can effectively extract the temporal features of user information and government service data, capture long-term dependencies, and improve the accuracy of personalized recommendation.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent government services, and specifically refers to a personalized government service recommendation method and system based on artificial intelligence. Background Art

[0002] Personalized government service recommendation utilizes technologies such as artificial intelligence, big data analysis, and machine learning to accurately recommend suitable government services to users according to their behaviors, needs, and preferences; this technology improves the efficiency and accuracy of government services by predicting demands and adjusting recommended content in real time, and significantly enhances the convenience and personalization level of government services.

[0003] However, traditional government service recommendation methods have technical problems that in the face of diverse and complexly distributed user information data, it is difficult to capture the potential patterns among users, resulting in low accuracy and personalization degree of recommendations. At the same time, high-dimensional user demand data is prone to information redundancy and increased noise; traditional government service recommendation methods have technical problems that the traditional recommendation algorithms used are difficult to handle complex user behavior patterns and multi-dimensional features, lack in-depth modeling of temporality, context information, and user behavior details, resulting in the inability of the recommendation effect to dynamically adapt to changes in user demands, and low computational efficiency. Summary of the Invention

[0004] In view of the above situation, to overcome the defects of the prior art, the present invention provides a personalized government service recommendation method and system based on artificial intelligence. Aiming at the technical problems that traditional government service recommendation methods have difficulty in capturing the potential patterns among users in the face of diverse and complexly distributed user information data, resulting in low accuracy and personalization degree of recommendations, and at the same time, high-dimensional user demand data is prone to information redundancy and increased noise, this solution creatively adopts a clustering model based on an optimization algorithm for user classification, which can effectively group complex user data, improve the interpretability and structuredness of the data, reduce the redundancy of input data, so that the deep learning model can more efficiently learn the personalized needs and preferences of users, and improve the accuracy and precision of the recommendation results; aiming at the technical problems that traditional government service recommendation methods have difficulty in handling complex user behavior patterns and multi-dimensional features with the traditional recommendation algorithms used, lack in-depth modeling of temporality, context information, and user behavior details, resulting in the inability of the recommendation effect to dynamically adapt to changes in user demands, and low computational efficiency, this solution creatively adopts an improved transformer model combined with a bidirectional gated recurrent unit for government service recommendation, which can effectively extract the temporal features of user information and government service data, capture long-term dependencies, and improve the accuracy and precision of personalized recommendations.

[0005] The technical solution adopted by the present invention is as follows: The personalized government service recommendation method based on artificial intelligence provided by the present invention includes the following steps:

[0006] Step S1: Data collection;

[0007] Step S2: Data preprocessing;

[0008] Step S3: Construction of user classification model;

[0009] Step S4: Construction of service recommendation model;

[0010] Step S5: Personalized service recommendation.

[0011] Further, in step S1, the data collection is used to collect the data required for personalized government service recommendation. Specifically, through data collection, an original personalized recommendation data set is obtained;

[0012] The original personalized recommendation data set specifically includes a historical recommendation original data set and a current recommendation original data set. The historical recommendation original data set and the current recommendation original data set both specifically include user basic information data, user behavior data, and government service data. The historical recommendation original data set further includes historical service record data, historical user preference data, and government service type label data. The user basic information data specifically includes user name, user gender, user age, user occupation, user family status, user income status, and user address. The user behavior data specifically includes user browsing record data and user government interaction data. The government service data specifically includes government service catalog data and government update data.

[0013] Further, in step S2, the data preprocessing is used to preprocess the collected original data, and specifically includes the following steps:

[0014] Step S21: Data cleaning, which is used to clean the original data. Specifically, the missing values and duplicate values in the historical recommendation original data set and the current recommendation original data set are removed to obtain a historical roughly processed data set and a current roughly processed data set;

[0015] Step S22: Data encoding, which is used to encode the roughly processed data. Specifically, one-hot encoding is used to encode the historical roughly processed data set and the current roughly processed data set to obtain a historical encoded data set and a current encoded data set;

[0016] Step S23: Data normalization, which is used to normalize the encoded data. Specifically, the minimum-maximum normalization method is used to normalize the historical encoded data set and the current encoded data set to obtain a historical normalized data set and a preliminary data set to be processed;

[0017] Step S24: Dataset splitting, which is used to split the dataset. Specifically, the historical normalized dataset is split to obtain a preliminary recommendation training set and a recommendation test set.

[0018] Step S25: Perform preprocessing. Specifically, through the data cleaning, the data encoding, and the data normalization, the current recommended original dataset is preprocessed to obtain a preliminary dataset to be processed. Through the data cleaning, the data encoding, the data normalization, and the dataset splitting, the historical recommended original dataset is preprocessed to obtain a preliminary recommendation training set and a recommendation test set.

[0019] Further, in step S3, the user classification model construction is used to construct the model required for user classification. Specifically, by constructing a clustering model based on an optimization algorithm, a user classification model is obtained. The user classification model construction specifically includes the following steps:

[0020] Step S31: Generate an initial solution. Specifically, an initial solution of the optimization algorithm is generated based on the tent chaotic map, and the formula used is as follows:

[0021] ;

[0022] In the formula, represents the initial value of the tent chaotic map, which is a random value within the range [0, 1], r represents the chaos control parameter, represents the chaotic value of the (j + 1)-th dimension, represents the chaotic value of the j-th dimension, represents the position of the i-th individual unit in the j-th dimension, represents the upper limit of the j-th dimension, represents the lower limit of the j-th dimension. The individual unit is used to represent the combination of clustering centers to be optimized, and the position of its dimension represents the dimension of the clustering center;

[0023] Step S32: Determine the fitness function. Specifically, the within-cluster sum of squares function is selected as the fitness function of the individual unit. The formula of the within-cluster sum of squares function is as follows:

[0024] ;

[0025] In the formula, represents the fitness function, represents the i-th individual unit, I represents the total number of clustering centers, represents the k-th data point in the i-th cluster;

