Interaction system and device for e-commerce platform

Abstract

The invention provides an interaction system and device for an e-commerce platform, and the system carries out the following steps: obtaining user behavior data and corresponding commodity information, and generating a training sample set; classifying the user behavior data in the training sample set to obtain the type of each user behavior data; collecting each user behavior data and each type of commodity information in the behavior data to train a classification neural network model to obtain a user behavior classification model; the multi-channel neural network model is correspondingly fed back and trained to obtain a commodity recommendation model; inputting the to-be-interacted user behavior data into a user behavior classification model to obtain the type of the to-be-interacted user behavior; and inputting the to-be-interacted user behavior data into a commodity recommendation model according to the type of the to-be-interacted user behavior to obtain commodity recommendation of the user behavior. The user satisfaction and the purchase conversion rate can be improved, and the user experience is optimized.

Description

Technical Field

[0001] The present invention relates to the field of e-commerce technology, and more specifically, to an interactive system and device for an e-commerce platform. Background Art

[0002] In the field of e-commerce, with the rapid development of Internet technology and the surge in the number of users, how to effectively analyze user behavior and provide personalized product recommendations has become an important means for e-commerce platforms to improve user experience and increase sales. Existing technologies usually rely on simple algorithms or rules, such as making recommendations based on users' purchase history, browsing history, etc. These methods often fail to deeply understand users' complex preferences, resulting in low accuracy and personalization of recommendation results. In addition, the diversity and complexity of user behavior data also pose challenges to behavior analysis, and traditional processing methods often fail to effectively utilize all the information of these data.

[0003] In the process of implementing the embodiments of the present invention, the inventors found that there are at least the following problems or defects in the prior art: when processing user behavior data, the existing e-commerce interactive system often ignores the diversity and dynamic changes of user behavior, resulting in the inability to accurately capture the user's real needs and preference changes. At the same time, most of the existing recommendation systems use a single recommendation model, lacking in-depth mining of user behavior types and multi-dimensional characteristics of product information, so that the accuracy and personalization of the recommendation results need to be improved. In addition, when processing large-scale user behavior data, the existing systems often face problems with computing efficiency and real-time performance, making it difficult to meet the e-commerce platform's requirements for high efficiency and real-time performance. Summary of the invention

[0004] The present invention provides an interactive system and device for an e-commerce platform.

[0005] In a first aspect of the present invention, there is provided an interactive system for an e-commerce platform, comprising: Acquire user behavior data and corresponding product information, and generate a training sample set based on the user behavior data and the corresponding product information; Classify the user behavior data in the training sample set based on user preferences to obtain the type of each user behavior data; Based on each user behavior data in the sample set and each type of product information in the behavior data, a classification neural network model is trained to obtain a user behavior classification model; The product recommendation model is obtained by training a multi-channel neural network model based on the behavior data of each user in the sample set, the product information of each type in the behavior data, and the feedback corresponding to the user behavior; The user behavior data to be interacted is input into the user behavior classification model to obtain the type of the user behavior to be interacted; and the user behavior data to be interacted is input into the product recommendation model according to the type of the user behavior to be interacted to obtain product recommendations for the user behavior.

[0006] Furthermore, the user behavior data in the training sample set are classified based on user preferences, including: for each user behavior data, the first behavior with corresponding product feedback is taken as the first category; each behavior is traversed in turn, and the preference change rate between the current behavior and the previous behavior is calculated based on the product information corresponding to the user behavior; If the preference change rate is greater than the threshold, the current behavior and the previous behavior are of different types; otherwise, the current behavior and the previous behavior are of the same type.

[0007] Furthermore, the preference change rate includes the price change rate and category change rate ; Calculating the preference change rate between the current behavior and the previous behavior based on the product information corresponding to the user behavior, including: calculating the statistical characteristics of the preference corresponding to each product information of the user behavior; The rate of change in preference between the current action and the previous action is calculated using the following formula: Price Change Rate Category change rate in, and Respectively represent the commodity prices corresponding to the current behavior and the previous behavior, and Respectively represent the product categories corresponding to the current behavior and the previous behavior.

[0008] Furthermore, the number of input channels of the multi-channel neural network model is the same as the number of behavior types; A product recommendation model is obtained by training a multi-channel neural network model based on each user behavior data in a sample set, each type of product information in the behavior data, and feedback corresponding to the user behavior, including: inputting behaviors of the same type in each user behavior data into the same input channel to train the multi-channel neural network model.

[0009] Furthermore, the loss of the multi-channel neural network model is calculated using the following formula: in, The model predicts the The product recommendation matrix corresponding to each behavior type, Indicates The gold standard product recommendation matrix corresponding to each behavior type, represents the number of behavior categories, and represents the weight parameter, represents the regularization parameter, Represents the total sample loss.

