E-commerce user behavior deep insight analysis method and system based on big data artificial intelligence

Abstract

The invention relates to an e-commerce user behavior deep insight analysis method and system based on big data artificial intelligence. The method comprises the steps of performing preprocessing and multi-dimensional feature engineering on historical behavior data of e-commerce users; based on the feature set, using a clustering algorithm to subdivide the user into a plurality of groups with different behavior features; independently training a re-purchase behavior prediction model for each group in a layered modeling mode and outputting a re-purchase probability by taking grouping as a condition; applying a Shapley value decomposition technology to quantitatively analyze the contribution degree of each feature to a prediction result, and identifying key influence factors of different groups; and integrating the user grouping, the re-purchasing probability and the feature contribution degree, and automatically generating a differentiated marketing strategy for different groups. The problems that a traditional method neglects user heterogeneity, the model interpretability is poor, and strategy making lacks data basis are solved, and the re-purchase prediction precision and the effectiveness of a marketing strategy are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of behavioral analysis technology, and in particular to a method and system for in-depth insight analysis of e-commerce user behavior based on big data and artificial intelligence. Background Technology

[0002] With the rapid development of e-commerce and increasingly fierce market competition, the core competition of e-commerce platforms has shifted from early traffic acquisition to in-depth user operations. Accurately predicting users' repeat purchase behavior and formulating effective retention and conversion strategies accordingly has become key for enterprises to enhance customer lifetime value and optimize marketing resource allocation.

[0003] Currently, existing technologies for e-commerce user behavior analysis and repeat purchase prediction mainly develop in three directions: First, customer segmentation and value assessment based on user profiles and statistical models; second, directly building user-level repeat purchase prediction models using various machine learning algorithms to improve prediction accuracy; and third, introducing model interpretability tools to understand model decisions and assist in strategy formulation. These technologies have all been applied to some extent in practice.

[0004] The aforementioned technologies employ a unified prediction framework, which implicitly assumes that all users follow the same decision-making logic. This fails to effectively identify and model the significant behavioral heterogeneity among different user groups, resulting in limited prediction accuracy when faced with diverse consumption patterns.

[0005] Based on this, this application provides a method and system for in-depth insight analysis of e-commerce user behavior based on big data and artificial intelligence. Summary of the Invention

[0006] To address the problem that a unified prediction framework implicitly assumes all users follow the same decision-making logic, fails to effectively identify and model significant behavioral heterogeneity among different user groups, and thus limits the prediction accuracy of the model when faced with diverse consumption patterns, this application provides a method and system for in-depth insight analysis of e-commerce user behavior based on big data and artificial intelligence.

[0007] Firstly, this application provides a method for in-depth insight analysis of e-commerce user behavior based on big data and artificial intelligence, employing the following technical solution: including: Receive user historical behavior data, which includes user attribute data and user interaction behavior data with products and merchants. Perform preprocessing and feature engineering on the user historical behavior data to construct a multi-dimensional feature set for characterizing user behavior patterns. Based on the multi-dimensional feature set, a clustering algorithm is used to segment users, resulting in multiple groups with different behavioral characteristics and corresponding group labels for each user. Using the aforementioned group labels as conditions, and based on the multi-dimensional feature set and corresponding historical repurchase markers, a hierarchical modeling approach is adopted to train an independent repurchase behavior prediction model for each user group and calculate the repurchase probability. Based on the repurchase behavior prediction model, the Shapley value decomposition technique is applied for analysis to calculate the feature contribution of each feature to the model prediction results, so as to identify the key factors affecting the repurchase behavior of different user groups. Based on the segmentation results, the repurchase probability of the repurchase behavior prediction model, and the feature contribution, differentiated marketing strategies are generated for different user segments.

[0008] Preferably, the step of receiving user historical behavior data includes user attribute data and user interaction data with products and merchants. The user historical behavior data is preprocessed and feature-engineered to construct a multi-dimensional feature set representing user behavior patterns, including: Receive user historical behavior data from e-commerce platforms, perform data cleaning on the user historical behavior data, fill or impute missing values, identify and process outliers, and standardize and transform the timestamp field. Based on the cleaned user historical behavior data, multi-dimensional features are extracted. The extracted feature dimensions include at least: user static attribute features, user dynamic behavior features, user-product interaction features, user-merchant relationship features, and time-series behavior features. The extracted multi-dimensional features are normalized and encoded to construct the multi-dimensional feature set for subsequent clustering and predictive analysis.

[0009] Preferably, based on the multi-dimensional feature set, a clustering algorithm is used to segment users, resulting in multiple groups with different behavioral characteristics and corresponding group labels for each user, including: A subset of behavioral features for cluster analysis is selected from the multi-dimensional feature set. The subset of behavioral features includes at least the core indicators that characterize user activity, purchase conversion rate, category focus and brand loyalty. Based on the aforementioned subset of behavioral features, the K-means clustering algorithm is used to perform cluster analysis on users. During the clustering process, the elbow method and silhouette coefficient are combined to determine the optimal number of clusters K. Clustering is performed based on the optimal number of clusters K to obtain a grouping result that divides all users into K mutually exclusive clusters, and a corresponding grouping label is assigned to each user. Analyze and output the center point or typical behavioral feature description of each user group on the subset of behavioral features.