[0026] Step S33: Update the position of the individual unit. The steps include:

[0027] Step S331: Update the dynamic evolution factor, and the formula used is as follows:

[0028] ;

[0029] In the formula, represents the dynamic evolution factor at the dt-th iteration, represents a random number within the range [-1, 1], dt represents the current iteration number, represents the maximum iteration number of the optimization algorithm, represents a random number that follows the standard normal distribution;

[0030] Step S332: Calculate the percentage difference between the current position and the best position, and the formula used is as follows:

[0031] ;

[0032] In the formula, represents the percentage difference between the current position and the best position, represents the dimensional mean of the i-th individual unit, J represents the total number of dimensions of the individual unit, represents the position of the j-th dimension of the best position;

[0033] Step S333: Calculate the reduction factor, and the formula used is as follows:

[0034] ;

[0035] In the formula, represents the reduction factor of the position of the i-th individual unit in the j-th dimension, represents the position of a random individual unit in the j-th dimension;

[0036] Step S334: Update the position, and the formula used is as follows:

[0037] ;

[0038] In the formula, represents the position of the i-th individual unit in the j-th dimension at the (dt + 1)-th iteration, represents the exploration control factor, represents a random number within the range [0, 1], represents a random number that follows the standard normal distribution, represents the position of the i-th individual unit in the j-th dimension at the dt-th iteration, represents a random number within the range [0, 1], represents the exploitation control factor, Represents a random number within the range [0, 1];

[0039] Step S34: Iterative update, specifically, iteratively update the optimization algorithm until the iterative termination condition of the optimization algorithm is reached, thereby obtaining the final global optimal position. The final global optimal position is specifically the optimal clustering center combination. The iterative termination condition of the optimization algorithm is specifically reaching the maximum number of iterations of the optimization algorithm or the fitness function value of the individual unit being less than the preset threshold;

[0040] Step S35: Construct a clustering model. Specifically, through the generation of the initial solution, the determination of the fitness function, the update of the individual unit position, and the iterative update, obtain the optimal clustering center combination, and based on the optimal clustering center combination, construct a clustering model to obtain a user classification model;

[0041] Step S36: User classification. Specifically, use the user classification model to perform clustering processing on the user basic information data and user behavior data in the preliminary recommendation training set to obtain historical user category data, and use the historical user category data to replace the user basic information data and user behavior data to obtain a recommendation training set;

[0042] Use the user classification model to perform clustering processing on the user basic information data and user behavior data in the preliminary data set to be processed to obtain current user category data, and use the current user category data to replace the user basic information data and user behavior data to obtain a data set to be processed.

[0043] Furthermore, in step S4, the construction of the service recommendation model is used to construct the model required for government service recommendation. Specifically, construct an improved transformer model combined with a bidirectional gated recurrent unit and use it as the service recommendation model. The improved transformer model combined with a bidirectional gated recurrent unit specifically includes a feature extraction module, a transformer module, a feature fusion module, and an output module;

[0044] The construction of the service recommendation model specifically includes the following steps:

[0045] Step S41: Construction of the feature extraction module. Specifically, construct a bidirectional gated recurrent unit for extracting features from the input data. The formula used is as follows:

[0046] ;

[0047] In the formula, represents the forward hidden state, represents the forward function of the bidirectional gated recurrent unit, represents the input data of the feature extraction module, represents the backward hidden state, Represents the backward function of the bidirectional gated recurrent unit, Represents the output features of the feature extraction module, Represents the concatenation operation;

[0048] Step S42: Transformer module construction, which is used to construct a transformer module for further processing the output features of the feature extraction module. Specifically, a transformer module using a dynamic sparse multi-head self-attention mechanism is constructed. The steps include:

[0049] Step S421: Calculate the query, key, and value for each head. The formulas used are as follows:

[0050] ;

[0051] In the formula, Represents the query of the h-th head, Represents the key of the h-th head, Represents the value of the h-th head, Represents the input features of the dynamic sparse multi-head self-attention mechanism, Represents the query transformation matrix of the h-th head, Represents the key transformation matrix of the h-th head, Represents the value transformation matrix of the h-th head;

[0052] Step S422: Calculate the similarity between each query vector and the key in each head. The formula used is as follows:

[0053] ;

[0054] In the formula, Represents the similarity calculation function, Represents the -th query vector of the h-th head, Represents the maximum value function, Represents the -th key vector of the h-th head, T represents the transpose operation, Represents the dimension of the key vector of the h-th head, Represents the total number of key vectors of the key of the h-th head;

[0055] Step S423: Calculate the dynamic sparse self-attention of each head. The steps include:

[0056] Step S4231: Extract the query vectors. Specifically, sort the query vectors from high to low according to the similarity with the key, extract the first Qn query vectors before sorting to form the sampled queries, and the remaining query vectors form the remaining queries;

[0057] Step S4232: Calculate the self-attention based on the sampled queries. The formula used is as follows:

[0058] ;

[0059] In the formula, represents the self-attention of the h-th head calculated based on the sampled query, represents the softmax function, represents the sampled query of the h-th head;

[0060] Step S4233: Calculate the self-attention based on the remaining queries. The formula used is as follows:

[0061] ;

[0062] In the formula, represents the self-attention of the h-th head calculated based on the remaining queries, represents the total number of value vectors of the values of the h-th head, represents the -th value vector of the h-th head;

[0063] Step S4234: Calculate the dynamic sparse self-attention of the head. The formula used is as follows:

[0064] ;

[0065] In the formula, represents the dynamic sparse self-attention of the h-th head;

[0066] Step S424: Calculate the output of the dynamic sparse multi-head self-attention mechanism. The formula used is as follows:

[0067] ;

[0068] In the formula, represents the output feature of the dynamic sparse multi-head self-attention mechanism, represents the concatenation function, represents the dynamic sparse self-attention of the first head, represents the dynamic sparse self-attention of the second head, represents the self-attention linear transformation weight matrix;

