E-commerce platform information analysis method and system based on artificial intelligence

Abstract

The present invention discloses an information analysis method and system for an e-commerce platform based on artificial intelligence, which relates to the field of information analysis technology, including obtaining user behavior data based on the e-commerce platform, calculating a prior distribution based on the user behavior change rate and combining it with a preliminary likelihood distribution calculated by a probability density function, constructing an objective function to optimize and calculate short-term, medium-term, and long-term behavior windows, calculating the power spectrum density of the time series for cluster analysis, optimizing and calculating the three windows, and determining different characteristics of user behavior through kernel density estimation; and adopting a multi-head attention mechanism to fuse different characteristics of users. The method of the present invention converts the time series into frequency domain space by calculating the power spectrum density, reveals the main periodic patterns of user behavior, makes behavior prediction more interpretable, and can optimize according to the actual frequency characteristics of user behavior by constructing the frequency and power density distribution of the power spectrum density.

Description

Technical Field

[0001] The present invention relates to the field of information analysis technology, and in particular to an e-commerce platform information analysis method and system based on artificial intelligence. Background Art

[0002] With the booming development of e-commerce, user behavior analysis has become one of the key technologies for improving the user experience of e-commerce platforms and optimizing business decisions. Analysis methods based on user behavior data are constantly being improved, especially in the application of recommendation systems. Content-based recommendation methods focus on analyzing product attribute information and recommending the degree of match with user interests, but they also lack the ability to accurately capture the long-term and short-term changes in user dynamic interests. Therefore, how to deeply analyze user behavior patterns and optimize recommendation systems to better adapt to changes in user needs has become a hot topic in current research.

[0003] Although existing technical solutions have solved the problems of user behavior modeling and recommendation system optimization to a certain extent, they still have many limitations. Traditional time window division methods usually use fixed time windows to analyze user behavior changes, but this method fails to fully consider the dynamic characteristics and multi-scale changes of user behavior, resulting in insufficient flexibility and accuracy in analyzing user interests. In addition, existing recommendation systems mostly rely on single-dimensional user interest modeling, lacking comprehensive integration of different user characteristics and multi-level information extraction, resulting in limited improvement in personalized recommendation effects. Especially when dealing with short-term, medium-term and long-term user behavior, traditional methods often ignore the dynamic adjustment of behavior change rate and fail to accurately capture user interest fluctuations at different time scales. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an artificial intelligence-based e-commerce platform information analysis method to solve the problem that the dynamic characteristics and multi-scale changes of user behavior are not fully taken into account, resulting in insufficient flexibility and accuracy in analyzing user interests. In addition, existing recommendation systems mostly rely on single-dimensional user interest modeling, lack of comprehensive integration of different user characteristics and multi-level information extraction, resulting in limited improvement in personalized recommendation effects. Especially when dealing with short-term, medium-term and long-term user behaviors, traditional methods often ignore the dynamic adjustment of behavior change rates and fail to accurately capture the fluctuations of user interests at different time scales.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides an e-commerce platform information analysis method based on artificial intelligence, which comprises:

[0008] Based on e-commerce platforms, we obtain user behavior data, calculate the prior distribution based on the rate of change of user behavior, and combine it with the preliminary likelihood distribution calculated using the probability density function. We then construct an objective function to optimize the calculation of short-term, medium-term, and long-term behavior windows. We then calculate the power spectral density of the time series for cluster analysis, optimize the calculation of the three windows, and identify different characteristics of user behavior through kernel density estimation.

[0009] A multi-head attention mechanism is used to integrate different user characteristics and construct a user interest graph. Multiple regression analysis is used to calculate the partial correlation coefficient. The Peter-Clark algorithm is used to calculate the causal weight between users and products. User interest characteristics are updated through heterogeneous GNN.

[0010] Determine the user's interest distribution for products, calculate the causal effect value based on user interest characteristics, and calculate the final recommendation score;

[0011] Generate a product recommendation list, encrypt data and transmit it securely.

