Dynamic big data-oriented consumer online shopping preference prediction system

Abstract

The invention discloses a dynamic big data-oriented consumer online shopping preference prediction system. The system comprises a database module, a prediction module, a personalized recommendation module and a visualization module, the database module adopts a dynamic big data technology to collect and store all shopping preference behavior data of consumers; the prediction module preprocesses and classifies the data based on a machine learning algorithm, generates shopping features, and obtains an analysis result through a neural network prediction model; the personalized recommendation module combines the commodity information and the analysis result, resorts the commodities, and generates a personalized recommendation list; the visualization module displays the recommendation list and identifies and stores the subsequent operation of the consumer to a database for subsequent prediction; the dynamic big data technology is adopted, the shopping preference behavior data of the consumers can be collected and stored in real time or almost in real time, and the system can more accurately predict the shopping preferences of the consumers through the combination of the neural network and the machine learning algorithm, so that the recommendation accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a consumer online shopping market prediction system, and more particularly to a consumer online shopping preference prediction system for dynamic big data. Background Art

[0002] Online shopping has become one of the indispensable lifestyles of modern people. It is deeply loved by consumers for its convenience, diversity, price advantages and other characteristics. However, while enjoying the convenience brought by online shopping, consumers also need to pay attention to protecting their personal information and payment security, and choose a regular platform for shopping. With the continuous progress of technology and the continuous change of consumer needs, online shopping will continue to maintain a rapid development trend, bringing people a more convenient, rich and personalized shopping experience.

[0003] In today's environment where there are a wide variety of commodities and the competition among e-commerce platforms is white-hot, how to effectively mine potential consumers in the massive user data and accurately predict their purchase intentions in order to implement precise marketing strategies is crucial for enhancing the competitiveness of enterprises and optimizing the user experience.

[0004] Existing systems or methods for predicting consumer shopping preferences, due to the large amount and complexity of data, traditional data collection and processing methods cannot obtain the latest shopping preference data of consumers in real time, resulting in inaccurate recommendation results. Moreover, existing consumer shopping preference prediction systems or methods cannot fully consider the personalized behaviors and preferences of consumers, resulting in a deviation between the recommended commodities and the actual needs of consumers.

[0005] Therefore, there is an urgent need for a consumer online shopping preference prediction system for dynamic big data to solve the technical problems existing in the above-mentioned prior art. Summary of the Invention

[0006] The present invention overcomes the deficiencies of the prior art and provides a consumer online shopping preference prediction system for dynamic big data.

[0007] To achieve the above object, the technical solution adopted by the present invention is: a consumer online shopping preference prediction system for dynamic big data, including: a database module, a prediction module, a personalized recommendation module, and a visualization module;

[0008] The database module uses dynamic big data technology to collect and store all shopping preference behaviors from consumers; including: historical shopping data, historical browsing data, real-time search data, real-time browsing data, and shopping completion cycle data of consumers;

[0009] The prediction module preprocesses and classifies all shopping preference behaviors stored in the database module based on a machine learning algorithm, sorts out the shopping habits of consumers' personalized behaviors, and generates shopping features; the prediction model builds a prediction model based on a neural network, takes all the shopping preference behaviors and shopping features as inputs into the prediction model, and obtains an analysis result;

[0010] The personalized recommendation module reorders the products based on all the shopping preference behaviors, combines the product information with the analysis result, and generates a personalized recommendation list;

[0011] The visualization module displays the recommendation list to consumers, identifies and saves the subsequent operations of consumers into the database module for subsequent prediction.

[0012] In a preferred embodiment of the present invention, a prediction threshold is set in the prediction model. When the analysis result is greater than or equal to the prediction threshold, the prediction model outputs the analysis result as the prediction result. When the analysis result is less than the prediction threshold, the sorting process of the shopping habits is iteratively optimized to obtain new shopping features, and the new shopping features and all the shopping preference behaviors are input into the prediction model again for analysis until the analysis result is greater than or equal to the prediction threshold, at which point the iterative optimization is stopped and the analysis result is output.