[0069] Step S425: Construct the transformer module. Specifically, design the dynamic sparse multi-head self-attention mechanism by calculating the query, key, and value of each head, calculating the similarity between each query vector and the key in each head, calculating the dynamic sparse self-attention of each head, and calculating the output of the dynamic sparse multi-head self-attention mechanism, and construct the transformer module based on the dynamic sparse multi-head self-attention mechanism;

[0070] Step S43: Construct a feature fusion module for fusing the output features of the feature extraction module and the output features of the transformer module. Specifically, construct a feature fusion module based on the cross-attention mechanism and the gating mechanism. The steps include:

[0071] Step S431: Design the cross-attention mechanism. The formula used is as follows:

[0072] ;

[0073] In the formula, Cq represents the cross-attention query, Ck represents the cross-attention key, Cv represents the cross-attention value, represents the cross-attention query transformation matrix, represents the cross-attention key transformation matrix, represents the cross-attention value transformation matrix, represents the output features of the transformer module, represents the output features of the cross-attention mechanism, represents the dimension of the cross-attention key;

[0074] Step S432: Design the gating mechanism. The steps include:

[0075] Step S4321: Calculate the selection gating value. The formula used is as follows:

[0076] ;

[0077] In the formula, represents the selection gating value, represents the sigmoid function, represents the learnable matrix in the calculation of the selection gating value, represents the bias term in the calculation of the selection gating value, represents element-wise addition;

[0078] Step S4322: Calculate the intensity gating value. The formula used is as follows:

[0079] ;

[0080] In the formula, represents the intensity gating value, represents the hyperbolic tangent function, represents the learnable matrix in the calculation of the intensity gating value, represents the bias term in the calculation of the intensity gating value;

[0081] Step S433: Calculate the output of the feature fusion module. The formula used is as follows:

[0082] ;

[0083] In the formula, Fu represents the fused feature output by the feature fusion module;

[0084] Step S44: Output module construction, which is used to construct an output module that processes the fused feature output by the feature fusion module and obtains the model output. Specifically, an output module based on the softmax classification function is constructed, and the formula used is as follows:

[0085] ;

[0086] In the formula, represents the recommended prediction result output by the model, represents the model output weight, represents the fully connected function, which is used to map the fused feature output by the feature fusion module to the output space, represents the model output bias term;

[0087] Step S45: Construct and train the model. Specifically, based on the construction of the feature extraction module, the construction of the transformer module, the construction of the feature fusion module, and the construction of the output module, an improved transformer model combined with a bidirectional gated recurrent unit is constructed, and the model is trained based on the recommended training set, and the model performance is verified based on the recommended test set. The model loss function selects the cross-entropy loss function to obtain an improved transformer model combined with a bidirectional gated recurrent unit, which is used as the service recommendation model.

[0088] Further, in step S5, the personalized service recommendation specifically uses the service recommendation model to process the to-be-processed data set to obtain personalized service recommendation reference data, and based on the personalized service recommendation reference data, government service recommendations are made to users.

[0089] The personalized government service recommendation system based on artificial intelligence provided by the present invention includes a data collection module, a data preprocessing module, a user classification model construction module, a service recommendation model construction module, and a personalized service recommendation module;

[0090] The data collection module is used for data collection. Through data collection, an original personalized recommendation data set is obtained, and the original personalized recommendation data set is sent to the data preprocessing module;

[0091] The data preprocessing module is used for data preprocessing. Through data preprocessing, a preliminary to-be-processed data set, a preliminary recommended training set, and a recommended test set are obtained, and the preliminary to-be-processed data set and the preliminary recommended training set are sent to the user classification model construction module, and the recommended test set is sent to the service recommendation model construction module;

[0092] The user classification model construction module is used to construct a user classification model. By constructing a clustering model based on an optimization algorithm, a user classification model, a dataset to be processed, and a recommended training set are obtained. The dataset to be processed is sent to the personalized service recommendation module, and the recommended training set is sent to the service recommendation model construction module;

[0093] The service recommendation model construction module is used to construct a service recommendation model. By constructing an improved transformer model combined with a bidirectional gated recurrent unit, a service recommendation model is obtained, and the service recommendation model is sent to the personalized service recommendation module;

[0094] The personalized service recommendation module is used for personalized service recommendation. By using the service recommendation model for personalized service recommendation, personalized service recommendation reference data is obtained.

[0095] The beneficial effects achieved by the present invention using the above solution are as follows:

[0096] (1) Aiming at the technical problems of the traditional government service recommendation method, which is difficult to capture the potential rules among users in the face of diverse and complexly distributed user information data, resulting in low accuracy and personalization of recommendations, and at the same time, the high-dimensional user demand data is prone to information redundancy and noise increase. This solution creatively uses a clustering model based on an optimization algorithm for user classification, which can effectively group complex user data, improve the interpretability and structuredness of the data, reduce the redundancy of input data, so that the deep learning model can more efficiently learn the personalized needs and preferences of users, and improve the accuracy and precision of the recommendation results.

[0097] (2) Aiming at the technical problems of the traditional government service recommendation method, which is difficult to process complex user behavior patterns and multi-dimensional features using traditional recommendation algorithms, lacks in-depth modeling of temporal, contextual information and user behavior details, resulting in the inability of the recommendation effect to dynamically adapt to changes in user needs, and low computational efficiency. This solution creatively uses an improved transformer model combined with a bidirectional gated recurrent unit for government service recommendation, which can effectively extract the temporal features of user information and government service data, capture long-term dependence relationships, and improve the accuracy and precision of personalized recommendations. Description of the Drawings

[0098] Figure 1 It is a schematic flow chart of the personalized government service recommendation method based on artificial intelligence provided by the present invention;

[0099] Figure 2 It is a schematic module diagram of the personalized government service recommendation system based on artificial intelligence provided by the present invention;

[0100] Figure 3It is a schematic flowchart of the data preprocessing in step S2;

[0101] Figure 4 It is a schematic flowchart of the construction of the user classification model in step S3;

[0102] Figure 5 It is a schematic flowchart of the construction of the service recommendation model in step S4.