[0012] As a preferred embodiment of the artificial intelligence-based e-commerce platform information analysis method of the present invention, the method of determining different characteristics of user behavior includes constructing a variational Bayesian time window (VBTW) model based on user behavior data, calculating the rate of change of user behavior at different times based on the window size, segmenting the rate of change sequence into time windows, calculating the mean and variance of each rate of change window, and using a variational Bayesian method to calculate the prior distribution of the time window.

[0013] The probability density function is calculated as the initial likelihood distribution, and the product of the prior distribution and the initial likelihood distribution is used as the objective function. The objective function value of each candidate time window is calculated, and the gradient ascent method is used to optimize and maximize the objective function.

[0014] According to the branches of the objective function in different window intervals, the smallest optimal window is determined as the short-term behavior window, the largest optimal window is determined as the long-term behavior window, and the average of the sum of the long-term behavior window and the short-term behavior window is used as the medium-term behavior window;

[0015] Use the autoregressive AR model to model user behavior data, calculate the autocorrelation function (ACF) of the time series, and calculate the power spectral density through Fourier transform.

[0016] Frequency and power density are constructed based on the calculation of power spectrum density. K-Means clustering is used for cluster analysis. Long-term behavior window, short-term behavior window, and medium-term behavior window are used as clustering targets. Optimization calculation is performed by calculating the loss function. Optimization calculation is performed based on the three behavior windows as the number of clusters. The windows at the maximum frequency, minimum frequency, and medium frequency of the change frequency f are extracted respectively as the optimized short-term window, long-term window, and medium-term window;

[0017] The Gaussian kernel is used to calculate the kernel density estimation KDE of the short-term window, long-term window and medium-term window, which are used as the short-term features, medium-term features and long-term features of user behavior respectively.

[0018] As a preferred solution of the artificial intelligence-based e-commerce platform information analysis method of the present invention, wherein: the updating of user interest features includes using a Transformer with a multi-head attention mechanism to fuse the user's short-term features, medium-term features, and long-term features as comprehensive behavioral features;

[0019] Based on comprehensive behavioral characteristics as user nodes, the product text, images, and historical interaction behavior information of users as product nodes, and product category, brand, and price range information as attribute nodes, the relationships between user nodes and product nodes, product nodes and attribute nodes, and user nodes and attribute nodes are used as edges to construct a user interest graph.

[0020] Based on the edges between user nodes and product nodes, we determine comprehensive behavioral characteristics as user behavior data U, product text, image, and historical interaction behavior information as product feature data, and promotional impact and ad exposure times as control variable data F.

[0021] Multiple regression analysis is used to calculate the partial correlation coefficient, and the Peter-Clark algorithm is used to calculate the conditional partial correlation coefficient between the user node and the product node. The conditional partial correlation coefficient is used as the causal weight of the edge between the user node and the product node;

[0022] The relationship between users, products and attributes is propagated through heterogeneous GNN. The user node c and the product node v are message-transmitted through the causal weight and attention mechanism of the edge between the user node and the product node to update the user interest representation.

[0023] As a preferred solution of the artificial intelligence-based e-commerce platform information analysis method of the present invention, wherein: the calculation of the final recommendation score includes, based on the updated user interest features, using Softmax normalization to calculate the user's interest distribution for each product attribute and calculating the causal effect;

[0024] The final recommendation score is calculated by combining the causal effect value of a single product with the product of the similarity between the user's interest and the product and the ratio of the remaining products.

[0025] As a preferred solution of the artificial intelligence-based e-commerce platform information analysis method described in the present invention, the generation of a product recommendation list includes sorting the recommendation scores of all products in descending order to generate a final recommendation list, and at the same time generating a product classification recommendation list based on the attribute node information of the products including the category, brand, and price range information of the products.

[0026] As a preferred solution of the artificial intelligence-based e-commerce platform information analysis method described in the present invention, the method includes obtaining user behavior data based on the e-commerce platform, including obtaining user behavior time series data based on the e-commerce platform, and collecting user behavior data within a time window, including the number of browsing behaviors, the number of click behaviors, and the number of purchase behaviors, performing data standardization processing, and constructing a user behavior matrix, in which each row corresponds to user behavior data at different times within the time window.