[0010] Preferably, the step of using the grouping labels as conditions, based on the multi-dimensional feature set and corresponding historical repurchase markers, and employing a hierarchical modeling approach to train an independent repurchase behavior prediction model for each user group, and calculating the repurchase probability, includes: Based on the group labels corresponding to each user, the multi-dimensional feature set and the corresponding historical repurchase tags are divided into subsets corresponding to each user group. For each user group's subset, a machine learning classification algorithm is selected as the base model. For each user segment, a base model is trained based on its corresponding subset, and the repurchase behavior prediction model corresponding to the segment is obtained through hyperparameter optimization. Meanwhile, during the training process, oversampling techniques are used to address the imbalance between positive and negative samples in the subsets of each cluster. Based on the trained repurchase behavior prediction model, the feature data of the target user is input to obtain the predicted probability value of the user's repurchase behavior within a preset time window in the future, which is used as the repurchase probability.

[0011] Preferably, for each user segment, a base model is trained based on its corresponding subset, and the repurchase behavior prediction model corresponding to the segment is obtained through hyperparameter optimization, including: For each user segment, the base model is determined to be the gradient boosting decision tree algorithm. A set of hyperparameters to be optimized is set for the gradient boosting decision tree algorithm. The set of hyperparameters includes at least the tree depth, learning rate, number of iterations, and regularization parameter. On a subset of the dataset, cross-validation is used to optimize the set of hyperparameters in order to maximize the model’s evaluation metrics on the validation set. The optimal hyperparameter combination obtained by the optimization search is configured into the gradient boosting decision tree algorithm, and the model is trained using all the data of the subset, thus obtaining the repurchase behavior prediction model corresponding to the user group and with optimized parameters.

[0012] Preferably, the step of using cross-validation to optimize the hyperparameter set on a subset of data sets to maximize the model's evaluation metrics on the validation set includes: For the subset, a hierarchical K-fold cross-validation method is used to divide it into K training-validation pairs to maintain the ratio of positive and negative samples in each fold consistent with the original dataset. The evaluation index is set as the area under the receiver operating characteristic curve, and multiple hyperparameter combinations are generated within the set range of the hyperparameter set using grid search or random search strategies. For each set of hyperparameters, cross-validation is performed on K training-validation pairs, and the average AUC on the K validation sets is calculated. Based on the average AUC, the hyperparameter combination that maximizes the average AUC is selected from multiple hyperparameter combinations as the optimal hyperparameter combination.

[0013] Preferably, based on the repurchase behavior prediction model, the Shapley value decomposition technique is applied for analysis to calculate the feature contribution of each feature to the model prediction result, so as to identify the key factors affecting the repurchase behavior of different user groups; based on the grouping results, the repurchase probability of the repurchase behavior prediction model, and the feature contribution, differentiated marketing strategies for different user groups are generated, including: For each user segment corresponding to the repurchase behavior prediction model, the Shapley value decomposition technique is applied to calculate the contribution value of each feature in the multi-dimensional feature set to the prediction result of the corresponding segment model. Aggregate the feature contribution values ​​of all user groups to generate a global feature importance ranking and identify the core features that have the greatest impact on overall repurchase prediction; By comparing and analyzing the differences in the contribution values ​​of the same feature among different user groups, key influencing factors with group specificity can be identified. Based on the grouping results, users are divided into corresponding behavior types; based on the repurchase probability, users in each group are further divided into high, medium, and low repurchase tendency intervals; and based on the feature contribution, the global feature importance ranking, and group-specific key influencing factors, the top N key features driving the repurchase behavior of each group are identified. Based on the behavior type, the repurchase tendency range, and the corresponding top N key features, a set of targeted and differentiated marketing strategies are matched and generated from a preset strategy rule base. The strategy rule base includes at least a membership benefit strategy for high-value loyal users, a decision-making support strategy for potential value-seeking users, and an instant conversion strategy for price-sensitive impulsive users.

[0014] Secondly, this application discloses a device for deep insight analysis of e-commerce user behavior based on big data and artificial intelligence, which adopts the following technical solution, including: The behavioral feature module is used to receive user historical behavior data, which includes user attribute data and user interaction behavior data with products and merchants. The module performs preprocessing and feature engineering on the user historical behavior data to construct a multi-dimensional feature set for characterizing user behavior patterns. The clustering module is used to segment users based on the multi-dimensional feature set using a clustering algorithm, and to obtain the results of dividing users into multiple groups with different behavioral characteristics and the corresponding group labels for each user. The repeat purchase prediction module is used to train an independent repeat purchase behavior prediction model for each user group based on the group labels, the multi-dimensional feature set and the corresponding historical repeat purchase tags, and a hierarchical modeling approach, and to calculate the repeat purchase probability. The repeat purchase analysis module is used to analyze the repeat purchase behavior prediction model using Shapley value decomposition technology, calculate the feature contribution of each feature to the model prediction results, and identify the key factors affecting the repeat purchase behavior of different user groups. The marketing strategy module is used to generate differentiated marketing strategies for different user groups based on the segmentation results, the repurchase probability of the repurchase behavior prediction model, and the feature contribution.