[0103] The accompanying drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, but do not constitute a limitation to the present invention. Detailed implementation manners

[0104] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0105] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be understood as a limitation to the present invention.

[0106] Embodiment 1, refer to Figure 1 , the technical solution adopted by the present invention is as follows: The personalized government service recommendation method based on artificial intelligence provided by the present invention includes the following steps:

[0107] Step S1: Data collection;

[0108] Step S2: Data preprocessing;

[0109] Step S3: Construction of the user classification model;

[0110] Step S4: Construction of the service recommendation model;

[0111] Step S5: Personalized service recommendation.

[0112] Embodiment 2, refer to Figure 1 and Figure 2 , in step S1, the data collection is used to collect the data required for personalized government service recommendation, specifically, through data collection, the original personalized recommendation data set is obtained;

[0113] The original personalized recommendation dataset specifically includes a historical recommendation original dataset and a current recommendation original dataset. Both the historical recommendation original dataset and the current recommendation original dataset specifically include user basic information data, user behavior data, and government service data. The historical recommendation original dataset further includes historical service record data, historical user preference data, and government service category label data. The user basic information data specifically includes user name, user gender, user age, user occupation, user family status, user income status, and user address. The user behavior data specifically includes user browsing record data and user government interaction data. The government service data specifically includes government service catalog data and government update data.

[0114] Example 3. Refer to Figure 1 、 Figure 2 and Figure 3 Based on the above example, in step S2, the data preprocessing is used to preprocess the collected original data, and specifically includes the following steps:

[0115] Step S21: Data cleaning, which is used to clean the original data, specifically to remove the missing values and duplicate values in the historical recommendation original dataset and the current recommendation original dataset, and obtain a historical roughly processed dataset and a current roughly processed dataset;

[0116] Step S22: Data encoding, which is used to encode the roughly processed data, specifically to use the one-hot encoding method to encode the historical roughly processed dataset and the current roughly processed dataset, and obtain a historical encoded dataset and a current encoded dataset;

[0117] Step S23: Data normalization, which is used to normalize the encoded data, specifically to use the min-max normalization method to normalize the historical encoded dataset and the current encoded dataset, and obtain a historical normalized dataset and a preliminary dataset to be processed;

[0118] Step S24: Dataset splitting, which is used to split the dataset, specifically to split the historical normalized dataset to obtain a preliminary recommendation training set and a recommendation test set;

[0119] Step S25: Perform preprocessing, specifically to preprocess the current recommendation original dataset through the data cleaning, the data encoding, and the data normalization to obtain a preliminary dataset to be processed, and to preprocess the historical recommendation original dataset through the data cleaning, the data encoding, the data normalization, and the dataset splitting to obtain a preliminary recommendation training set and a recommendation test set.

[0120] Example 4. Refer to Figure 1, Figure 2 and Figure 4 , based on the above embodiment, in step S3, the user classification model construction is for constructing a model required for user classification. Specifically, by constructing a clustering model based on an optimization algorithm, a user classification model is obtained. The user classification model construction specifically includes the following steps:

[0121] Step S31: Generate an initial solution. Specifically, generate an initial solution of the optimization algorithm based on the tent chaotic map. The formula used is as follows:

[0122] ;

[0123] In the formula, represents the initial value of the tent chaotic map, which is a random value within the range [0, 1]. r represents the chaos control parameter, represents the chaotic value of the (j + 1)-th dimension, represents the chaotic value of the j-th dimension, represents the position of the i-th individual unit in the j-th dimension, represents the upper limit of the j-th dimension, represents the lower limit of the j-th dimension. The individual unit is used to represent the combination of clustering centers to be optimized, and the position of its dimension represents the dimension of the clustering center;

[0124] Step S32: Determine the fitness function. Specifically, select the within-cluster sum of squared errors function as the fitness function of the individual unit. The formula of the within-cluster sum of squared errors function is as follows:

[0125] ;

[0126] In the formula, represents the fitness function, represents the i-th individual unit, I represents the total number of clustering centers, represents the k-th data point in the i-th cluster;

[0127] Step S33: Update the position of the individual unit. The steps include:

[0128] Step S331: Update the dynamic evolution factor. The formula used is as follows:

[0129] ;

[0130] In the formula, represents the dynamic evolution factor at the dt-th iteration, represents a random number within the range [-1, 1], dt represents the current iteration number, represents the maximum number of iterations of the optimization algorithm, Represents a random number that satisfies the standard normal distribution;

[0131] Step S332: Calculate the percentage difference between the current position and the best position, using the following formula:

[0132] ;

[0133] In the formula, Represents the percentage difference between the current position and the best position, Represents the dimensional mean of the i-th individual unit, J represents the total number of individual unit dimensions, Represents the position of the j-th dimension of the best position;

[0134] Step S333: Calculate the contraction factor, using the following formula:

[0135] ;

[0136] In the formula, Represents the contraction factor of the position of the i-th individual unit in the j-th dimension, Represents the position of the random individual unit in the j-th dimension;

[0137] Step S334: Update the position, using the following formula:

[0138] ;

[0139] In the formula, Represents the position of the i-th individual unit in the j-th dimension at the (dt + 1)-th iteration, Represents the exploration control factor, Represents a random number in the range [0, 1], Represents a random number that satisfies the standard normal distribution, Represents the position of the i-th individual unit in the j-th dimension at the dt-th iteration, Represents a random number in the range [0, 1], Represents the exploitation control factor, Represents a random number in the range [0, 1];

[0140] Step S34: Iteratively update, specifically iteratively update the optimization algorithm until the iteration termination condition of the optimization algorithm is reached, so as to obtain the final global best position. The final global best position is specifically the optimal clustering center combination. The iteration termination condition of the optimization algorithm is specifically reaching the maximum number of iterations of the optimization algorithm or the fitness function value of the individual unit being less than the preset threshold;

[0141] Step S35: Construct a clustering model. Specifically, through the generation of the initial solution, the determination of the fitness function, the update of the individual unit positions, and the iterative update, obtain the optimal combination of clustering centers, and based on the optimal combination of clustering centers, construct a clustering model to obtain a user classification model;

[0142] Step S36: User classification. Specifically, use the user classification model to perform clustering processing on the user basic information data and user behavior data in the preliminary recommendation training set to obtain historical user category data, and use the historical user category data to replace the user basic information data and user behavior data to obtain a recommendation training set;

[0143] Use the user classification model to perform clustering processing on the user basic information data and user behavior data in the preliminary data set to be processed to obtain current user category data, and use the current user category data to replace the user basic information data and user behavior data to obtain a data set to be processed.