[0027] As a preferred solution of the artificial intelligence-based e-commerce platform information analysis method described in the present invention, the data is encrypted and securely transmitted, including: storing recommendation list data and user interest feature data through AES-256 encryption, using RBAC to manage user permissions, and using HTTPS for secure data transmission, and adding noise to user behavior data for private storage.

[0028] In a second aspect, the present invention provides a system for an e-commerce platform information analysis method based on artificial intelligence, comprising:

[0029] User behavior data processing module, which obtains user browsing, clicking, purchasing and other behavior data from the e-commerce platform and calculates the user behavior change rate;

[0030] The time window optimization module constructs the objective function based on the prior distribution and preliminary likelihood distribution, calculates the short-term, medium-term and long-term behavior windows, and optimizes the window division through power spectral density analysis and clustering;

[0031] The user behavior feature extraction module uses kernel density estimation to calculate user behavior features in different time windows and integrates short-, medium-, and long-term behavior patterns through a multi-head attention mechanism;

[0032] The user interest graph construction module builds a heterogeneous graph containing user, product, and attribute relationships based on user behavior characteristics, and calculates the causal weight of user-product relationships;

[0033] The GNN propagation module uses multiple regression analysis to calculate partial correlation coefficients, uses the Peter-Clark algorithm to calculate causal weights, and updates user interest features through heterogeneous GNN propagation;

[0034] The recommendation score calculation module calculates the user's recommendation score for the product based on the user's interest distribution and causal effect value, performs normalization to ensure the rationality of the sorting, sorts the calculated recommendation scores, and generates a recommendation list;

[0035] The data security protection module adds noise to the data to protect privacy.

[0036] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the artificial intelligence-based e-commerce platform information analysis method as described in the first aspect of the present invention is implemented.

[0037] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the artificial intelligence-based e-commerce platform information analysis method as described in the first aspect of the present invention.

[0038] The beneficial effects of the present invention are: by calculating the power spectral density, the time series is converted into the frequency domain space, the main periodic patterns of user behavior are revealed, and the behavior prediction is made more explainable. By constructing the frequency and power density distribution of the power spectral density, it can be optimized according to the actual frequency characteristics of the user behavior, ensuring the scientificity and rationality of the behavior analysis, so that the extracted short, medium and long-term behavior windows can be calculated under the kernel density estimation. Because of the true probability distribution of the data, it avoids the assumption that user behavior must obey a certain specific distribution. By combining the causal effect value of a single product with the similarity between user interest and product, and normalizing it with the ratio of all candidate products, it can be ensured that the final recommendation score not only takes into account the causal relationship, but also the degree of match between user interest and the product itself. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0040] Figure 1 Schematic diagram of the process of the e-commerce platform information analysis method based on artificial intelligence in Example 1.

[0041] Figure 2 This is a structural diagram of the artificial intelligence-based e-commerce platform information analysis system in Example 1. DETAILED DESCRIPTION

[0042] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0043] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0044] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0045] Example 1, with reference to Figures 1 to 2 , which is the first embodiment of the present invention, provides an e-commerce platform information analysis method based on artificial intelligence, comprising the following steps:

[0046] S1, based on the e-commerce platform, obtains user behavior data, calculates the prior distribution based on the user behavior change rate, combines it with the preliminary likelihood distribution calculated by the probability density function, constructs an objective function to optimize the calculation of short-term, medium-term, and long-term behavior windows, calculates the power spectral density of the time series for cluster analysis, optimizes the calculation of the three windows, and determines the different characteristics of user behavior through kernel density estimation;

[0047] Preferably, the user behavior data is obtained based on the e-commerce platform, including obtaining the user's behavior time series data based on the e-commerce platform, and collecting the user's behavior data within the time window, including the number of browsing behaviors, the number of click behaviors and the number of purchase behaviors, performing data standardization processing, and constructing a user behavior matrix, in which each row corresponds to the user behavior data at different times within the time window.

[0048] By obtaining user behavior time series data based on the e-commerce platform and collecting user behavior data within the time window, including the number of browses, clicks and purchases, we can fully capture the changing trends of user interests, so that the recommendation system can be personalized and optimized based on the user's actual behavior patterns. Through data standardization, the differences in different user behavior scales can be eliminated, and data deviations caused by different user activity levels can be avoided, making the behavior data of different users comparable at the same scale, improving the stability and generalization ability of the model. By constructing a user behavior matrix, user behavior data can be input into the subsequent modeling and analysis process in a structured manner, ensuring that the recommendation system can efficiently process time series information, thereby improving the accuracy of recommendations and the accuracy of user interest predictions.