[0015] Thirdly, this application also provides a control device, the device comprising: It includes a memory and a processor, wherein the memory stores a computer program that can be loaded and executed by the processor, such as the above-described method for in-depth insight analysis of e-commerce user behavior based on big data and artificial intelligence.

[0016] Fourthly, this application also provides a computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above regarding the method for in-depth insight analysis of e-commerce user behavior based on big data artificial intelligence.

[0017] In summary, this application receives multi-source user behavior data from e-commerce platforms. Preprocessing is performed through data cleaning, missing value imputation, and time-series standardization to construct a multi-dimensional feature set encompassing user static attributes, dynamic behaviors, interaction relationships, and time-series patterns. Subsequently, based on the feature set, an improved K-means clustering algorithm is used, combined with the elbow method and silhouette coefficient to determine the optimal number of clusters, automatically dividing users into groups with significant behavioral differences. Furthermore, hierarchical modeling is implemented based on cluster labels, independently training a repurchase prediction model centered on gradient boosting decision trees for each group. Cross-validation and hyperparameter optimization ensure model performance, and the repurchase probability at the user level is output. Next, Shapley value decomposition is applied to calculate the feature contribution of each group model, identifying core driving factors and group-specific key features from a global and group-specific perspective. Finally, by integrating user group affiliation, repurchase probability intervals, and feature importance analysis results, a rule engine is driven to generate and output differentiated and actionable marketing strategies for different groups from a pre-set strategy library. This solves the problems of traditional methods ignoring user heterogeneity, poor model interpretability, and high implementation threshold. It can significantly improve the accuracy of repeat purchase prediction and the effectiveness of marketing strategies with lower data and computing power costs, and achieve refined user operations. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a method for in-depth insight analysis of e-commerce user behavior based on big data and artificial intelligence.

[0019] Figure 2 This is a structural block diagram of an e-commerce user behavior deep insight analysis device based on big data and artificial intelligence. Detailed Implementation

[0020] The following combination Figure 1 - Figure 2 This application will be described in further detail.

[0021] Reference Figure 1 The embodiments of this application include at least steps S10 to S50.

[0022] S10 receives user historical behavior data, which includes user attribute data and user interaction data with products and merchants. The user historical behavior data is preprocessed and feature-engineered to construct a multi-dimensional feature set to characterize user behavior patterns.

[0023] S20 uses a clustering algorithm based on a multi-dimensional feature set to segment users, resulting in multiple groups with different behavioral characteristics and corresponding group labels for each user.

[0024] S30 uses the group label as a condition, and based on the multi-dimensional feature set and the corresponding historical repurchase markers, adopts a hierarchical modeling approach to train an independent repurchase behavior prediction model for each user group and calculate the repurchase probability.

[0025] S40, based on the repurchase behavior prediction model, uses Shapley value decomposition technology for analysis to calculate the feature contribution of each feature to the model prediction results, in order to identify the key factors affecting the repurchase behavior of different user groups.

[0026] S50 generates differentiated marketing strategies for different user groups based on the segmentation results, repurchase probability of the repurchase behavior prediction model, and feature contribution.

[0027] Specifically, the process involves preprocessing and feature engineering the raw e-commerce data to construct a multi-dimensional feature set characterizing user behavior. Next, based on this feature set, a clustering algorithm is used to segment users into multiple groups with distinct behavioral patterns, and each user is labeled with a subgroup label. Then, using subgrouping as a condition, a hierarchical modeling approach is employed to independently train a high-precision repurchase prediction model for each group, outputting user-level repurchase probabilities. Subsequently, Shapley value technology is applied to quantify the contribution of each feature to the prediction results, accurately identifying the key factors driving repurchases in different groups. Finally, by integrating user subgrouping, repurchase probabilities, and feature contributions, highly differentiated marketing strategies are automatically generated and output. This effectively solves the problems of traditional methods, such as neglecting user heterogeneity, the black-box nature of prediction models, and the lack of data support for strategy formulation, significantly improving the intelligence level of user operations and the return on investment in marketing.

[0028] In some embodiments, step S10 specifically includes the following steps: receiving user historical behavior data from an e-commerce platform; cleaning the user historical behavior data, filling or imputing missing values, identifying and processing outliers, and standardizing the timestamp field; extracting multi-dimensional features based on the cleaned user historical behavior data, wherein the extracted feature dimensions include at least: user static attribute features, user dynamic behavior features, user-product interaction features, user-merchant relationship features, and time-series behavior features; and standardizing and encoding the extracted multi-dimensional features to construct a multi-dimensional feature set for subsequent clustering and predictive analysis.

[0029] Specifically, by cleaning, transforming, and reconstructing the raw data, the chaotic user behavior records are transformed into a systematic and computable structured feature set. Data cleaning ensures the accuracy and consistency of the analytical foundation; multidimensional feature extraction comprehensively portrays user profiles from the dimensions of user attributes, dynamic behavior, interaction relationships, and time, aiming to deeply capture potential behavioral patterns and preferences. Feature normalization and encoding further eliminate the impact of differences in data scale and type on the model. Transforming the raw data into high-quality, high-information-density analytical fuel lays a solid data foundation for subsequent accurate user segmentation, reliable predictive modeling, and effective attribution analysis.