[0010] Further, generating a training sample set based on the user behavior data and the corresponding product information includes: resampling the user behavior data and the corresponding product information to standard voxels using a linear interpolation method; Normalize the voxel values ​​of user behavior data; The normalized user behavior data is segmented to obtain a user behavior mask, the minimum circumscribed cuboid of the user behavior mask is calculated, the user behavior data and the corresponding product information in the minimum circumscribed cuboid are extracted, and a training sample set is generated.

[0011] Furthermore, the multi-channel neural network model includes a multi-channel convolution layer, a downsampling convolution module and an upsampling convolution module connected in sequence; The multi-channel convolution layer is used to extract a multi-channel feature image from multiple input channels by using convolution; the downsampling convolution module is used to extract features of different levels from the multi-channel feature image, and the upsampling convolution module is used to upsample the extracted features and output the recommendation results; The downsampling convolution module includes a plurality of downsampling convolution units, each of which includes a three-dimensional convolution layer, a LeakyReLU layer, a batch normalization layer, and a maximum pooling layer connected in sequence; The upsampling convolution module includes multiple upsampling convolution units, each of which includes a three-dimensional convolution layer, a LeakyReLU layer, a batch normalization layer and a transposed convolution layer connected in sequence.

[0012] In a second aspect of the present invention, there is provided an interactive device for an e-commerce platform, comprising: A sample set generation module, used to obtain user behavior data and corresponding commodity information, and generate a training sample set based on the user behavior data and the corresponding commodity information; A behavior classification module is used to classify the user behavior data in the training sample set based on user preferences to obtain the type of each user behavior data; A classification model training module is used to train a classification neural network model based on each type of commodity information in each user behavior data in the sample set to obtain a user behavior classification model; A recommendation model training module is used to train a multi-channel neural network model based on each user behavior data in the sample set, each type of product information in the behavior data, and feedback corresponding to the user behavior to obtain a product recommendation model; The automatic interaction module is used to input the user behavior data to be interacted into the user behavior classification model to obtain the type of the user behavior to be interacted; and input the user behavior data to be interacted into the product recommendation model according to the type of the user behavior to be interacted to obtain product recommendations for the user behavior.

[0013] Furthermore, the behavior classification module classifies the user behavior data in the training sample set by adopting the following steps: for each user behavior data, the first behavior with corresponding product feedback is taken as the first category; each behavior is traversed in turn, and the preference change rate between the current behavior and the previous behavior is calculated based on the product information corresponding to the user behavior; If the preference change rate is greater than the threshold, the current behavior and the previous behavior are of different types; otherwise, the current behavior and the previous behavior are of the same type.

[0014] Furthermore, the preference change rate includes the price change rate and category change rate ; Calculating the preference change rate between the current behavior and the previous behavior based on the product information corresponding to the user behavior, including: calculating the statistical characteristics of the preference corresponding to each product information of the user behavior; The rate of change in preference between the current action and the previous action is calculated using the following formula: Price Change Rate Category change rate in, and Respectively represent the commodity prices corresponding to the current behavior and the previous behavior, and Respectively represent the product categories corresponding to the current behavior and the previous behavior.

[0015] The above-mentioned embodiments of the present invention have at least the following beneficial effects: the e-commerce platform interactive system can conduct in-depth analysis and classification of user behavior data by adopting advanced machine learning technology, so as to more accurately identify the user's behavior type and preference changes. The system uses a multi-channel neural network model to comprehensively consider multiple dimensions of user behavior, including price, category, etc., to provide users with more accurate product recommendations. In addition, the system can dynamically adjust the recommendation strategy by updating user behavior data and product information in real time to adapt to changes in user preferences, thereby improving the relevance of recommendations and user satisfaction.

[0016] Through this method, the system can not only improve the user's shopping experience and increase user stickiness, but also improve the conversion rate of goods through accurate recommendations, bringing higher economic benefits to the e-commerce platform. At the same time, the system's automated interaction mechanism can reduce manual intervention, reduce operating costs, and improve the efficiency of processing large-scale data, so that the e-commerce platform can better respond to market changes and the rapid growth of user demand. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, in which: Figure 1 A schematic diagram of a flow chart of an e-commerce platform interaction system provided by an embodiment of the present invention; Figure 2 A schematic diagram of the structure of an e-commerce platform interaction device provided by an embodiment of the present invention; Figure 3 The schematic diagram schematically shows the structure of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0019] Those skilled in the art know that the embodiments of the present invention can be implemented as a system, device, apparatus, method or computer program product. Therefore, the present invention can be specifically implemented in the following forms, namely: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0020] It should be noted that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction and does not have any limiting meaning.