[0144] By performing the above operations, for the technical problems existing in the traditional government service recommendation method, such as facing diverse and complexly distributed user information data, it is difficult to capture the potential rules among users, resulting in low accuracy and personalization degree of recommendations, and at the same time, the high-dimensional user demand data is prone to information redundancy and noise increase. This solution creatively uses a clustering model based on an optimization algorithm for user classification, which can effectively group complex user data, improve the interpretability and structuredness of the data, reduce the redundancy of the input data, so that the deep learning model can more efficiently learn the personalized needs and preferences of users, and improve the accuracy and precision of the recommendation results.

[0145] Example 5, refer to Figure 1 、 Figure 2 and Figure 5 In this example, based on the above example, in step S4, the service recommendation model construction is used to construct the model required for government service recommendation. Specifically, an improved transformer model combined with a bidirectional gated recurrent unit is constructed and used as the service recommendation model. The improved transformer model combined with a bidirectional gated recurrent unit specifically includes a feature extraction module, a transformer module, a feature fusion module, and an output module;

[0146] The construction of the service recommendation model specifically includes the following steps:

[0147] Step S41: Construction of the feature extraction module. Specifically, construct a bidirectional gated recurrent unit for extracting features from the input data, and the formula used is as follows:

[0148] ;

[0149] In the formula, Represents the forward hidden state, Represents the forward function of the bidirectional gated recurrent unit, Represents the input data of the feature extraction module, Represents the backward hidden state, Represents the backward function of the bidirectional gated recurrent unit, Represents the output features of the feature extraction module, Represents the concatenation operation;

[0150] Step S42: Transformer module construction, used to construct a transformer module for further processing the output features of the feature extraction module. Specifically, it is to construct a transformer module using a dynamic sparse multi-head self-attention mechanism. The steps include:

[0151] Step S421: Calculate the query, key, and value for each head. The formulas used are as follows:

[0152] ;

[0153] In the formula, Represents the query of the h-th head, Represents the key of the h-th head, Represents the value of the h-th head, Represents the input features of the dynamic sparse multi-head self-attention mechanism, Represents the query transformation matrix of the h-th head, Represents the key transformation matrix of the h-th head, Represents the value transformation matrix of the h-th head;

[0154] Step S422: Calculate the similarity between each query vector and the key in each head. The formula used is as follows:

[0155] ;

[0156] In the formula, Represents the similarity calculation function, Represents the -th query vector of the h-th head, Represents the maximum value function, Represents the -th key vector of the h-th head, T represents the transpose operation, Represents the dimension of the key vector of the h-th head, Represents the total number of key vectors of the key of the h-th head;

[0157] Step S423: Calculate the dynamic sparse self-attention of each head. The steps include:

[0158] Step S4231: Extract query vectors. Specifically, sort the query vectors in descending order of similarity to the keys, extract the top Qn query vectors before sorting to form a sampled query, and the remaining query vectors form the remaining query;

[0159] Step S4232: Calculate self-attention based on the sampled query. The formula used is as follows:

[0160] ;

[0161] In the formula, represents the self-attention of the h-th head calculated based on the sampled query, represents the softmax function, represents the sampled query of the h-th head;

[0162] Step S4233: Calculate self-attention based on the remaining query. The formula used is as follows:

[0163] ;

[0164] In the formula, represents the self-attention of the h-th head calculated based on the remaining query, represents the total number of value vectors of the values of the h-th head, represents the -th value vector of the h-th head;

[0165] Step S4234: Calculate the dynamic sparse self-attention of the head. The formula used is as follows:

[0166] ;

[0167] In the formula, represents the dynamic sparse self-attention of the h-th head;

[0168] Step S424: Calculate the output of the dynamic sparse multi-head self-attention mechanism. The formula used is as follows:

[0169] ;

[0170] In the formula, represents the output feature of the dynamic sparse multi-head self-attention mechanism, represents the concatenation function, represents the dynamic sparse self-attention of the first head, represents the dynamic sparse self-attention of the second head, represents the weight matrix of the linear transformation of self-attention;

[0171] Step S425: Construct a transformer module. Specifically, design the dynamic sparse multi-head self-attention mechanism through calculating the queries, keys, and values of each head, calculating the similarity between each query vector and the key in each head, calculating the dynamic sparse self-attention of each head, and calculating the output of the dynamic sparse multi-head self-attention mechanism. Then, construct a transformer module based on the dynamic sparse multi-head self-attention mechanism;

[0172] Step S43: Construct a feature fusion module, which is used to construct a feature fusion module that fuses the output features of the feature extraction module and the output features of the transformer module. Specifically, construct the feature fusion module based on the cross-attention mechanism and the gating mechanism. The steps include:

[0173] Step S431: Design the cross-attention mechanism, and the formula used is as follows:

[0174] ;

[0175] In the formula, Cq represents the cross-attention query, Ck represents the cross-attention key, Cv represents the cross-attention value, represents the cross-attention query transformation matrix, represents the cross-attention key transformation matrix, represents the cross-attention value transformation matrix, represents the output features of the transformer module, represents the output features of the cross-attention mechanism, represents the dimension of the cross-attention key;

[0176] Step S432: Design the gating mechanism. The steps include:

[0177] Step S4321: Calculate the selection gating value, and the formula used is as follows:

[0178] ;