[0030] In some embodiments, step S20 specifically includes the following steps: selecting a subset of behavioral features from a multi-dimensional feature set for cluster analysis, wherein the subset of behavioral features includes at least core indicators representing user activity, purchase conversion rate, category focus, and brand loyalty; and performing cluster analysis on users using the K-means clustering algorithm based on the subset of behavioral features, the optimization objective of which is to minimize the within-cluster squared error (SSE), the formula for which the objective function is calculated is as follows: ; Where k represents the preset number of clusters, that is, the total number of groups into which users will ultimately be divided; Let i represent the i-th cluster, which is the set of all user samples belonging to this group. Indicates belonging to a cluster A user sample point, whose coordinates are composed of the user's values ​​on a subset of behavioral features. Cluster The centroid is the mean vector of all user sample points in the feature space. This represents the distance from sample point x to the centroid of its cluster. The square of the Euclidean distance is used to measure the degree of deviation of a single sample from the population center.

[0031] During the clustering process, the optimal number of clusters K is determined by combining the elbow method and the silhouette coefficient evaluation. Clustering is performed according to the optimal number of clusters K to obtain the clustering results that divide all users into K mutually exclusive clusters, and a corresponding cluster label is assigned to each user. The centroid or typical behavioral feature description of each user cluster on the behavioral feature subset is analyzed and output.

[0032] Specifically, this approach achieves automated user segmentation through a data-driven method. First, key behavioral indicators are selected to construct feature subsets, accurately depicting core user behavior patterns. Then, the K-means algorithm combined with evaluation methods is used to determine the optimal number of clusters, ensuring an optimal balance between intra-group similarity and inter-group differences. Finally, labeled clustering results and typical behavioral profiles for each user group are output. This process efficiently categorizes massive numbers of users into segmented groups with clear behavioral characteristics, providing a scientific basis for subsequent differentiated and precise predictions and strategic interventions.

[0033] In some embodiments, step S30 specifically includes the following steps: based on the segmentation labels corresponding to each user, the multi-dimensional feature set and the corresponding historical repurchase markers are divided into subsets corresponding to each user segment; for each subset of user segments, a machine learning classification algorithm is selected as the base model; for each user segment, the base model is trained based on its corresponding subset, and the repurchase behavior prediction model corresponding to the segment is obtained through hyperparameter optimization; at the same time, during the training process, oversampling technology is used to handle the imbalance of positive and negative repurchase samples in the subset of each segment; based on the trained repurchase behavior prediction model, the feature data of the target user is input to obtain the predicted probability value of the user's repurchase behavior within a preset time window in the future, which is used as the repurchase probability.

[0034] Specifically, based on user segmentation results, the data and problem space are decomposed, and customized predictive models are trained independently for each homogeneous group. Through targeted selection of base models, hyperparameter optimization, and sample balancing, it is ensured that each sub-model can learn the complex behavioral patterns and repurchase decision mechanisms of its corresponding group most efficiently, effectively avoiding prediction bias caused by a single model fitting the average user. Finally, a quantified, personalized repurchase probability is output for each user. This probability is the most core and direct decision-making basis for subsequent strategy generation, enabling marketing interventions to accurately focus on high-potential individuals.

[0035] In some embodiments, step S30 further includes the following steps: for each user segment, determine the base model as the gradient boosting decision tree algorithm, set a set of hyperparameters to be optimized for the gradient boosting decision tree algorithm, the set of hyperparameters including at least the tree depth, learning rate, number of iterations and regularization parameter; on the subset dataset, use cross-validation to optimize the set of hyperparameters to maximize the evaluation index of the model on the validation set; configure the optimal combination of hyperparameters obtained by optimization search to the gradient boosting decision tree algorithm, and use all the data of the subset dataset to complete the model training, thereby obtaining the repurchase behavior prediction model corresponding to the user segment with optimized parameters.

[0036] Specifically, by uniformly selecting gradient boosting decision trees—a powerful prediction algorithm adept at capturing complex nonlinear relationships and friendly to category features—as the technical foundation, a high-starting-point modeling capability is provided for each cluster. Furthermore, by systematically defining a set of key hyperparameters and employing rigorous optimization through cross-validation, the aim is to find the most suitable model configuration for each group's specific data distribution, maximizing its predictive and discriminative power. This optimization process is crucial for transforming a general algorithm into a high-precision, customized tool for specific user behavior patterns, ensuring that each deployed cluster model achieves its theoretically optimal performance, thus providing a solid technical guarantee for the accuracy of the entire hierarchical prediction framework.