[0021] Reference below Figure 1 , Figure 1 This is a flow chart of an e-commerce platform interaction system provided by an embodiment of the present invention. Figure 1 As shown, an interactive system 100 for an e-commerce platform includes: Step 101, obtaining user behavior data and corresponding product information, and generating a training sample set based on the user behavior data and the corresponding product information; Step 102, classifying the user behavior data in the training sample set based on user preferences to obtain the type of each user behavior data; Step 103, training a classification neural network model based on each user behavior data in the sample set and each type of commodity information in the behavior data to obtain a user behavior classification model; Step 104, training a multi-channel neural network model based on each user behavior data in the sample set, each type of product information in the behavior data, and feedback corresponding to the user behavior to obtain a product recommendation model; Step 105 , input the user behavior data to be interacted into the user behavior classification model to obtain the type of the user behavior to be interacted; input the user behavior data to be interacted into the product recommendation model according to the type of the user behavior to be interacted to obtain product recommendations for the user behavior.

[0022] It should be noted that the system first obtains user behavior data and corresponding product information. These data include the user's browsing history, purchase history, click behavior, etc., as well as product details related to these behaviors, such as price, category, description, etc.

[0023] Specifically, user behavior data can be collected through log files, database queries, etc., while product information comes from the product database of the e-commerce platform. In the process of generating training sample sets, the system will use certain data preprocessing techniques, such as data cleaning and denoising, to ensure the quality and availability of the data. In addition, the system will use some feature engineering techniques, such as feature selection and feature extraction, to enhance the performance of the model.

[0024] Preferably, when processing user behavior data, the system will use some advanced data processing technologies, such as sequence models in deep learning, to capture the temporal characteristics of user behavior. At the same time, the system will introduce some regularization technologies, such as L1 and L2 regularization, to prevent model overfitting and improve the generalization ability of the model. In addition, the system will also use some optimization algorithms, such as stochastic gradient descent SGD, Adam, etc., to accelerate the training process of the model.

[0025] It should be noted that the system described in this embodiment adopts a method based on user preferences when classifying user behavior data, which includes classifying user behavior data according to certain rules to identify user behavior patterns and preference changes.

[0026] Specifically, the system will define a threshold to determine whether the preference change between user behaviors is significant. If the preference change rate of two consecutive behaviors exceeds this threshold, they are considered to be of different types. The calculation of the preference change rate involves multiple dimensions such as price change rate and category change rate, which can be quantified through statistical methods.

[0027] More specifically, the system uses some advanced mathematical formulas and algorithms to calculate the preference change rate. For example, the price change rate is calculated using the relative change in price, while the category change rate takes into account the similarity of product categories. In addition, the system also introduces some machine learning techniques, such as cluster analysis, to automatically determine the classification of behavior types.

[0028] It should be noted that the multi-channel neural network model described in this embodiment is the core component of the system for implementing product recommendations. The model can process data from multiple input channels and output corresponding recommendation results.

[0029] Specifically, the number of input channels of the multi-channel neural network model will match the number of types of user behavior. During the training process, the system will input the same type of user behavior data into the corresponding input channel to train the model to identify and recommend the corresponding type of products.

[0030] More specifically, the system uses some advanced neural network structures, such as downsampling convolution modules and upsampling convolution modules, to extract and fuse multi-channel features. These structures can help the model better understand the complex patterns in user behavior data and improve the accuracy of recommendation results. In addition, the system also uses some regularization techniques, such as Dropout, to prevent the model from overfitting.

[0031] In some embodiments, classifying user behavior data in a training sample set based on user preferences includes: for each user behavior data, taking the first behavior with corresponding product feedback as the first category; traversing each behavior in turn, and calculating the preference change rate between the current behavior and the previous behavior based on the product information corresponding to the user behavior; If the preference change rate is greater than the threshold, the current behavior and the previous behavior are of different types; otherwise, the current behavior and the previous behavior are of the same type.

[0032] It should be noted that the e-commerce platform interaction system in this embodiment pays special attention to the changes in user preferences when classifying user behavior data. The system analyzes the product information in the user behavior data, calculates the change rate of user preferences, and thus divides user behaviors into different types.

[0033] Specifically, the change rate of user preferences includes price change rate and category change rate. Price change rate refers to the difference in the price of the goods that the user pays attention to in two consecutive behaviors; category change rate refers to the change in the category of the goods that the user pays attention to. These two indicators can help the system understand the dynamic changes of user preferences and classify user behaviors more accurately.