[0013] In a preferred embodiment of the present invention, the prediction module cleans, de-duplicates, and processes missing values of all the shopping preference behavior data in the database module, extracts the shopping habits of consumers' shopping preference behaviors, and performs feature transformation, standardization, and normalization processing on the shopping habits to obtain the shopping features; among them, the feature transformation process of the shopping habits is represented by the formula where x′ represents the feature value after feature transformation; x represents the shopping habit, that is, the original feature value; μ represents the mean of the shopping habit x; σ represents the standard deviation of the shopping habit x.

[0014] In a preferred embodiment of the present invention, the prediction model is built based on the neural network, and the prediction model includes an input layer, a hidden layer, and an output layer;

[0015] The input layer combines all the shopping preference behaviors and shopping features to obtain an input feature vector set X = [x′, D]; in the feature vector set, D represents the feature vector set of all the shopping preference behaviors of consumers after preprocessing;

[0016] The hidden layer is associated with the input layer through linear transformation and non-linear activation functions, and extracts high-level features from the input data through layer-by-layer transmission; among them, the extraction process of the high-level features is represented by the formula ; in the formula, Z (i) represents the result value after linear transformation of the neurons in the i-th layer; W (i) represents the weight matrix of the neurons in the i-th layer; X (i) represents the i-th feature vector value in the input feature vector set X; a (i-1) represents the activation value of the previous layer of neurons received by each layer of neurons; b (i) represents the bias vector of the neurons in the i-th layer; a (i) represents the activation value of the neurons in the i-th layer; f (i) represents the activation function of the neurons in the i-th layer;

[0017] The output layer is associated with the hidden layer through linear transformation, maps the high-level features to the output space through linear transformation, and obtains the analysis result; among them, the calculation formula of the analysis result is: In the formula, Z (I) represents the result value after linear transformation of the neurons in the last layer; W (I) represents the weight matrix of the neurons in the last layer; X (I) represents the last feature vector value in the input feature vector set X; a (I-1) represents the activation value of the previous layer of neurons received by the neurons in the last layer; b (I) represents the bias vector of the neurons in the last layer; f (I) represents the activation function of the neurons in the last layer; Y represents the analysis result.

[0018] In a preferred embodiment of the present invention, the setting of the prediction threshold includes the following steps:

[0019] S1. According to all the shopping preference behaviors of consumers, train the prediction model using machine learning algorithms, and evaluate and optimize the prediction model through cross-validation methods;

[0020] S2. According to the analysis result, initially set the prediction threshold as the value A, collect the feedback of consumers, and dynamically adjust the value A through the F1 score to obtain the value B;

[0021] S3. Real-time monitor the comparison result between the analysis result of the prediction model and the value B, judge the accuracy rate, and adjust the value B according to the change of the accuracy rate.

[0022] In a preferred embodiment of the present invention, the process of the personalized recommendation module generating a personalized recommendation list includes the following steps:

[0023] A1. Based on the analysis results of all consumers' shopping preference behaviors and the prediction module, conduct a preliminary ranking of the products.

[0024] A2. Considering the popularity indicators of the real-time search volume, click volume, and purchase volume of the products, dynamically adjust the preliminary ranking results.

[0025] A3. According to the personalized shopping preferences of consumers, perform weighted processing on the products.

[0026] A4. Considering the recent shopping behaviors of consumers, fine-tune the recommendation list to finally obtain a personalized recommendation list.

[0027] In a preferred embodiment of the present invention, the visualization module is built-in with an instant feedback mechanism, through which consumers can evaluate the analysis results, and the prediction module iteratively optimizes the analysis results according to the evaluation results.

[0028] In a preferred embodiment of the present invention, the database module further includes a data encryption and security processing unit, which is used to encrypt the collected and stored consumers' shopping preference behavior data, and at the same time set access permission control to only allow authorized system components to access the data.

[0029] In a preferred embodiment of the present invention, the historical shopping data includes: types of purchased products, quantity, price, and purchase time; the historical browsing data includes: browsed product pages, stay time, and browsing times; the search data includes: search keywords, search time, and click situation of search results; the real-time browsing data includes: currently browsed products, page stay time, and browsing path; the shopping completion cycle data includes: time required from browsing to purchase and purchase frequency.