[0037] In some embodiments, step S30 further includes the following steps: for a subset of data, a hierarchical K-fold cross-validation method is used to divide it into K training-validation pairs to maintain the ratio of positive and negative samples in each fold consistent with the original dataset; the evaluation metric is set as the area under the receiver operating characteristic curve, and multiple sets of hyperparameter combinations are generated within a set range of hyperparameter sets using a grid search or random search strategy; for each set of hyperparameter combinations, cross-validation is performed on the K training-validation pairs, and the average AUC is calculated on the K validation sets; based on the average AUC, the hyperparameter combination that maximizes the average AUC is selected from the multiple sets of hyperparameter combinations as the optimal hyperparameter combination.

[0038] ; AUC represents the area under the receiver operating characteristic curve, ranging from [0, 1]. It is used to comprehensively evaluate the overall performance of the model at different classification thresholds; the closer the value is to 1, the stronger the model's discriminative ability. The True Positive Rate (TPR) is calculated using the following formula: FPR represents the proportion of samples correctly predicted as positive by the model out of all actual positive samples. The FPR stands for False Positive Rate, calculated using the formula: , represents the proportion of samples that the model incorrectly predicted as positive out of all actual negative samples, and d(FPR) represents the integral of the false positive rate FPR. The geometric meaning of AUC is the area enclosed by the ROC curve (with FPR as the horizontal axis and TPR as the vertical axis) and the horizontal axis.

[0039] Specifically, hierarchical K-fold cross-validation is used to robustly estimate the model's generalization ability while maintaining data distribution consistency. AUC is chosen as the core metric to effectively overcome the interference of imbalanced positive and negative samples in repeat purchase prediction scenarios. Combined with a systematic parameter search strategy, this method can objectively and stably select the optimal parameter combination that maximizes the model's discriminative power. This provides a reliable and automated quality inspection and tuning process for the customized model for each cluster, ensuring that the final deployed model configuration is a fully validated optimal solution. Therefore, from a fundamental technical perspective, this guarantees the high accuracy and reliability of the repeat purchase probability output by the entire hierarchical prediction system.

[0040] In some embodiments, steps S40 and S50 specifically include the following steps: For the repurchase behavior prediction model corresponding to each user segment, apply Shapley value decomposition technology to calculate the contribution value of each feature in the multi-dimensional feature set to the prediction result of the corresponding segment model; aggregate the feature contribution values ​​of all user segments to generate a global feature importance ranking and identify the core features that have the greatest impact on the overall repurchase prediction; compare and analyze the differences in the contribution values ​​of the same feature among different user segments to identify key influencing factors with group specificity; based on the segmentation results, classify users into corresponding behavior types; based on the repurchase probability, further divide users in each segment into high, medium, and low repurchase tendency intervals; and based on feature contribution, global feature importance ranking, and key influencing factors with group specificity, identify the top N key features driving the repurchase behavior of each segment; according to the behavior type, repurchase tendency interval, and corresponding top N key features, match and generate a set of targeted differentiated marketing strategies from a preset strategy rule base, wherein the strategy rule base includes at least a membership benefit strategy for high-value loyal users, a decision-making boosting strategy for potential value-seeking users, and an instant conversion strategy for price-sensitive impulsive users.

[0041] Shapley value The calculation formula is as follows: ; in, Let represent the Shapley value of feature j, which is the average marginal contribution of that feature to the model's prediction result. Let F represent the complete set of all features. Let j represent the target feature whose contribution is to be calculated. Let S represent any subset of the feature set F that does not contain feature j. |S| and |F| represent the number of features contained in subset S and the entire set F, respectively. This represents the evaluation value of the model trained using only a subset of features S for predicting the target (such as the probability of repeat purchase). This represents the evaluation value of the model obtained by training using a subset of features S plus feature j. This represents the marginal gain of the model evaluation value when feature j is added to subset S.

[0042] Specifically, the algorithm first uses Shapley scores to perform a micro-analysis of each prediction model, quantifying the contribution of each feature. This not only yields a global importance ranking but, more importantly, reveals the unique factors driving repeat purchases among different groups. Next, it comprehensively utilizes segmentation tags, repeat purchase probability intervals, and the aforementioned attribution analysis results to create a multi-dimensional and refined profile of users' states. Finally, by intelligently matching this deep insight with a pre-defined strategy rule base, it automatically generates highly personalized marketing action plans with clear causal relationships.

[0043] The implementation principle of the deep insight analysis method for e-commerce user behavior based on big data and artificial intelligence in this application is as follows: Multi-source user behavior data from e-commerce platforms is received and preprocessed through data cleaning, missing value imputation, and time series standardization. A multi-dimensional feature set covering user static attributes, dynamic behaviors, interaction relationships, and time series patterns is constructed. Subsequently, based on the feature set, an improved K-means clustering algorithm is used, combined with the elbow method and silhouette coefficient to determine the optimal number of clusters, automatically dividing users into groups with significant behavioral differences. Then, hierarchical modeling is implemented based on cluster labels, independently training a repurchase prediction model with gradient boosting decision trees as the core for each group. Cross-validation and hyperparameter optimization are used to ensure model performance, and the repurchase probability at the user granularity is output. Afterwards, Shapley value decomposition technology is applied to calculate the feature contribution of each group model, identifying core driving factors and group-specific key features from a global and group comparison perspective. Finally, by integrating the results of user group affiliation, repurchase probability intervals, and feature importance analysis, the rule engine is driven to generate and output differentiated and operable marketing strategies for different groups from a preset strategy library. This solves the problems of traditional methods ignoring user heterogeneity, poor model interpretability, and high implementation threshold. It can significantly improve the accuracy of repeat purchase prediction and the effectiveness of marketing strategies with lower data and computing power costs, and achieve refined user operations.