[0034] Preferably, the system uses a specific mathematical formula when calculating the price change rate and category change rate. For example, the price change rate can be obtained by calculating the relative change of the commodity price in two consecutive behaviors, while the category change rate is calculated by comparing the similarity of commodity categories. In addition, the system sets a threshold value, and when the change rate exceeds this threshold value, it is considered that the user's behavior type has changed.

[0035] It should be noted that the system in this embodiment uses a threshold-based classification method when processing user behavior data. This method determines whether user behaviors belong to the same type by setting a threshold.

[0036] Specifically, the system will set this threshold based on historical data and the statistical characteristics of user behavior. For example, the system will calculate the historical average of the price change rate and category change rate in user behavior, and then use this average as a reference for the threshold.

[0037] More specifically, the system will use some machine learning techniques to dynamically adjust the threshold. For example, the system will use online learning methods to update the threshold in real time based on the user's latest behavior data to adapt to changes in user preferences.

[0038] It should be noted that the system in this embodiment particularly considers the relationship between user behavior and product information when classifying user behavior. The system calculates the change rate of user preference by analyzing the product information corresponding to the user behavior.

[0039] Specifically, the system collects detailed information from user behavior data, such as information about products that users browse, click, and purchase, and then calculates the price change rate and category change rate based on this information.

[0040] Preferably, the system will use some advanced data processing techniques when calculating these change rates. For example, the system will use time series analysis to process price change rates and cluster analysis to process category change rates. In addition, the system will also use some statistical methods, such as variance analysis, to evaluate the significance of the change rate.

[0041] In some embodiments, the preference change rate includes a price change rate and category change rate ; Calculating the preference change rate between the current behavior and the previous behavior based on the product information corresponding to the user behavior, including: calculating the statistical characteristics of the preference corresponding to each product information of the user behavior; The rate of change in preference between the current action and the previous action is calculated using the following formula: Price Change Rate Category change rate in, and Respectively represent the commodity prices corresponding to the current behavior and the previous behavior, and Respectively represent the product categories corresponding to the current behavior and the previous behavior.

[0042] It should be noted that the e-commerce platform interaction system in this embodiment pays special attention to the changes in the two dimensions of price and category when calculating the change rate of user preferences. The system can more accurately capture the subtle changes in user preferences by accurately calculating the change rates of these two dimensions.

[0043] Specifically, the price change rate is calculated by calculating the relative change in the price of the product that the user pays attention to in two consecutive behaviors. The category change rate is calculated by comparing the difference in the categories of the products that the user pays attention to in two consecutive behaviors. The calculation of these two change rates can help the system understand the user's behavior patterns and preference changes in more detail.

[0044] Preferably, the system uses a specific mathematical formula when calculating the price change rate and category change rate. The price change rate calculation formula is: The formula for calculating the category change rate is: in, and Respectively represent the commodity prices corresponding to the current behavior and the previous behavior, and Respectively represent the product categories corresponding to the current behavior and the previous behavior.

[0045] Specifically, the system collects detailed information from user behavior data, such as the products that users browse, click, and purchase, and then calculates the price change rate and category change rate based on this information. These statistical features include the price range and category distribution of the products.

[0046] Furthermore, the system uses some advanced data processing techniques when calculating these rates of change. For example, the system uses time series analysis to process price rates of change and cluster analysis to process category rates of change. In addition, the system uses some statistical methods, such as variance analysis, to evaluate the significance of the rate of change.

[0047] More specifically, the system will dynamically adjust the price change rate and category change rate calculation formula based on the user's real-time behavior data. For example, the system will adjust the weight of the change rate based on the frequency and intensity of user behavior to better reflect changes in user preferences.

[0048] Furthermore, the system will use some machine learning techniques to achieve this dynamic adjustment. For example, the system will use reinforcement learning algorithms to optimize the calculation formula of the change rate, so that the system can automatically adjust the calculation strategy based on user feedback, thereby more accurately capturing changes in user preferences.

[0049] In some embodiments, the number of input channels of the multi-channel neural network model is the same as the number of behavior types; A product recommendation model is obtained by training a multi-channel neural network model based on each user behavior data in a sample set, each type of product information in the behavior data, and feedback corresponding to the user behavior, including: inputting behaviors of the same type in each user behavior data into the same input channel to train the multi-channel neural network model.

[0050] It should be noted that the e-commerce platform interaction system in this embodiment uses a multi-channel neural network model to process user behavior data, and the number of input channels of this model matches the number of types of user behavior. This design enables the system to conduct in-depth analysis and learning for different types of user behavior.