[0030] In a preferred embodiment of the present invention, the prediction module evaluates the prediction performance of the prediction model by using the cross-entropy loss function. Among them, the formula of the cross-entropy loss function is:

[0031]

[0032] In the formula, N represents the number of all shopping preference behaviors; y i represents the true label of the i-th shopping preference behavior; represents the predicted probability of the i-th shopping preference behavior;

[0033] At the same time, calculate the gradient of the cross-entropy loss function with respect to the parameters of the prediction model through the backpropagation algorithm, and then use the gradient descent method to update the parameters of the prediction model; among them, the update formula of the gradient descent method is: Wherein, θ represents the prediction model parameter; θ' represents the updated prediction model parameter; η represents the learning rate; represents the gradient of the cross-entropy loss function with respect to the prediction model parameter.

[0034] The present invention solves the defects existing in the background art and has the following beneficial effects:

[0035] (1) By adopting the dynamic big data technology, it can collect and store the shopping preference behavior data of consumers in real time or nearly in real time, and through the combination of neural network and machine learning algorithms, the system can more accurately predict the shopping preferences of consumers, thereby improving the accuracy of recommendations.

[0036] (2) The system can generate personalized recommendation lists according to the personalized behaviors and preferences of consumers, enhancing the user experience and satisfaction.

[0037] (3) A prediction model is established based on the neural network, and high-level features are extracted through the neural network structure to achieve accurate prediction of consumers' shopping preferences. In addition, by setting a prediction threshold and an iterative optimization mechanism, the accuracy of the prediction is further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings;

[0039] Figure 1 is the system diagram of the preferred embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all 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 protection scope of the present invention.

[0041] Many specific details are set forth in the following description in order to provide a thorough understanding of the present invention, but the present invention can also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0042] Such as Figure 1As shown in the figure, a consumer online shopping preference prediction system for dynamic big data includes: a database module, a prediction module, a personalized recommendation module, and a visualization module;

[0043] The database module adopts dynamic big data technology to collect and store all shopping preference behaviors of consumers; including: historical shopping data, historical browsing data, real-time search data, real-time browsing data, and shopping completion cycle data of consumers;

[0044] Furthermore, the historical shopping data includes: types of purchased goods, quantity, price, and purchase time; the historical browsing data includes: browsed product pages, stay time, and number of browsing times; the search data includes: search keywords, search time, and click situation of search results; the real-time browsing data includes: currently browsed products, page stay time, and browsing path; the shopping completion cycle data includes: time required from browsing to purchase and purchase frequency.

[0045] Even further, the database module also includes a data encryption and security processing unit, which is used to encrypt the collected and stored consumer shopping preference behavior data, and at the same time set access permission control, only allowing authorized system components to access the data.

[0046] The prediction module preprocesses and classifies all shopping preference behaviors stored in the database module based on machine learning algorithms, sorts out the shopping habits of consumers' personalized behaviors, and generates shopping features; the prediction model builds a prediction model based on a neural network, takes all shopping preference behaviors and shopping features as inputs into the prediction model, and obtains an analysis result;

[0047] Furthermore, a prediction threshold is set in the prediction model. When the analysis result is greater than or equal to the prediction threshold, the prediction model outputs the analysis result as the prediction result. When the analysis result is less than the prediction threshold, the sorting process of shopping habits is iteratively optimized to obtain new shopping features, and the new shopping features and all shopping preference behaviors are input into the prediction model again for analysis until the analysis result is greater than or equal to the prediction threshold, then the iterative optimization is stopped and the analysis result is output.

[0048] Furthermore, the prediction module cleans, de-duplicates, and processes missing values for all shopping preference behavior data in the database module, extracts the shopping habits of consumers' shopping preference behaviors, and performs feature transformation, standardization, and normalization processing on the shopping habits to obtain shopping features; among them, the feature transformation process of shopping habits is represented by the formula where x′ represents the feature value after feature transformation; x represents the shopping habit, that is, the original feature value; μ represents the mean of the shopping habit x; σ represents the standard deviation of the shopping habit x.