[0044] Figure 1 This is a flowchart illustrating a method for in-depth insight analysis of e-commerce user behavior based on big data and artificial intelligence in one embodiment. It should be understood that, although... Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows; unless explicitly stated herein, there is no strict order requirement for the execution of these steps, and they can be executed in other orders; and Figure 1At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0045] Based on the same technical concept, referring to Figure 2 This application also provides a device for deep insight analysis of e-commerce user behavior based on big data artificial intelligence, which adopts the following technical solution: The device includes: The behavioral feature module is used to receive user historical behavior data, which includes user attribute data and user interaction data with products and merchants. The module preprocesses and performs feature engineering on the user historical behavior data to construct a multi-dimensional feature set to characterize user behavior patterns. The clustering module is used to segment users based on a multi-dimensional feature set and a clustering algorithm to obtain the results of dividing users into multiple groups with different behavioral characteristics and the corresponding group labels for each user. The repeat purchase prediction module is used to train an independent repeat purchase behavior prediction model for each user group based on the group labels, multi-dimensional feature sets and corresponding historical repeat purchase tags, and adopts a hierarchical modeling approach, and calculates the repeat purchase probability. The repeat purchase analysis module is used to analyze the repeat purchase behavior prediction model using Shapley value decomposition technology, calculate the feature contribution of each feature to the model prediction results, and identify the key factors affecting the repeat purchase behavior of different user groups. The marketing strategy module is used to generate differentiated marketing strategies for different user groups based on the segmentation results, repurchase probability from the repurchase behavior prediction model, and feature contribution.

[0046] In some embodiments, the behavior feature module is specifically used to receive user historical behavior data from the e-commerce platform, perform data cleaning on the user historical behavior data, fill or impute missing values, identify and process outliers, and standardize and transform the timestamp field. Based on the cleaned user historical behavior data, multi-dimensional features are extracted. The extracted feature dimensions include at least: user static attribute features, user dynamic behavior features, user-product interaction features, user-merchant relationship features, and time-series behavior features. The extracted multi-dimensional features are normalized and encoded to construct a multi-dimensional feature set for subsequent clustering and predictive analysis.

[0047] In some embodiments, the clustering module is specifically used to select a subset of behavioral features from a multi-dimensional feature set for clustering analysis. The subset of behavioral features includes at least core indicators that characterize user activity, purchase conversion rate, category focus, and brand loyalty. Based on a subset of behavioral features, the K-means clustering algorithm is used to perform cluster analysis on users. During the clustering process, the elbow method and silhouette coefficient are combined to determine the optimal number of clusters K. Clustering is performed based on the optimal number of clusters K, resulting in a grouping of all users into K mutually exclusive clusters, and a corresponding grouping label is assigned to each user. Analyze and output the center point or typical behavioral feature description of each user group on the behavioral feature subset.

[0048] In some embodiments, the repeat purchase prediction module is specifically used to divide the multi-dimensional feature set and the corresponding historical repeat purchase tags into subsets corresponding to each user group based on the group labels corresponding to each user group. For each user group's subset, a machine learning classification algorithm is selected as the base model. For each user segment, a base model is trained based on its corresponding subset data set, and a repurchase behavior prediction model corresponding to the segment is obtained through hyperparameter optimization. Meanwhile, during the training process, oversampling techniques are used to address the imbalance between positive and negative samples in the subsets of each cluster. Based on the trained repurchase behavior prediction model, the target user's feature data is input to obtain the predicted probability value of the user's repurchase behavior within a preset time window in the future, which is used as the repurchase probability.

[0049] In some embodiments, the repurchase prediction module is specifically used to segment each user group, determine the base model as the gradient boosting decision tree algorithm, and set a set of hyperparameters to be optimized for the gradient boosting decision tree algorithm. The set of hyperparameters includes at least the tree depth, learning rate, number of iterations, and regularization parameter. On a subset of the dataset, cross-validation is used to optimize the set of hyperparameters in order to maximize the model’s evaluation metrics on the validation set. The optimal hyperparameter combination obtained by the optimization search is configured into the gradient boosting decision tree algorithm, and the model is trained using all the data of the subset, thus obtaining a repurchase behavior prediction model corresponding to the user group with optimized parameters.

[0050] In some embodiments, the repeat purchase prediction module is specifically used to divide a subset of data into K training-validation pairs using a hierarchical K-fold cross-validation method, so as to keep the ratio of positive and negative samples in each fold consistent with the original dataset. The evaluation metric is set as the area under the receiver operating characteristic curve. Multiple hyperparameter combinations are generated within the set range of the hyperparameter set using grid search or random search strategies. For each set of hyperparameters, cross-validation is performed on K training-validation pairs, and the average AUC on the K validation sets is calculated. Based on the average AUC, the hyperparameter combination that maximizes the average AUC is selected from multiple hyperparameter combinations as the optimal hyperparameter combination.