[0049] Preferably, determining different characteristics of user behavior includes constructing a variational Bayesian time window (VBTW) model based on user behavior data, calculating the user behavior change rate at different times based on the window size, and segmenting the change rate sequence into time windows, calculating the mean and variance of each change rate window, and using the variational Bayesian method to calculate the prior distribution of the time window, which is expressed as:

[0050] ;

[0051] in represents the prior distribution of the rate-of-change window, represents the Gaussian prior distribution calculation, represents the change rate window mean, represents the variance of the rate of change window, represents the regularization term, where Indicates the penalty factor for window growth, usually set to a small constant;

[0052] The probability density function is calculated as the initial likelihood distribution, and the product of the prior distribution and the initial likelihood distribution is used as the objective function. The objective function value of each candidate time window is calculated, and the gradient ascent method is used to optimize and maximize the objective function, which is expressed as:

[0053] ;

[0054] ;

[0055] in represents the user behavior data at time t, represents the maximum time step, represents the optimal time window function value, Indicates that under a given window W, the behavior is observed probability;

[0056] According to the branches of the objective function in different window intervals, the smallest optimal window is determined as the short-term behavior window, the largest optimal window is determined as the long-term behavior window, and the average of the sum of the long-term behavior window and the short-term behavior window is used as the medium-term behavior window;

[0057] Use the autoregressive AR model to model user behavior data (predict current behavior based on past data), calculate the autocorrelation function (ACF) of the time series, and calculate the power spectral density through Fourier transform, which is expressed as:

[0058] ;

[0059] ;

[0060] ;

[0061] in represents the autoregressive order, represents the autoregressive coefficient (confirmed by the least squares method), represents the user behavior data at time ti, represents the error term, (which can be assumed to be Gaussian noise), represents the autoregressive order at lag k, express The global mean of the series, represents the power density at different behavior change frequencies f, and j represents the imaginary unit;

[0062] Frequency and power density are constructed based on the calculation of power spectrum density. K-Means clustering is used for cluster analysis. Long-term behavior window, short-term behavior window, and medium-term behavior window are used as clustering targets. Optimization calculation is performed by calculating the loss function. Optimization calculation is performed based on the three behavior windows as the number of clusters. The windows at the maximum frequency, minimum frequency, and medium frequency of the change frequency f are extracted respectively as the optimized short-term window, long-term window, and medium-term window;

[0063] The Gaussian kernel is used to calculate the kernel density estimation KDE of the short-term window, long-term window, and medium-term window, which are respectively used as the short-term features, medium-term features, and long-term features of user behavior, expressed as:

[0064] ;

[0065] in Indicates that under a certain window W, behavioral data is observed The kernel density estimate of Indicates the number of historical data points within the time window, represents the bandwidth parameter (Silverman rule is used to determine the optimal bandwidth), K represents the kernel function, Represents the i-th historical data point in the time window.

[0066] By calculating the rate of change of user behavior, we avoid the information loss or lag problems caused by using fixed time windows. By using the variational Bayesian method to calculate the prior distribution of the time window, we can effectively avoid the instability of time window division caused by sparse data or abnormal distribution, and improve the robustness of the model. By combining prior information with the likelihood distribution of observed data, we can provide a more reasonable window division strategy, ensuring that the window selection is not interfered with by individual abnormal behaviors. The use of autoregressive (AR) models can extract the temporal dependence of user behavior, and calculating the autocorrelation function can measure the correlation of behavioral data at different lag times.

[0067] By calculating the power spectral density, the time series is transformed into the frequency domain space, revealing the main periodic patterns of user behavior, making behavior prediction more explainable. By constructing the frequency and power density distribution of the power spectral density and using K-Means clustering to optimize the time window, the time window division no longer relies on fixed rules, but can be optimized according to the actual frequency characteristics of user behavior, ensuring the scientificity and rationality of the behavior analysis. The extracted short, medium and long-term behavior windows are calculated under the kernel density estimation to be the true probability distribution of the data, avoiding the assumption that user behavior must obey a certain specific distribution.