[0051] Specifically, a multi-channel neural network model is a deep learning model that can process data from multiple input sources. In this system, each input channel specializes in processing a specific type of user behavior data. For example, if the user behavior data includes browsing, clicking, and purchasing, then the model will have three corresponding input channels.

[0052] Preferably, when designing a multi-channel neural network model, the system will use different network structures to adapt to different types of user behavior data. For example, for behavior data containing time series information, a recurrent neural network structure such as LSTM or GRU will be used; while for image or text behavior data, a convolutional neural network CNN or Transformer structure will be used.

[0053] Specifically, the system uses a special data preprocessing method to group user behavior data by type and then input them into the corresponding channels of the model. For example, all click behavior data will be input into the first channel, all browsing behavior data will be input into the second channel, and so on.

[0054] Preferably, the system will use some advanced data processing techniques to enhance the performance of the model. For example, data enhancement techniques will be used to generate more training samples, or regularization techniques will be used to prevent the model from overfitting. In addition, the system will also use some optimization algorithms, such as Adam or RMSprop, to speed up the model training process.

[0055] It should be noted that the system in this embodiment will consider the complexity of user behavior data and product information when training the multi-channel neural network model. The system will use this information to train the model so that the model can better understand and predict user behavior.

[0056] Specifically, the system uses a complex network structure, such as a deep convolutional neural network, to process high-dimensional user behavior data and product information. This network structure can automatically extract key features from the data and learn the complex relationships between the data.

[0057] Preferably, the system will use some advanced feature fusion techniques to integrate information from different channels. For example, an attention mechanism will be used to dynamically adjust the importance of different channels, or a gating mechanism will be used to control the flow of information. These technologies can help the model learn and predict user behavior more effectively.

[0058] In some embodiments, the loss of the multi-channel neural network model is calculated using the following formula: in, The model predicts the The product recommendation matrix corresponding to each behavior type, Indicates The gold standard product recommendation matrix corresponding to each behavior type, represents the number of behavior categories, and represents the weight parameter, represents the regularization parameter, Represents the total sample loss.

[0059] Specifically, the calculation of the loss function involves the difference between the product recommendation matrix predicted by the model and the gold standard product recommendation matrix. This difference is quantified by calculating the error between the predicted matrix and the true matrix. The smaller the error, the better the prediction performance of the model.

[0060] Preferably, the system will use some advanced error measurement methods, such as mean square error MSE or cross entropy loss, to calculate the error between the predicted matrix and the true matrix. In addition, the system will also introduce some regularization terms, such as L1 or L2 regularization, to prevent the model from overfitting.

[0061] It should be noted that the loss function in this embodiment includes multiple components, each of which is targeted at different prediction behavior types of the model. Such a design enables the system to optimize the prediction performance of the model for different types of user behaviors.

[0062] Specifically, the loss function will contain multiple terms, each corresponding to a behavior type. For example, if the system needs to predict three different types of user behavior, then the loss function will have three corresponding terms. Each term will calculate the error of that type of prediction and include it in the calculation of the total loss.

[0063] Preferably, the system will use some weight parameters to balance the importance of different behavior types. For example, if the prediction of certain behavior types is more critical to the final recommendation result, then the loss items of these behavior types will be given higher weights.

[0064] It should be noted that the loss function in this embodiment also takes into account the regularization term to improve the generalization ability of the model. The regularization term can help the model avoid overfitting, so as to maintain good prediction performance on new and unseen data.

[0065] More specifically, the regularization term includes a penalty on the model parameters, such as the L1 norm or L2 norm of the parameters. These penalty terms can limit the size of the model parameters, thereby preventing the model from becoming too complex.

[0066] Furthermore, the system will use some advanced regularization techniques, such as Dropout or BatchNormalization, to further improve the generalization ability of the model. These techniques can dynamically adjust the complexity of the model during the training process, thereby improving model performance while avoiding overfitting.

[0067] In some embodiments, generating a training sample set based on the user behavior data and the corresponding product information includes: resampling the user behavior data and the corresponding product information to standard voxels using a linear interpolation method; Normalize the voxel values ​​of user behavior data; The normalized user behavior data is segmented to obtain a user behavior mask, the minimum circumscribed cuboid of the user behavior mask is calculated, the user behavior data and the corresponding product information in the minimum circumscribed cuboid are extracted, and a training sample set is generated.

[0068] Specifically, resampling is to adjust the user behavior data and the corresponding product information to a unified standard voxel size through the linear interpolation method. Normalization is to adjust the numerical range of these data to a specific interval, such as 0 to 1, to eliminate the numerical differences between different features. Image segmentation is to segment the voxel values ​​in the user behavior data to identify the key areas of user behavior.