[0051] In some embodiments, the repeat purchase analysis module is specifically used to apply Shapley value decomposition technology to the repeat purchase behavior prediction model corresponding to each user group, and calculate the contribution value of each feature in the multi-dimensional feature set to the prediction result of the corresponding group model. Aggregate the feature contribution values ​​of all user groups to generate a global feature importance ranking and identify the core features that have the greatest impact on overall repurchase prediction; By comparing and analyzing the differences in the contribution values ​​of the same feature among different user groups, key influencing factors with group specificity can be identified. Based on the segmentation results, users are divided into corresponding behavior types; based on repurchase probability, users in each segment are further divided into high, medium and low repurchase tendency intervals; and based on feature contribution, global feature importance ranking and group-specific key influencing factors, the top N key features driving repurchase behavior in each segment are identified. Based on behavior type, repurchase tendency range and corresponding top N key features, a set of targeted and differentiated marketing strategies are matched and generated from a pre-set strategy rule base. The strategy rule base includes at least a membership benefit strategy for high-value loyal users, a decision-making support strategy for potential value-seeking users, and an instant conversion strategy for price-sensitive impulsive users.

[0052] This application also discloses a control device.

[0053] Specifically, the control device includes a memory and a processor. The memory stores a computer program that can be loaded and executed by the processor to perform the aforementioned method for in-depth analysis of e-commerce user behavior based on big data and artificial intelligence.

[0054] This application also discloses a computer-readable storage medium.

[0055] Specifically, the computer-readable storage medium stores a computer program that can be loaded and executed by a processor, such as the aforementioned method for in-depth insight analysis of e-commerce user behavior based on big data artificial intelligence. The computer-readable storage medium includes, for example, various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0056] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for in-depth insight analysis of e-commerce user behavior based on big data and artificial intelligence, characterized in that, include: Receive user historical behavior data, which includes user attribute data and user interaction behavior data with products and merchants. Perform preprocessing and feature engineering on the user historical behavior data to construct a multi-dimensional feature set for characterizing user behavior patterns. Based on the multi-dimensional feature set, a clustering algorithm is used to segment users, resulting in multiple groups with different behavioral characteristics and corresponding group labels for each user. Using the aforementioned group labels as conditions, and based on the multi-dimensional feature set and corresponding historical repurchase markers, a hierarchical modeling approach is adopted to train an independent repurchase behavior prediction model for each user group and calculate the repurchase probability. Based on the repurchase behavior prediction model, the Shapley value decomposition technique is applied for analysis to calculate the feature contribution of each feature to the model prediction results, so as to identify the key factors affecting the repurchase behavior of different user groups. Based on the segmentation results, the repurchase probability of the repurchase behavior prediction model, and the feature contribution, differentiated marketing strategies are generated for different user segments.

2. The method for deep insight analysis of e-commerce user behavior based on big data artificial intelligence according to claim 1, characterized in that, The process involves receiving user historical behavior data, which includes user attribute data and user interaction data with products and merchants. This data is then preprocessed and feature-engineered to construct a multi-dimensional feature set representing user behavior patterns, including: Receive user historical behavior data from e-commerce platforms, perform data cleaning on the user historical behavior data, fill or impute missing values, identify and process outliers, and standardize and transform the timestamp field. Based on the cleaned user historical behavior data, multi-dimensional features are extracted. The extracted feature dimensions include at least: user static attribute features, user dynamic behavior features, user-product interaction features, user-merchant relationship features, and time-series behavior features. The extracted multi-dimensional features are normalized and encoded to construct the multi-dimensional feature set for subsequent clustering and predictive analysis.

3. The method for deep insight analysis of e-commerce user behavior based on big data artificial intelligence according to claim 2, characterized in that, Based on the multi-dimensional feature set, a clustering algorithm is used to segment users, resulting in multiple groups with different behavioral characteristics and corresponding group labels for each user, including: A subset of behavioral features for cluster analysis is selected from the multi-dimensional feature set. The subset of behavioral features includes at least the core indicators that characterize user activity, purchase conversion rate, category focus and brand loyalty. Based on the aforementioned subset of behavioral features, the K-means clustering algorithm is used to perform cluster analysis on users. During the clustering process, the elbow method and silhouette coefficient are combined to determine the optimal number of clusters K. Clustering is performed based on the optimal number of clusters K to obtain a grouping result that divides all users into K mutually exclusive clusters, and a corresponding grouping label is assigned to each user. Analyze and output the center point or typical behavioral feature description of each user group on the subset of behavioral features.