[0068] S2 uses a multi-head attention mechanism to integrate different user features, construct a user interest graph, use multiple regression analysis to calculate the partial correlation coefficient, and use the Peter-Clark algorithm to calculate the causal weight between users and products, and update user interest features through heterogeneous GNN;

[0069] Preferably, updating the user interest features includes using a Transformer with a multi-head attention mechanism to fuse the user's short-term features, medium-term features, and long-term features as comprehensive behavioral features;

[0070] Based on comprehensive behavioral characteristics as user nodes, the product text, images, and historical interaction behavior information of users as product nodes, and product category, brand, and price range information as attribute nodes, the relationships between user nodes and product nodes, product nodes and attribute nodes, and user nodes and attribute nodes are used as edges to construct a user interest graph.

[0071] Based on the edges between user nodes and product nodes, we determine comprehensive behavioral characteristics as user behavior data U, product text, image, and historical interaction behavior information as product feature data, and promotional impact and ad exposure times as control variable data F.

[0072] Multiple regression analysis is used to calculate the partial correlation coefficient, and the Peter-Clark algorithm is used to calculate the conditional partial correlation coefficient between the user node and the product node. The conditional partial correlation coefficient is used as the causal weight of the edge between the user node and the product node, which is expressed as:

[0073] ;

[0074] ;

[0075] in represents the partial correlation coefficient among user behavior data, product feature data, and control variable data. 、 as well as They represent the covariance between user behavior and product characteristics, user behavior and a single control variable, and a single control variable and product characteristics, respectively. and represent the variance of user behavior and product characteristics respectively, represents the conditional partial correlation coefficient of user behavior data, product feature data, and control variable data, represents the number of samples, Indicates the number of control variable data;

[0076] The relationship between users, products and attributes is propagated through heterogeneous GNN. The user node c and the product node v are message-transmitted through the causal weight and attention mechanism of the edge between the user node and the product node, and the user interest representation is updated, which is expressed as:

[0077] ;

[0078] ;

[0079] in Represents the transpose calculation of the feature data of the user node, Represents the similarity between user nodes and product nodes calculated based on the trainable attention weight matrix G, Represents adjacent product nodes, represents the updated user interest features, represents the ReLU activation function, represents the adjacent product set of the user node, that is, all products that have interacted with user c. Represents the attention weight between the user node and the product node, Represents the causal weight between the user node and the product node, Represents the characteristic data of the user node, Indicates the characteristic data of a product node.

[0080] By integrating user features through the Transformer, comprehensive behavioral features are used as user nodes. Product nodes are constructed by combining product text, images, and historical interaction behavior information. Attribute nodes are constructed using product category, brand, and price range information. This allows the user interest graph to include the complete user-product-attribute ternary relationship, improving the expressiveness of the graph structure and enabling the recommendation system to maintain strong generalization capabilities even in cold start scenarios. Furthermore, by introducing product attribute information as attribute nodes, the model can learn more fine-grained product features, thereby improving the interpretability and accuracy of recommendations.

[0081] By using comprehensive behavioral features as user behavior data, product text, images, and historical interaction behaviors as product feature data, and introducing promotional influence and ad exposure times as control variables, causal reasoning can remove external interference factors, ensuring that the user interest expressions learned by the model are driven by real user preferences rather than short-term interference from external factors, thereby reducing the bias of the recommendation system and improving recommendation quality.

[0082] The calculation method of the conditional partial correlation coefficient can quantify the influence of different control variables on user behavior and product characteristics, thereby improving the long-term benefits of the recommendation system. In heterogeneous GNNs, message transmission is carried out through the causal weight and attention mechanism between users and products, thereby improving the accuracy of user interest modeling. At the same time, the attention mechanism can dynamically adjust the intensity of information dissemination according to the weight of different user-product relationships, so that the model can more accurately capture the key features of user interests.