[0069] Preferably, the system will use some advanced image processing techniques to improve the accuracy of data preprocessing. For example, more complex interpolation methods, such as cubic spline interpolation, can be used to improve the quality of resampling. At the same time, the normalization process will use different methods, such as minimum-maximum normalization or Z-score normalization, to adapt to different types of data distribution.

[0070] It should be noted that the system in this embodiment pays special attention to the voxel values ​​of the user behavior data when generating the training sample set. By normalizing the voxel values ​​and performing image segmentation, the system can more accurately identify the key features of the user behavior.

[0071] Specifically, the normalization of voxel values ​​can be performed using the following formula: Image segmentation is achieved by identifying different regions in voxel values ​​and calculating the minimum circumscribed cuboid of the user behavior mask.

[0072] Preferably, the system will use some advanced image segmentation algorithms, such as threshold-based segmentation or region-based segmentation, to improve the accuracy of segmentation. In addition, the system will also use some morphological operations, such as erosion and dilation, to optimize the quality of the user behavior mask.

[0073] It should be noted that when the system in this embodiment generates a training sample set, it also extracts the user behavior data and corresponding product information in the smallest circumscribed cuboid of the user behavior mask. This step helps the system to more accurately locate the key areas of user behavior and use them for model training.

[0074] Specifically, the minimum circumscribed cuboid can be obtained by calculating the bounding box of the user behavior mask. Extracting user behavior data and product information in this area can ensure that the data in the training sample set is the most representative and informative.

[0075] Preferably, the system will use some advanced geometric processing techniques to optimize the calculation of the minimum circumscribed cuboid. For example, a convex hull algorithm is used to determine the boundaries of the user behavior mask, or a more complex geometric algorithm is used to improve the accuracy of the bounding box. In addition, the system will also introduce some machine learning techniques, such as cluster analysis, to automatically identify and extract key user behavior areas.

[0076] In some embodiments, the multi-channel neural network model includes a multi-channel convolution layer, a downsampling convolution module, and an upsampling convolution module connected in sequence; The multi-channel convolution layer is used to extract a multi-channel feature image from multiple input channels by using convolution; the downsampling convolution module is used to extract features of different levels from the multi-channel feature image, and the upsampling convolution module is used to upsample the extracted features and output the recommendation results; The downsampling convolution module includes a plurality of downsampling convolution units, each of which includes a three-dimensional convolution layer, a LeakyReLU layer, a batch normalization layer, and a maximum pooling layer connected in sequence; The upsampling convolution module includes multiple upsampling convolution units, each of which includes a three-dimensional convolution layer, a LeakyReLU layer, a batch normalization layer and a transposed convolution layer connected in sequence.

[0077] Specifically, the multi-channel convolution layer can process data with multiple input channels, each channel corresponding to a type of user behavior data. The downsampling convolution module is used to extract features at different levels, while the upsampling convolution module is used to upsample the extracted features in order to output the final recommendation results.

[0078] Preferably, the system will use some advanced convolutional neural network technology to optimize the performance of the model. For example, deep separable convolution is used to reduce the number of model parameters and computational complexity. At the same time, the system will also introduce an attention mechanism to enhance the model's ability to recognize key features.

[0079] It should be noted that the multi-channel convolution layer in this embodiment is used to extract multi-channel feature images from multiple input channels. These feature images contain important information of user behavior data and provide a basis for subsequent recommendations.

[0080] Specifically, the multi-channel convolution layer uses convolution kernels of different sizes to capture features of different scales. In addition, the system sets different numbers of channels to adapt to different types of user behavior data.

[0081] Furthermore, the system will use some regularization techniques, such as Dropout or Batch Normalization, to improve the generalization ability of the model. At the same time, the system will also use some optimization algorithms, such as Adam or RMSprop, to accelerate the training process of the model.

[0082] It should be noted that the downsampling convolution module in this embodiment includes multiple downsampling convolution units, each of which is composed of a three-dimensional convolution layer, a LeakyReLU layer, a batch normalization layer, and a maximum pooling layer connected in sequence. Such a design helps the model extract deep features from multi-channel feature images.

[0083] Specifically, the 3D convolutional layer is able to process data with three dimensions, the LeakyReLU layer is used to introduce nonlinearity, the batch normalization layer helps to speed up the training process, and the maximum pooling layer is used to reduce the spatial dimension of the features.

[0084] More specifically, the system will use some advanced pooling techniques, such as adaptive pooling, to better capture the dynamic range of features. At the same time, the system will also introduce some advanced activation functions, such as ELU or SELU, to improve the nonlinear expression ability of the model.