4. The method for deep insight analysis of e-commerce user behavior based on big data artificial intelligence according to claim 1, characterized in that, The method, using the group labels as conditions, and based on the multi-dimensional feature set and corresponding historical repurchase markers, employs a hierarchical modeling approach to train an independent repurchase behavior prediction model for each user group, and calculates the repurchase probability, including: Based on the group labels corresponding to each user, the multi-dimensional feature set and the corresponding historical repurchase tags are divided into subsets corresponding to each user group. For each user group's subset, a machine learning classification algorithm is selected as the base model. For each user segment, a base model is trained based on its corresponding subset, and the repurchase behavior prediction model corresponding to the segment is obtained through hyperparameter optimization. Meanwhile, during the training process, oversampling techniques are used to address the imbalance between positive and negative samples in the subsets of each cluster. Based on the trained repurchase behavior prediction model, the feature data of the target user is input to obtain the predicted probability value of the user's repurchase behavior within a preset time window in the future, which is used as the repurchase probability.

5. The method for deep insight analysis of e-commerce user behavior based on big data artificial intelligence according to claim 4, characterized in that, For each user segment, a base model is trained based on its corresponding subset dataset, and the repurchase behavior prediction model corresponding to the segment is obtained through hyperparameter optimization, including: For each user segment, the base model is determined to be the gradient boosting decision tree algorithm. A set of hyperparameters to be optimized is set for the gradient boosting decision tree algorithm. The set of hyperparameters includes at least the tree depth, learning rate, number of iterations, and regularization parameter. On a subset of the dataset, cross-validation is used to optimize the set of hyperparameters in order to maximize the model’s evaluation metrics on the validation set. The optimal hyperparameter combination obtained by the optimization search is configured into the gradient boosting decision tree algorithm, and the model is trained using all the data of the subset, thus obtaining the repurchase behavior prediction model corresponding to the user group and with optimized parameters.

6. The method for deep insight analysis of e-commerce user behavior based on big data artificial intelligence according to claim 4, characterized in that, The method of using cross-validation to optimize the hyperparameter set on a subset of datasets to maximize the model's evaluation metrics on the validation set includes: For the subset, a hierarchical K-fold cross-validation method is used to divide it into K training-validation pairs to maintain the ratio of positive and negative samples in each fold consistent with the original dataset. The evaluation index is set as the area under the receiver operating characteristic curve, and multiple hyperparameter combinations are generated within the set range of the hyperparameter set using grid search or random search strategies. For each set of hyperparameters, cross-validation is performed on K training-validation pairs, and the average AUC on the K validation sets is calculated. Based on the average AUC, the hyperparameter combination that maximizes the average AUC is selected from multiple hyperparameter combinations as the optimal hyperparameter combination.

7. The method for deep insight analysis of e-commerce user behavior based on big data artificial intelligence according to claim 1, characterized in that, The repurchase behavior prediction model is analyzed using Shapley value decomposition technology to calculate the feature contribution of each feature to the model prediction results, so as to identify the key factors affecting the repurchase behavior of different user groups. Based on the segmentation results, the repurchase probability of the repurchase behavior prediction model, and the feature contribution, differentiated marketing strategies are generated for different user segments, including: For each user segment corresponding to the repurchase behavior prediction model, the Shapley value decomposition technique is applied to calculate the contribution value of each feature in the multi-dimensional feature set to the prediction result of the corresponding segment model. Aggregate the feature contribution values ​​of all user groups to generate a global feature importance ranking and identify the core features that have the greatest impact on overall repurchase prediction; By comparing and analyzing the differences in the contribution values ​​of the same feature among different user groups, key influencing factors with group specificity can be identified. Based on the grouping results, users are divided into corresponding behavior types; based on the repurchase probability, users in each group are further divided into high, medium, and low repurchase tendency intervals; and based on the feature contribution, the global feature importance ranking, and group-specific key influencing factors, the top N key features driving the repurchase behavior of each group are identified. Based on the behavior type, the repurchase tendency range, and the corresponding top N key features, a set of targeted and differentiated marketing strategies are matched and generated from a preset strategy rule base. The strategy rule base includes at least a membership benefit strategy for high-value loyal users, a decision-making support strategy for potential value-seeking users, and an instant conversion strategy for price-sensitive impulsive users.

8. A device for deep insight analysis of e-commerce user behavior based on big data and artificial intelligence, characterized in that, The device includes: The behavioral feature module is used to receive user historical behavior data, which includes user attribute data and user interaction behavior data with products and merchants. The module performs preprocessing and feature engineering on the user historical behavior data to construct a multi-dimensional feature set for characterizing user behavior patterns. The clustering module is used to segment users based on the multi-dimensional feature set using a clustering algorithm, and to obtain the results of dividing users into multiple groups with different behavioral characteristics and the corresponding group labels for each user. The repeat purchase prediction module is used to train an independent repeat purchase behavior prediction model for each user group based on the group labels, the multi-dimensional feature set and the corresponding historical repeat purchase tags, and a hierarchical modeling approach, and to calculate the repeat purchase probability. The repeat purchase analysis module is used to analyze the repeat purchase behavior prediction model using Shapley value decomposition technology, calculate the feature contribution of each feature to the model prediction results, and identify the key factors affecting the repeat purchase behavior of different user groups. The marketing strategy module is used to generate differentiated marketing strategies for different user groups based on the segmentation results, the repurchase probability of the repurchase behavior prediction model, and the feature contribution.

9. A control device, characterized in that, The device includes: A memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 7.

Images