[0179] In the formula, represents the selection gating value, represents the sigmoid function, represents the learnable matrix in the calculation of the selection gating value, represents the bias term in the calculation of the selection gating value, represents the element-wise addition;

[0180] Step S4322: Calculate the intensity gating value, and the formula used is as follows:

[0181] ;

[0182] In the formula, represents the intensity gating value, represents the hyperbolic tangent function, represents the learnable matrix in the calculation of the intensity gating value, represents the bias term in the calculation of the intensity gating value;

[0183] Step S433: Calculate the output of the feature fusion module, and the formula used is as follows:

[0184] ;

[0185] In the formula, Fu represents the fused feature output by the feature fusion module;

[0186] Step S44: Construct the output module, which is used to construct the output module that processes the fused feature output by the feature fusion module and obtains the model output. Specifically, an output module based on the softmax classification function is constructed, and the formula used is as follows:

[0187] ;

[0188] In the formula, represents the recommended prediction result output by the model, represents the model output weight, represents the fully connected function, which is used to map the fused feature output by the feature fusion module to the output space, represents the model output bias term;

[0189] Step S45: Construct and train the model. Specifically, based on the construction of the feature extraction module, the transformer module, the feature fusion module, and the output module, an improved transformer model combined with a bidirectional gated recurrent unit is constructed, and the model is trained based on the recommended training set, and the model performance is verified based on the recommended test set. The model loss function selects the cross-entropy loss function to obtain an improved transformer model combined with a bidirectional gated recurrent unit and use it as the service recommendation model.

[0190] By performing the above operations, for the technical problems existing in the traditional government service recommendation method, such as the traditional recommendation algorithm used is difficult to handle complex user behavior patterns and multi-dimensional features, lacking in-depth modeling of temporal characteristics, context information, and user behavior details, resulting in the recommendation effect being unable to dynamically adapt to changes in user needs and having low computational efficiency, this solution creatively uses an improved transformer model combined with a bidirectional gated recurrent unit for government service recommendation, which can effectively extract the temporal features of user information and government service data, capture long-term dependence relationships, and improve the accuracy and precision of personalized recommendations.

[0191] Example six, refer to Figure 1 and Figure 2, based on the above embodiment, in step S5, the personalized service recommendation specifically involves using the service recommendation model to process the dataset to be processed, obtaining personalized service recommendation reference data, and making government service recommendations to users based on the personalized service recommendation reference data.

[0192] Embodiment VII, refer to Figure 1 and Figure 2 , based on the above embodiment, the personalized government service recommendation system based on artificial intelligence provided by the present invention includes a data acquisition module, a data preprocessing module, a user classification model construction module, a service recommendation model construction module, and a personalized service recommendation module;

[0193] The data acquisition module is used for data acquisition. Through data acquisition, an original personalized recommendation dataset is obtained, and the original personalized recommendation dataset is sent to the data preprocessing module;

[0194] The data preprocessing module is used for data preprocessing. Through data preprocessing, a preliminary dataset to be processed, a preliminary recommendation training set, and a recommendation test set are obtained. The preliminary dataset to be processed and the preliminary recommendation training set are sent to the user classification model construction module, and the recommendation test set is sent to the service recommendation model construction module;

[0195] The user classification model construction module is used for constructing a user classification model. By constructing a clustering model based on an optimization algorithm, a user classification model, a dataset to be processed, and a recommendation training set are obtained. The dataset to be processed is sent to the personalized service recommendation module, and the recommendation training set is sent to the service recommendation model construction module;

[0196] The service recommendation model construction module is used for constructing a service recommendation model. By constructing an improved transformer model combined with a bidirectional gated recurrent unit, a service recommendation model is obtained, and the service recommendation model is sent to the personalized service recommendation module;

[0197] The personalized service recommendation module is used for personalized service recommendation. By using the service recommendation model for personalized service recommendation, personalized service recommendation reference data is obtained.

[0198] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device.

[0199] Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention.