[0083] The attention weights of user nodes and product nodes are calculated using a trainable attention weight matrix, allowing information transfer between different user-product pairs to be weighted based on similarity rather than average propagation, thereby improving the model's ability to capture changes in user interests. The introduction of causal weights enables the model to model the relationship between user behavior and product features as a true causal relationship, rather than just a simple matching relationship between covariates, thereby ensuring that the recommendation system can provide more interpretable recommendation results.

[0084] S3, determine the user's interest distribution for the product, calculate the causal effect value based on the user's interest characteristics, and calculate the final recommendation score;

[0085] Preferably, the final recommendation score is calculated by using Softmax normalization to calculate the user's interest distribution for each product attribute based on the updated user interest features. , represents the probability distribution of the user's interest in the specific attribute a in the attribute node A of the product v, and calculates the causal effect, which is expressed as:

[0086] ;

[0087] in Represents the causal effect value between users and products, It represents the probability of the product node v being purchased when user c has a certain interest a;

[0088] The final recommendation score is calculated by combining the product of the causal effect value of a single product and the similarity between the user's interest and the product with the ratio of the remaining products, which is expressed as:

[0089] ;

[0090] in represents the recommendation score of user c for product node v, represents the causal effect value between the user and the adjacent products, Represents cosine similarity calculation.

[0091] By using Softmax normalization based on updated user interest features to calculate the distribution of users' interests in various product attributes, we can quantify user preferences for different product attributes, ensuring that the recommendation system can accurately understand the user's interest structure. By calculating the causal effect value of users on products, the model not only relies on historical interaction data, but can also measure the user's potential interests and preferences to measure the user's appeal to the product, thus eliminating the interest bias caused by the recommendation system's exposure mechanism.

[0092] By combining the causal effect value of a single product with the product of the similarity between user interests and the product, and normalizing it with the ratio of all candidate products, we can ensure that the final recommendation score not only takes into account the causal relationship, but also the degree of match between user interests and the product itself. By using cosine similarity to calculate the match between user interests and product characteristics, we can ensure that the model will not lead to recommendation bias due to the high causal effect of a certain product, but will instead comprehensively consider the overall interest characteristics of the user, so that the recommendation results strike a balance between accuracy and personalization.

[0093] S4, generates a product recommendation list, encrypts the data and transmits it securely;

[0094] Preferably, generating a product recommendation list includes sorting the recommendation scores of all products in descending order to generate a final recommendation list, and generating a product classification recommendation list based on the attribute node information of the products including the category, brand, and price range information of the products.

[0095] By sorting the recommendation scores of all products in descending order, we can ensure that the final recommendation list prioritizes products that best match the user's interests, thereby improving the accuracy of recommendations and user satisfaction. Based on the attribute node information of the product, including the product category, brand and price range information, a product classification recommendation list is generated, enabling the recommendation system to provide more structured and diverse recommendation results to meet the preferences of different users.

[0096] Furthermore, the data is encrypted and securely transmitted, including AES-256 encryption to store recommendation list data and user interest feature data, using RBAC to manage user permissions, and using HTTPS for secure data transmission, and adding noise to user behavior data for private storage.

[0097] Recommendation list data and user interest feature data are stored through AES-256 encryption to ensure that sensitive data is not leaked during storage, thereby improving the data security of the system. RBAC (role-based access control) is used to manage user permissions, which can effectively limit the access scope of users of different levels to data. HTTPS (TLS encryption protocol) is used for secure data transmission to prevent data from being intercepted, tampered with, or forged during transmission. By adding noise to user behavior data for private storage, users' personal behavior patterns can be protected and their individual identities can be prevented from being speculated.

[0098] This embodiment also provides a system for an e-commerce platform information analysis method based on artificial intelligence, including:

[0099] User behavior data processing module, which obtains user browsing, clicking, purchasing and other behavior data from the e-commerce platform and calculates the user behavior change rate;

[0100] The time window optimization module constructs the objective function based on the prior distribution and preliminary likelihood distribution, calculates the short-term, medium-term and long-term behavior windows, and optimizes the window division through power spectral density analysis and clustering;

[0101] The user behavior feature extraction module uses kernel density estimation to calculate user behavior features in different time windows and integrates short-, medium-, and long-term behavior patterns through a multi-head attention mechanism;