[0049] Furthermore, a prediction model is established based on a neural network. The prediction model includes an input layer, a hidden layer, and an output layer;

[0050] The input layer combines all shopping preference behaviors and shopping characteristics to obtain an input feature vector set X = [x′, D]; in the feature vector set, D represents the feature vector set of all shopping preference behaviors of consumers after preprocessing;

[0051] The hidden layer is associated with the input layer through linear transformation and a non - linear activation function, and extracts high - level features from the input data through layer - by - layer transmission; among them, the process of extracting high - level features is represented by the formula ; in the formula, Z (i) represents the result value after the linear transformation of the i - th layer neuron; W (i) represents the weight matrix of the i - th layer neuron; X (i) represents the i - th feature vector value in the input feature vector set X; a (i-1) represents the activation value of the previous layer neuron received by each layer of neurons; b (i) represents the bias vector of the i - th layer neuron; a (i) represents the activation value of the i - th layer neuron; f (i) represents the activation function of the i - th layer neuron;

[0052] Here, it needs to be further explained that the first layer of the hidden layer (i.e., i = 1) will receive the vector a 0 transmitted from the input layer, and then linearly transform the input vector through the weight matrix W (1) , the first feature vector value X (1) and the bias vector b (1) to calculate the result value Z (1) ; the purpose of this linear transformation is to map the input features to a new space to prepare for subsequent non - linear transformation.

[0053] After that, the result value Z (1) after linear transformation will undergo a non - linear transformation through the activation function f (1) to obtain the activation value a (1) of the first layer of the hidden layer; the role of the activation function is to introduce non - linearity so that the neural network can learn more complex patterns.

[0054] Each layer in the hidden layer will repeat the above - mentioned linear transformation and non - linear activation process, that is, for i = 2, 3, 4,..., i - 1; each layer will receive the activation value a (i-1) of the previous layer, and through the weight matrix W (1) , the first feature vector value X (1) and the bias vector b (1)Perform a linear transformation on the input vector, and then perform a non-linear transformation through the activation function f (i) to obtain the activation value a of the current layer (i) .

[0055] The output layer is associated with the hidden layer through a linear transformation, maps the high-level features to the output space through a linear transformation, and obtains the analysis result; among them, the calculation formula of the analysis result is:

[0056] In the formula, Z (I) represents the result value after the linear transformation of the neurons in the last layer; W (I) represents the weight matrix of the neurons in the last layer; X (I) represents the value of the last feature vector in the input feature vector set X; a (I-1) represents the activation value of the neurons in the previous layer received by the neurons in the last layer; b (I) represents the bias vector of the neurons in the last layer; f (I) represents the activation function of the neurons in the last layer; Y represents the analysis result

[0057] Here, it needs to be further explained that the output layer (i.e., i = I) will receive the activation value a of the last layer of the hidden layer (I-1) , and the output layer passes through the weight matrix W (I) , the value X of the last feature vector in the input feature vector set X (I) and the bias vector b (I) to perform a linear transformation and calculate the result value Z (I) ; the purpose of this linear transformation is to map the high-level features extracted by the hidden layer to the output space for final prediction or classification

[0058] Moreover, the output layer usually uses a non-linear activation function f (I) to perform a non-linear transformation on Z (I) . Finally, after the linear transformation and non-linear transformation of the output layer, the neural network will output the final result Y

[0059] Furthermore, the setting of the prediction threshold includes the following steps:

[0060] S1. Based on the overall shopping preference behaviors of consumers, a machine learning algorithm is used to train a prediction model, and the prediction model is evaluated and optimized through cross-validation methods. Specifically: Collect all the shopping preference behavior data of consumers from the database module, including historical shopping data, historical browsing data, real-time search data, real-time browsing data, and shopping completion cycle data. Clean, deduplicate, and handle missing values for the collected data to ensure the accuracy and integrity of the data. Extract consumers' shopping habits, and perform feature transformation, standardization, and normalization processing to generate shopping features. Use machine learning algorithms (such as random forest, gradient boosting tree, etc.) to train the preprocessed data to build an initial prediction model. Use cross-validation methods (such as K-fold cross-validation) to evaluate the model, select the model parameters with the best performance. According to the evaluation results, adjust and optimize the model to improve the prediction accuracy of the model. Use multiple evaluation metrics (such as accuracy, recall rate, F1 score, etc.) to comprehensively evaluate the trained model, analyze the error sources of the model, and conduct targeted optimization for the parts with larger errors.