[0085] The above-mentioned embodiments of the present invention have the following beneficial effects: the e-commerce platform interaction system of the present invention can effectively integrate user behavior data and product information, generate training sample sets, and realize personalized product recommendations through user behavior classification models and product recommendation models. The system can automatically learn and adjust recommendation strategies based on user behavior data and feedback, thereby providing users with more accurate and personalized product recommendations. In addition, by adopting classification neural network models and multi-channel neural network models, the system can more deeply mine the potential information in user behavior data and improve the performance and accuracy of the recommendation system.

[0086] The e-commerce platform interactive system can improve the user interaction experience and enhance user satisfaction and loyalty. At the same time, the automation and intelligent features of the system can reduce the operating costs of the e-commerce platform and improve operational efficiency. The system can also quickly respond to market changes and user needs by analyzing and processing user behavior data in real time, bringing higher commercial value and competitive advantages to the e-commerce platform.

[0087] like Figure 2 As shown, in some embodiments, an interactive device 200 for an e-commerce platform includes: A sample set generation module 201 is used to obtain user behavior data and corresponding product information, and generate a training sample set based on the user behavior data and the corresponding product information; A behavior classification module 202 is used to classify the user behavior data in the training sample set based on user preferences to obtain the type of each user behavior data; A classification model training module 203 is used to train a classification neural network model based on each type of commodity information in each user behavior data in the sample set to obtain a user behavior classification model; The recommendation model training module 204 is used to train a multi-channel neural network model based on each user behavior data in the sample set, each type of product information in the behavior data, and feedback corresponding to the user behavior to obtain a product recommendation model; The automatic interaction module 205 is used to input the user behavior data to be interacted into the user behavior classification model to obtain the type of the user behavior to be interacted; and input the user behavior data to be interacted into the product recommendation model according to the type of the user behavior to be interacted to obtain product recommendations for the user behavior.

[0088] It is understandable that the modules recorded in the e-commerce platform interaction device 200 and the reference Figure 1 Therefore, the operations, features and beneficial effects described above for the e-commerce platform interaction system are also applicable to the e-commerce platform interaction device 200 and the modules contained therein, and will not be repeated here.

[0089] In some embodiments, the behavior classification module classifies the user behavior data in the training sample set using the following steps: for each user behavior data, the first behavior with corresponding product feedback is taken as the first category; each behavior is traversed in turn, and the preference change rate between the current behavior and the previous behavior is calculated based on the product information corresponding to the user behavior; If the preference change rate is greater than the threshold, the current behavior and the previous behavior are of different types; otherwise, the current behavior and the previous behavior are of the same type.

[0090] It should be noted that the behavior classification module of the e-commerce platform interaction device in this embodiment adopts a series of steps to classify the user behavior data in the training sample set. This process involves grouping the user behavior data according to its relevance to product feedback to identify different user behavior types.

[0091] Specifically, the behavior classification module first marks the first behavior with corresponding product feedback in each user behavior data as the first category. Then, the module will traverse each behavior in turn and calculate the preference change rate between the current behavior and the previous behavior based on the product information corresponding to the user behavior.

[0092] Preferably, the system will use some advanced data analysis techniques to improve the accuracy of behavior classification, for example, cluster analysis to automatically identify patterns in behavior data, or time series analysis to consider the time dependency of behavior data.

[0093] It should be noted that the behavior classification module in this embodiment will consider the price change rate and category change rate when calculating the preference change rate. These change rates are calculated based on the product information corresponding to the user behavior and are used to measure the dynamic changes of user preferences.

[0094] Specifically, the calculation of price change rate and category change rate involves complex mathematical formulas. For example, the price change rate is calculated by comparing the commodity prices corresponding to the current behavior and the previous behavior, while the category change rate is obtained by comparing the category identifiers of the commodities.

[0095] Preferably, the system will introduce some machine learning algorithms to optimize the calculation process of the change rate. For example, support vector machines (SVMs) are used to determine the thresholds of price and category changes, or neural networks are used to learn the relationship between the change rate and the user behavior type.

[0096] It should be noted that the behavior classification module in this embodiment sets a threshold to determine whether the preference change rate is significant when determining the user behavior type. If the change rate exceeds this threshold, the current behavior and the previous behavior are considered to be of different types; otherwise, they are of the same type.

[0097] Specifically, the threshold setting will be based on the statistical analysis of historical data. The system will calculate the median or average of the historical preference change rate and use it as the initial setting of the threshold.

[0098] Furthermore, the system will use some adaptive methods to dynamically adjust the threshold. For example, the threshold can be adjusted according to the real-time changes in user behavior data, or online learning algorithms can be used to update the threshold in real time to adapt to the dynamic changes in user behavior. In addition, the system will also introduce some anomaly detection mechanisms to process behavior data that does not conform to the normal pattern.