[0102] The user interest graph construction module builds a heterogeneous graph containing user, product, and attribute relationships based on user behavior characteristics, and calculates the causal weight of user-product relationships;

[0103] The GNN propagation module uses multiple regression analysis to calculate partial correlation coefficients, uses the Peter-Clark algorithm to calculate causal weights, and updates user interest features through heterogeneous GNN propagation;

[0104] The recommendation score calculation module calculates the user's recommendation score for the product based on the user's interest distribution and causal effect value, performs normalization to ensure the rationality of the sorting, sorts the calculated recommendation scores, and generates a recommendation list;

[0105] The data security protection module adds noise to the data to protect privacy.

[0106] This embodiment also provides a computer device, which is suitable for the case of an e-commerce platform information analysis method based on artificial intelligence, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the e-commerce platform information analysis method based on artificial intelligence proposed in the above embodiment.

[0107] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0108] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the artificial intelligence-based e-commerce platform information analysis method proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0109] In summary, the present invention transforms the time series into the frequency domain space through the calculation of power spectral density, reveals the main periodic patterns of user behavior, and makes behavior prediction more interpretable. By constructing the frequency and power density distribution of power spectral density, it can be optimized according to the actual frequency characteristics of user behavior, ensuring the scientificity and rationality of behavior analysis, so that the extracted short, medium and long-term behavior windows can be calculated under the kernel density estimation. Because of the true probability distribution of the data, it avoids the assumption that user behavior must obey a certain specific distribution. By combining the causal effect value of a single product with the similarity between user interest and product, and normalizing it with the ratio of all candidate products, it can be ensured that the final recommendation score not only takes into account the causal relationship, but also the degree of match between user interest and the product itself.

[0110] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An e-commerce platform information analysis method based on artificial intelligence, characterized in that: include: Based on e-commerce platforms, we obtain user behavior data, calculate the prior distribution based on the rate of change of user behavior, and combine it with the preliminary likelihood distribution calculated using the probability density function. We then construct an objective function to optimize the calculation of short-term, medium-term, and long-term behavior windows. We then calculate the power spectral density of the time series for cluster analysis, optimize the calculation of the three windows, and identify different characteristics of user behavior through kernel density estimation. A multi-head attention mechanism is used to integrate different user characteristics and construct a user interest graph. Multiple regression analysis is used to calculate the partial correlation coefficient. The Peter-Clark algorithm is used to calculate the causal weight between users and products. User interest characteristics are updated through heterogeneous GNN. Determine the user's interest distribution for products, calculate the causal effect value based on user interest characteristics, and calculate the final recommendation score; Generate a product recommendation list, encrypt data and transmit it securely; Determining different characteristics of user behavior, This includes building a variational Bayesian time window (VBTW) model based on user behavior data, calculating the user behavior change rate at different times based on the window size, segmenting the change rate sequence into time windows, calculating the mean and variance for each change rate window, and using the variational Bayesian method to calculate the prior distribution of the time window. The probability density function is calculated as the initial likelihood distribution, and the product of the prior distribution and the initial likelihood distribution is used as the objective function. The objective function value of each candidate time window is calculated, and the gradient ascent method is used to optimize and maximize the objective function. According to the branches of the objective function in different window intervals, the smallest optimal window is determined as the short-term behavior window, the largest optimal window is determined as the long-term behavior window, and the average of the sum of the long-term behavior window and the short-term behavior window is used as the medium-term behavior window; Use the autoregressive AR model to model user behavior data, calculate the autocorrelation function (ACF) of the time series, and calculate the power spectral density through Fourier transform. Frequency and power density are constructed based on the calculation of power spectrum density. K-Means clustering is used for cluster analysis. Long-term behavior window, short-term behavior window, and medium-term behavior window are used as clustering targets. Optimization calculation is performed by calculating the loss function. Optimization calculation is performed based on the three behavior windows as the number of clusters. The windows at the maximum frequency, minimum frequency, and medium frequency of the change frequency f are extracted respectively as the optimized short-term window, long-term window, and medium-term window; The Gaussian kernel is used to calculate the kernel density estimation KDE of the short-term window, long-term window and medium-term window, which are used as the short-term features, medium-term features and long-term features of user behavior respectively.