[0200] The above description of the present invention and its embodiments is not restrictive. What is shown in the drawings is only one of the embodiments of the present invention, and the actual structure is not limited thereto. In general, if those of ordinary skill in the art are inspired by it and design similar structural modes and embodiments without creative efforts without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. A personalized government service recommendation method based on artificial intelligence, characterized in that: The method includes the following steps: Step S1: Data collection. Through data collection, an original personalized recommendation data set is obtained. The original personalized recommendation data set specifically includes a historical recommendation original data set and a current recommendation original data set. The historical recommendation original data set and the current recommendation original data set both specifically include user basic information data, user behavior data, and government service data. The historical recommendation original data set also includes historical service record data, historical user preference data, and government service type label data; Step S2: Data preprocessing. The collected original data is preprocessed to obtain a preliminary data set to be processed, a preliminary recommendation training set, and a recommendation test set; Step S3: User classification model construction. It is used to construct a model required for user classification. Specifically, a clustering model based on an optimization algorithm is constructed to obtain a user classification model, and the user classification model is used to process the preliminary data set to be processed and the preliminary recommendation training set to obtain a data set to be processed and a recommendation training set; Step S4: Service recommendation model construction. It is used to construct a model required for government service recommendation. Specifically, an improved transformer model combined with a bidirectional gated recurrent unit is constructed and used as the service recommendation model. The improved transformer model combined with the bidirectional gated recurrent unit specifically includes a feature extraction module, a transformer module, a feature fusion module, and an output module; Step S5: Personalized service recommendation. Specifically, personalized service recommendation is performed through the service recommendation model to obtain personalized service recommendation reference data; The construction of the service recommendation model specifically includes the following steps: Step S41: Feature extraction module construction. Specifically, a bidirectional gated recurrent unit for extracting features from input data is constructed, and the formula used is as follows: ； In the formula, represents the forward hidden state, represents the forward function of the bidirectional gated recurrent unit, represents the input data of the feature extraction module, represents the backward hidden state, represents the backward function of the bidirectional gated recurrent unit, represents the output feature of the feature extraction module, represents the concatenation operation; Step S42: Transformer module construction. It is used to construct a transformer module for further processing the output features of the feature extraction module. Specifically, a transformer module using a dynamic sparse multi-head self-attention mechanism is constructed. The steps include: Step S421: Calculate the query, key, and value of each head, and the formula used is as follows: ； In the formula, represents the query of the h-th head, represents the key of the h-th head, represents the value of the h-th head, represents the input feature of the dynamic sparse multi-head self-attention mechanism, represents the query transformation matrix of the h-th head, represents the key transformation matrix of the h-th head, represents the value transformation matrix of the h-th head; Step S422: Calculate the similarity between each query vector and the key in each head, and the formula used is as follows: ； In the formula, represents the similarity calculation function, represents the th query vector of the hth head, represents the maximum value function, represents the th key vector of the hth head, T represents the transpose operation, represents the dimension of the key vector of the hth head, represents the total number of key vectors of the keys of the hth head; Step S423: Calculate the dynamic sparse self-attention of each head. The steps include: Step S4231: Extract query vectors. Specifically, the query vectors are sorted from high to low according to the similarity with the key, and the first Qn query vectors before sorting are extracted to form a sampled query, and the remaining query vectors form the remaining query; Step S4232: Calculate self-attention based on the sampled query, and the formula used is as follows: ； In the formula, represents the self-attention of the h-th head calculated based on the sampled query, represents the softmax function, represents the sampled query of the h-th head; Step S4233: Calculate self-attention based on the remaining query, and the formula used is as follows: ； In the formula, represents the self-attention of the h-th head calculated based on the remaining query, represents the total number of value vectors of the values of the h-th head, represents the -th value vector of the h-th head; Step S4234: Calculate the dynamic sparse self-attention of the head, and the formula used is as follows: ； In the formula, represents the dynamic sparse self-attention of the h-th head; Step S424: Calculate the output of the dynamic sparse multi-head self-attention mechanism, and the formula used is as follows: ； In the formula, represents the output feature of the dynamic sparse multi-head self-attention mechanism, represents the connection function, represents the dynamic sparse self-attention of the first head, represents the dynamic sparse self-attention of the second head, represents the self-attention linear transformation weight matrix; Step S425: Construct a transformer module. Specifically, design the dynamic sparse multi-head self-attention mechanism through calculating the query, key, and value of each head, calculating the similarity between each query vector and the key in each head, calculating the dynamic sparse self-attention of each head, and calculating the output of the dynamic sparse multi-head self-attention mechanism, and construct a transformer module based on the dynamic sparse multi-head self-attention mechanism; Step S43: Construct a feature fusion module, which is used to construct a feature fusion module that fuses the output features of the feature extraction module and the output features of the transformer module. Specifically, construct a feature fusion module based on the cross-attention mechanism and the gating mechanism. The steps include: Step S431: Design the cross-attention mechanism, and the formula used is as follows: ； Wherein, Cq represents the cross-attention query, Ck represents the cross-attention key, and Cv represents the cross-attention value. represents the cross-attention query transformation matrix. represents the cross-attention key transformation matrix. represents the cross-attention value transformation matrix. represents the output feature of the transformer module. represents the output feature of the cross-attention mechanism. represents the dimension of the cross-attention key. Step S432: Design the gating mechanism. The steps include: Step S4321: Calculate the selection gating value, and the formula used is as follows: ； In the formula, represents the selection gating value, represents the sigmoid function, represents the learnable matrix in the selection gating value calculation, represents the bias term in the selection gating value calculation, represents element-wise addition; Step S4322: Calculate the intensity gating value, and the formula used is as follows: ； In the formula, represents the intensity gating value, represents the hyperbolic tangent function, represents the learnable matrix in the calculation of the intensity gating value, represents the bias term in the calculation of the intensity gating value; Step S433: Calculate the output of the feature fusion module, and the formula used is as follows: ； In the formula, Fu represents the fused feature output by the feature fusion module; Step S44: Construct an output module, which is used to construct an output module that processes the fused feature output by the feature fusion module and obtains the model output. Specifically, construct an output module based on the softmax classification function, and the formula used is as follows: ； Wherein, represents the recommended prediction result output by the model, represents the model output weight, represents the fully connected function, which is used to map the fused features output by the feature fusion module to the output space, represents the model output bias term; Step S45: Construct and train the model. Specifically, construct an improved transformer model combined with a bidirectional gated recurrent unit based on the construction of the feature extraction module, the construction of the transformer module, the construction of the feature fusion module, and the construction of the output module, train the model based on the recommended training set, verify the model performance based on the recommended test set, select the cross-entropy loss function as the model loss function, and obtain an improved transformer model combined with a bidirectional gated recurrent unit, which is used as a service recommendation model.