[0061] S2. According to the analysis results, initially set the prediction threshold as value A, collect consumers' feedback, and dynamically adjust value A through the F1 score to obtain value B. Specifically: According to the analysis results of the model, initially set a prediction threshold (value A). This threshold should be able to distinguish the high-confidence and low-confidence parts in the prediction results. Collect consumers' satisfaction and feedback on the recommendation results through methods such as user research and questionnaires, organize and analyze the feedback, and understand consumers' expectations and needs for the recommendation results. Dynamically adjust the initially set threshold according to consumers' feedback and the F1 score to obtain value B. The F1 score is the harmonic mean of the accuracy and recall rate, which can comprehensively reflect the performance of the model.

[0062] S3. Real-time monitor the comparison result between the analysis result of the prediction model and value B, judge the accuracy rate, and adjust value B according to the change of the accuracy rate. Specifically: Real-time monitor the comparison result between the analysis result of the prediction model and the prediction threshold (value B), record performance metrics such as the prediction accuracy rate and recall rate of the model at different time periods. According to the real-time monitoring results, judge whether the prediction accuracy rate of the model meets the requirements. If the accuracy rate drops, the model needs to be retrained and parameter-tuned. Adjust the prediction threshold (value B) timely according to the change of the accuracy rate to ensure that the model can maintain a high prediction accuracy rate at different time periods.

[0063] The personalized recommendation module re-ranks the products based on all shopping preference behaviors, combines the analysis results with the product information, and generates a personalized recommendation list.

[0064] Further, the process of the personalized recommendation module generating a personalized recommendation list includes the following steps:

[0065] A1. Based on all the shopping preference behaviors of consumers and the analysis results of the prediction module, conduct a preliminary ranking of products; specifically: integrate all the shopping preference behavior data of consumers and the analysis results of the prediction module, extract feature information related to products, such as product categories, prices, brands, etc. According to the integrated data, the ranking basis can include consumers' historical purchase records, browsing times, staying time, etc., and conduct a preliminary ranking of products.

[0066] A2. Consider the popularity indicators of the real-time search volume, click volume, and purchase volume of products, and dynamically adjust the preliminary ranking results; specifically: collect the popularity indicators such as the search volume, click volume, and purchase volume of products in real time, and dynamically adjust the preliminary ranking results according to the real-time popularity indicators, promoting the products with higher popularity to the front of the ranking results to improve the timeliness and accuracy of recommendations.

[0067] A3. According to consumers' personalized shopping preferences, conduct weighted processing on products; specifically: extract consumers' personalized shopping preference characteristics, such as brand preference, price sensitivity, purchase cycle, etc. Conduct weighted processing on products according to consumers' personalized shopping preference characteristics, increasing the weight of products that meet consumers' preferences in the recommendation list and reducing the weight of products that do not meet the preferences.

[0068] A4. Consider consumers' recent shopping behaviors and fine-tune the recommendation list to finally obtain a personalized recommendation list. Specifically: analyze consumers' recent shopping behavior data, such as the types of products recently purchased, price ranges, etc., and fine-tune the recommendation list according to consumers' recent shopping behaviors, promoting the products related to consumers' recent shopping behaviors to the front of the recommendation list to improve the pertinence and accuracy of recommendations.

[0069] The visualization module displays the recommendation list to consumers, and identifies and saves the subsequent operations of consumers into the database module for subsequent prediction.

[0070] Further, the visualization module is built-in with an instant feedback mechanism. Consumers can evaluate the analysis results through the instant feedback mechanism, and the prediction module iteratively optimizes the analysis results according to the evaluation results.

[0071] To further ensure the prediction accuracy of the prediction model, the prediction module uses the cross-entropy loss function to evaluate the prediction performance of the prediction model. Among them, the formula of the cross-entropy loss function is:

[0072]

[0073] Wherein, N represents the number of all shopping preference behaviors; y i represents the true label of the i-th shopping preference behavior; represents the predicted probability of the i-th shopping preference behavior;

[0074] Meanwhile, calculate the gradient of the cross-entropy loss function with respect to the parameters of the prediction model through the backpropagation algorithm, and then use the gradient descent method to update the parameters of the prediction model; wherein, the update formula of the gradient descent method is: Wherein, θ represents the parameters of the prediction model; θ′ represents the updated parameters of the prediction model; η represents the learning rate; represents the gradient of the cross-entropy loss function with respect to the parameters of the prediction model.