2. The artificial intelligence-based e-commerce platform information analysis method according to claim 1, characterized in that: The updating of the user interest features includes using a Transformer with a multi-head attention mechanism to fuse the user's short-term features, medium-term features, and long-term features as comprehensive behavioral features; Based on comprehensive behavioral characteristics as user nodes, the product text, images, and historical interaction behavior information of users as product nodes, and product category, brand, and price range information as attribute nodes, the relationships between user nodes and product nodes, product nodes and attribute nodes, and user nodes and attribute nodes are used as edges to construct a user interest graph. Based on the edges between user nodes and product nodes, we determine comprehensive behavioral characteristics as user behavior data U, product text, image, and historical interaction behavior information as product feature data, and promotional impact and ad exposure times as control variable data F. Multiple regression analysis is used to calculate the partial correlation coefficient, and the Peter-Clark algorithm is used to calculate the conditional partial correlation coefficient between the user node and the product node. The conditional partial correlation coefficient is used as the causal weight of the edge between the user node and the product node; The relationship between users, products and attributes is propagated through heterogeneous GNN. The user node c and the product node v are message-transmitted through the causal weight and attention mechanism of the edge between the user node and the product node to update the user interest representation.

3. The artificial intelligence-based e-commerce platform information analysis method according to claim 2, characterized in that: Calculating the final recommendation score includes calculating the user's interest distribution for each product attribute using Softmax normalization based on the updated user interest features, and calculating the causal effect; The final recommendation score is calculated by combining the product of the causal effect value of a single product and the similarity between the user's interest and the product with the ratio of the remaining products.

4. The artificial intelligence-based e-commerce platform information analysis method according to claim 3, characterized in that: The generating of the product recommendation list includes sorting the recommendation scores of all products in descending order to generate a final recommendation list, and generating a product classification recommendation list based on the attribute node information of the products including the category, brand, and price range information of the products.

5. The method for analyzing e-commerce platform information based on artificial intelligence according to claim 4, characterized in that: The method of obtaining user behavior data based on the e-commerce platform includes obtaining user behavior time series data based on the e-commerce platform, collecting user behavior data within a time window, including the number of browsing behaviors, the number of click behaviors, and the number of purchase behaviors, performing data standardization processing, and constructing a user behavior matrix, in which each row corresponds to user behavior data at different times within the time window.

6. The artificial intelligence-based e-commerce platform information analysis method according to claim 5, characterized in that: The data is encrypted and securely transmitted, including: storing recommendation list data and user interest feature data through AES-256 encryption, managing user permissions using RBAC, and securely transmitting data using HTTPS, and adding noise to user behavior data for private storage.

7. A system for an artificial intelligence-based e-commerce platform information analysis method, based on the artificial intelligence-based e-commerce platform information analysis method according to any one of claims 1 to 6, characterized in that: include, User behavior data processing module, which obtains user behavior data from the e-commerce platform and calculates the user behavior change rate; The time window optimization module constructs the objective function based on the prior distribution and preliminary likelihood distribution, calculates the short-term, medium-term and long-term behavior windows, and optimizes the window division through power spectral density analysis and clustering; The user behavior feature extraction module uses kernel density estimation to calculate user behavior features in different time windows and integrates short-term, medium-term, and long-term behavior patterns through a multi-head attention mechanism; The user interest graph construction module builds a heterogeneous graph containing user, product, and attribute relationships based on user behavior characteristics, and calculates the causal weight of user-product relationships; The GNN propagation module uses multiple regression analysis to calculate partial correlation coefficients, uses the Peter-Clark algorithm to calculate causal weights, and updates user interest features through heterogeneous GNN propagation; The recommendation score calculation module calculates the user's recommendation score for the product based on the user's interest distribution and causal effect value, performs normalization to ensure the rationality of the sorting, sorts the calculated recommendation scores, and generates a recommendation list; The data security protection module adds noise to the data to protect privacy.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the artificial intelligence-based e-commerce platform information analysis method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the artificial intelligence-based e-commerce platform information analysis method according to any one of claims 1 to 6 are implemented.

Images