2. The personalized government affairs service recommendation method based on artificial intelligence according to claim 1, wherein: In step S3, the construction of the user classification model specifically includes the following steps: Step S31: Generate an initial solution. Specifically, generate an initial solution of the optimization algorithm based on the tent chaotic map, and the formula used is as follows: ； In the formula, represents the initial value of the tent chaos map, which is a random value within the range [0, 1], and r represents the chaos control parameter. represents the chaos value of the (j + 1)-th dimension. represents the chaos value of the j-th dimension. represents the position of the i-th individual unit in the j-th dimension. represents the upper limit of the j-th dimension. represents the lower limit of the j-th dimension. The individual unit is used to represent the combination of clustering centers to be optimized, and the position of its dimension represents the dimension of the clustering center. Step S32: Determine the fitness function. Specifically, select the within-cluster sum of squares error function as the individual unit fitness function. The within-cluster sum of squares error function is as follows: ； In the formula, represents the fitness function, represents the i-th individual unit, and I represents the total number of clustering centers, represents the k-th data point in the i-th cluster; Step S33: Update the position of the individual unit. The steps include: Step S331: Update the dynamic evolution factor, and the formula used is as follows: ； In the formula, represents the dynamic evolution factor at the dt-th iteration, represents a random number within the range [-1, 1], dt represents the current iteration number, represents the maximum number of iterations of the optimization algorithm, represents a random number that satisfies the standard normal distribution; Step S332: Calculate the percentage difference between the current position and the best position, and the formula used is as follows: ； wherein, represents the percentage difference between the current position and the optimal position, represents the dimensional mean of the i-th individual unit, and J represents the total number of dimensions of the individual unit, represents the position of the j-th dimension of the optimal position; Step S333: Calculate the reduction factor, and the formula used is as follows: ； wherein, denotes the reduction factor of the position of the i-th individual unit in the j-th dimension, denotes the position of the random individual unit in the j-th dimension; Step S334: Update the position, and the formula used is as follows: ； Wherein, represents the position of the i-th individual unit in the j-th dimension at the (dt + 1)-th iteration, represents the exploration control factor, represents a random number within the range [0, 1], represents a random number that follows the standard normal distribution, represents the position of the i-th individual unit in the j-th dimension at the dt-th iteration, represents a random number within the range [0, 1], represents the exploitation control factor, represents a random number within the range [0, 1]; Step S34: Iterative update. Specifically, iteratively update the optimization algorithm until the iterative termination condition of the optimization algorithm is reached, so as to obtain the final global best position. The final global best position is specifically the optimal clustering center combination. The iterative termination condition of the optimization algorithm is specifically to reach the maximum number of iterations of the optimization algorithm or the value of the individual unit fitness function is less than the preset threshold; Step S35: Construct a clustering model. Specifically, through the generation of the initial solution, the determination of the fitness function, the update of the individual unit positions, and the iterative update, obtain the optimal combination of clustering centers, and based on the optimal combination of clustering centers, construct a clustering model to obtain a user classification model; Step S36: User classification. Specifically, use the user classification model to perform clustering processing on the user basic information data and user behavior data in the preliminary recommendation training set to obtain historical user category data, and use the historical user category data to replace the user basic information data and user behavior data to obtain a recommendation training set; Use the user classification model to perform clustering processing on the user basic information data and user behavior data in the preliminary data set to be processed to obtain current user category data, and use the current user category data to replace the user basic information data and user behavior data to obtain a data set to be processed.

3. The personalized government affairs service recommendation method based on artificial intelligence according to claim 1, characterized in that: In step S1, the data collection is used to collect the data required for personalized government service recommendations. Specifically, through data collection, an original personalized recommendation data set is obtained; The original personalized recommendation data set specifically includes a historical recommendation original data set and a current recommendation original data set. The historical recommendation original data set and the current recommendation original data set both specifically include user basic information data, user behavior data, and government service data. The historical recommendation original data set also includes historical service record data, historical user preference data, and government service type label data. The user basic information data specifically includes user name, user gender, user age, user occupation, user family status, user income status, and user address. The user behavior data specifically includes user browsing record data and user government interaction data. The government service data specifically includes government service catalog data and government update data.

4. The personalized government affairs service recommendation method based on artificial intelligence according to claim 1, characterized in that: In step S2, the data preprocessing is used to preprocess the collected original data, and specifically includes the following steps: Step S21: Data cleaning is used to clean the original data. Specifically, remove the missing values and duplicate values in the historical recommendation original data set and the current recommendation original data set to obtain a historical roughly processed data set and a current roughly processed data set; Step S22: Data encoding is used to encode the roughly processed data. Specifically, use the one-hot encoding method to perform data encoding on the historical roughly processed data set and the current roughly processed data set to obtain a historical encoded data set and a current encoded data set; Step S23: Data normalization is used to perform data normalization on the encoded data. Specifically, use the min-max normalization method to perform data normalization on the historical encoded data set and the current encoded data set to obtain a historical normalized data set and a preliminary data set to be processed; Step S24: Data set splitting is used to split the data set. Specifically, perform data set splitting on the historical normalized data set to obtain a preliminary recommendation training set and a recommendation test set; Step S25: Perform preprocessing. Specifically, preprocess the current recommended original dataset through the data cleaning, the data encoding, and the data normalization to obtain a preliminary dataset to be processed. Preprocess the historical recommended original dataset through the data cleaning, the data encoding, the data normalization, and the dataset splitting to obtain a preliminary recommendation training set and a recommendation test set.

5. The personalized government affairs service recommendation method based on artificial intelligence according to claim 1, characterized in that: In step S5, the personalized service recommendation specifically uses the service recommendation model to process the dataset to be processed, obtain personalized service recommendation reference data, and perform government service recommendation for users based on the personalized service recommendation reference data.

6. An artificial intelligence-based personalized government affairs service recommendation system for implementing the artificial intelligence-based personalized government affairs service recommendation method according to any one of claims 1-5, characterized in that: It includes a data acquisition module, a data preprocessing module, a user classification model construction module, a service recommendation model construction module, and a personalized service recommendation module.

7. The personalized government affairs service recommendation system based on artificial intelligence according to claim 6, characterized in that: The data acquisition module is used for data acquisition. Through data acquisition, an original personalized recommendation dataset is obtained and sent to the data preprocessing module. The data preprocessing module is used for data preprocessing. Through data preprocessing, a preliminary dataset to be processed, a preliminary recommendation training set, and a recommendation test set are obtained. The preliminary dataset to be processed and the preliminary recommendation training set are sent to the user classification model construction module, and the recommendation test set is sent to the service recommendation model construction module. The user classification model construction module is used for constructing a user classification model. By constructing a clustering model based on an optimization algorithm, a user classification model, a dataset to be processed, and a recommendation training set are obtained. The dataset to be processed is sent to the personalized service recommendation module, and the recommendation training set is sent to the service recommendation model construction module. The service recommendation model construction module is used for constructing a service recommendation model. By constructing an improved transformer model combined with a bidirectional gated recurrent unit, a service recommendation model is obtained and sent to the personalized service recommendation module. The personalized service recommendation module is used for personalized service recommendation. By using the service recommendation model for personalized service recommendation, personalized service recommendation reference data is obtained.

Images