[0075] When the present invention is used, the database module: adopts dynamic big data technology to collect and store all the shopping preference behavior data of consumers, including historical shopping data, historical browsing data, real-time search data, real-time browsing data, and shopping completion cycle data. It also includes a data encryption and security processing unit to ensure the security and privacy protection of the data.

[0076] The prediction module: preprocess and classify the collected shopping preference behaviors based on machine learning algorithms, sort out the shopping habits of consumers, and generate shopping features. Use a neural network to establish a prediction model, take the shopping preference behaviors and shopping features as inputs, and output the analysis results. A prediction threshold is set in the prediction model to judge the accuracy of the analysis results and perform iterative optimization to improve the prediction performance. Evaluate the prediction performance of the prediction model through the cross-entropy loss function and use the gradient descent method to update the model parameters.

[0077] The personalized recommendation module: sort and recommend products based on the shopping preference behaviors of consumers and the analysis results of the prediction module. Dynamically adjust the recommendation list considering the real-time popularity indicators of products (such as search volume, click volume, purchase volume). Perform weighted processing on products according to the personalized shopping preferences of consumers and make fine-tuning considering the recent shopping behaviors of consumers.

[0078] The visualization module: display the recommendation list to consumers and collect the operation feedback of consumers.

[0079] An in-built instant feedback mechanism is provided to allow consumers to evaluate the analysis results for iterative optimization.

[0080] Based on the ideal embodiments of the present invention as an inspiration, through the above description, relevant personnel can completely make various changes and modifications without departing from the technical idea of the present invention. The technical scope of the present invention is not limited to the content in the specification, and must be determined according to the scope of the claims.

Claims

1. A consumer online shopping preference prediction system for dynamic big data, characterized by: include: Database module, prediction module, personalized recommendation module and visualization module; The database module uses dynamic big data technology to collect and store all shopping preference behaviors from consumers; This includes consumers’: historical shopping data, historical browsing data, real-time search data, real-time browsing data, and shopping completion cycle data; The prediction module pre-processes and classifies all shopping preference behaviors stored in the database module based on a machine learning algorithm, sorts out the shopping habits of consumers' personalized behaviors, and generates shopping features; the prediction model establishes a prediction model based on a neural network, and inputs all shopping preference behaviors and shopping features into the prediction model to obtain analysis results; The personalized recommendation module re-sorts the commodities based on the entire shopping preference behavior, according to the commodity information, and in combination with the analysis result, to generate a personalized recommendation list; The visualization module displays the recommendation list to the consumer, and identifies and obtains the consumer's subsequent operations and saves them in the database module for subsequent prediction.

2. The consumer online shopping preference prediction system for dynamic big data according to claim 1 is characterized by: A prediction threshold is set in the prediction model. When the analysis result is greater than or equal to the prediction threshold, the prediction model outputs the analysis result as the prediction result. When the analysis result is less than the prediction threshold, the shopping habit sorting process is iteratively optimized to obtain new shopping features. The new shopping features and all shopping preference behaviors are input into the prediction model again for analysis until the analysis result is greater than or equal to the prediction threshold. Then, the iterative optimization is stopped and the analysis result is output.

3. The consumer online shopping preference prediction system for dynamic big data according to claim 1 is characterized by: The prediction module cleans, removes duplicates and processes missing values ​​of all the shopping preference behavior data in the database module, extracts the shopping habits derived from the consumer's shopping preference behavior, and performs feature conversion, standardization and normalization on the shopping habits to obtain the shopping features; wherein the feature conversion process of the shopping habits is represented by the formula Indicates; in the formula, x′ represents the eigenvalue after feature transformation; x represents shopping habits, that is, the original eigenvalue; μ represents the mean of shopping habits x; σ represents the standard deviation of shopping habits x.

4. The consumer online shopping preference prediction system for dynamic big data according to claim 1 is characterized by: Establishing the prediction model based on the neural network, wherein the prediction model includes an input layer, a hidden layer, and an output layer; The input layer combines all the shopping preference behaviors and shopping features to obtain an input feature vector set X=[x′, D]; In the feature vector set, D represents the feature vector set of all shopping preference behaviors of consumers after preprocessing; The hidden layer is associated with the input layer through linear transformation and nonlinear activation function, and high-level features in the input data are extracted by layer-by-layer transmission; wherein the extraction process of the high-level features is given by the formula In the formula, Z (i) W represents the result value after linear transformation of the neurons in the i-th layer; (i) represents the weight matrix of the neurons in the i-th layer; X (i) represents the value of the i-th eigenvector in the input eigenvector set X; a (i-1) Indicates the activation value of the previous layer of neurons received by each layer of neurons; b (i) Represents the bias vector of the i-th layer neuron; a (i) represents the activation value of the neuron in the i-th layer; f (i) represents the activation function of the neurons in the i-th layer; The output layer is associated with the hidden layer through linear transformation, and the high-level features are mapped to the output space through linear transformation to obtain the analysis result; wherein the calculation formula of the analysis result is: In the formula, Z (I) W represents the result value after linear transformation of the last layer of neurons; (I) represents the weight matrix of the last layer of neurons; X (I) Represents the last eigenvector value in the input eigenvector set X; a (I-1) Indicates the activation value of the previous layer of neurons received by the last layer of neurons; b (I) represents the bias vector of the last layer of neurons; f (I) represents the activation function of the last layer of neurons; Y represents the analysis result.

5. The consumer online shopping preference prediction system for dynamic big data according to claim 2 is characterized by: The setting of the prediction threshold comprises the following steps: S1. Based on all shopping preference behaviors of consumers, the prediction model is trained using a machine learning algorithm, and the prediction model is evaluated and optimized using a cross-validation method; S2. According to the analysis results, the prediction threshold is initially set as value A, and consumer feedback is collected. Value A is dynamically adjusted through the F1 score to obtain value B; S3. Compare the analysis results of the real-time monitoring prediction model with the value B, determine the accuracy, and adjust the value B according to the change in accuracy.

6. The consumer online shopping preference prediction system for dynamic big data according to claim 1 is characterized by: The process of generating a personalized recommendation list by the personalized recommendation module includes the following steps: A1. Preliminary sorting of products based on all shopping preference behaviors of consumers and the analysis results of the prediction module; A2. Consider the real-time search volume, click volume, and purchase volume of the product, and dynamically adjust the preliminary ranking results; A3. Weighting products according to consumers’ personalized shopping preferences; A4. Consider the consumer’s recent shopping behavior and fine-tune the recommendation list to ultimately obtain a personalized recommendation list.

7. The consumer online shopping preference prediction system for dynamic big data according to claim 1 is characterized by: The visualization module has a built-in instant feedback mechanism, through which consumers can evaluate the analysis results, and the prediction module iteratively optimizes the analysis results according to the evaluation results.

8. The consumer online shopping preference prediction system for dynamic big data according to claim 1 is characterized by: The database module also includes a data encryption and security processing unit for encrypting the collected and stored consumer shopping preference behavior data and setting access permission control to allow only authorized system components to access the data.

9. The consumer online shopping preference prediction system for dynamic big data according to claim 1 is characterized by: The historical shopping data includes: the type, quantity, price and purchase time of purchased goods; the historical browsing data includes: the browsed product pages, dwell time and number of views; the search data includes: search keywords, search time and search result clicks; the real-time browsing data includes: the currently browsed product, page dwell time and browsing path; the shopping completion cycle data includes: the time required from browsing to purchase and the purchase frequency.

10. The consumer online shopping preference prediction system for dynamic big data according to claim 1 or 2, characterized in that: The prediction module evaluates the prediction performance of the prediction model by using a cross entropy loss function, wherein the formula of the cross entropy loss function is: Where N represents the number of all shopping preference behaviors; y i Represents the true label of the i-th shopping preference behavior; represents the predicted probability of the i-th shopping preference behavior; At the same time, the gradient of the cross entropy loss function to the prediction model parameters is calculated by the back propagation algorithm, and then the prediction model parameters are updated using the gradient descent method; wherein the update formula of the gradient descent method is: In the formula, θ represents the prediction model parameters; θ′ represents the updated prediction model parameters; η represents the learning rate; Represents the gradient of the cross entropy loss function with respect to the prediction model parameters.

Images