[0099] In some embodiments, the preference change rate includes a price change rate and category change rate ; Calculating the preference change rate between the current behavior and the previous behavior based on the product information corresponding to the user behavior, including: calculating the statistical characteristics of the preference corresponding to each product information of the user behavior; The rate of change in preference between the current action and the previous action is calculated using the following formula: Price Change Rate Category change rate in, and Respectively represent the commodity prices corresponding to the current behavior and the previous behavior, and Respectively represent the product categories corresponding to the current behavior and the previous behavior.

[0100] It should be noted that the behavior classification module of the e-commerce platform interaction device in this embodiment pays special attention to the two key indicators of price change rate and category change rate when calculating the preference change rate of user behavior data. These two indicators help the module to more accurately identify the type of user behavior, thereby providing more accurate user preference information for subsequent product recommendations.

[0101] Specifically, the price change rate refers to the difference in the price of the product that the user is interested in between two consecutive behaviors. The category change rate refers to the difference in the category of the products that the user is interested in. The calculation of these two indicators is based on the product information corresponding to the user behavior, such as price, category and other attributes.

[0102] Preferably, the system will use some advanced mathematical models and algorithms to calculate these two change rates, for example, using statistical methods to analyze the historical change trends of prices and categories, or using machine learning algorithms to predict users' sensitivity to price and category changes.

[0103] It should be noted that the preference change rate calculation in this embodiment includes the calculation of the statistical characteristics of the preference corresponding to each commodity information of the user behavior. This means that the system not only considers the changes in price and category, but also considers other commodity attributes such as brand, evaluation, etc., to obtain more comprehensive user preference information.

[0104] Specifically, the system uses feature extraction technology to identify and quantify user preferences for different product attributes. For example, natural language processing technology is used to analyze user feedback on product descriptions, or image recognition technology is used to analyze user preferences for product images.

[0105] Preferably, the system will use some advanced data analysis techniques to improve the accuracy of preference change rate calculation, for example, using deep learning algorithms to automatically learn complex patterns of user preferences, or using reinforcement learning algorithms to optimize the prediction model of user preferences.

[0106] It should be noted that the system in this embodiment uses specific formulas to quantify the price change rate and category change rate when calculating the preference change rate. These formulas provide a standardized method to measure changes in user preferences, thereby helping the system to more accurately distinguish different types of user behavior.

[0107] Furthermore, the calculation formulas for price change rate and category change rate involve complex mathematical operations, such as ratio calculation, normalization, etc. These formulas are designed to ensure that the calculation of the change rate is both accurate and comparable.

[0108] Furthermore, the system will use some advanced mathematical optimization techniques to improve these formulas. For example, numerical analysis methods are used to optimize the parameters in the formulas, or machine learning techniques are used to automatically adjust the formulas to adapt to different user behavior patterns. In addition, the system will also introduce some regularization terms to prevent overfitting problems in the calculation process.

[0109] Reference below Figure 3 , which shows a schematic diagram of a structure 300 of an electronic device suitable for implementing some embodiments of the present invention. The electronic devices in some embodiments of the present invention may include, but are not limited to, mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 3 The terminal device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0110] like Figure 3 As shown, the electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 to a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0111] Typically, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Figure 3 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 3 Each block shown in the figure may represent one device, or may represent multiple devices as required.

[0112] Furthermore, the storage medium of the embodiment of the present application stores program instructions that can implement all the above methods, wherein the program instructions can be stored in the above storage medium in the form of a software product, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or terminal devices such as a computer, a server, a mobile phone, and a tablet.

[0113] The above descriptions are only some preferred embodiments of the present invention and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention.

Claims

1. An interactive system for an e-commerce platform, characterized in that: The system performs the following steps: Acquire user behavior data and corresponding product information, and generate a training sample set based on the user behavior data and the corresponding product information; Classify the user behavior data in the training sample set based on user preferences to obtain the type of each user behavior data; Based on each user behavior data in the sample set and each type of product information in the behavior data, a classification neural network model is trained to obtain a user behavior classification model; The product recommendation model is obtained by training a multi-channel neural network model based on the behavior data of each user in the sample set, the product information of each type in the behavior data, and the feedback corresponding to the user behavior; Inputting the behavior data of the user to be interacted into the user behavior classification model to obtain the type of the user behavior to be interacted; According to the type of the user behavior to be interacted, the user behavior data to be interacted is input into the product recommendation model to obtain product recommendations for the user behavior.

2. The interactive system for an e-commerce platform according to claim 1, characterized in that: Classifying the user behavior data in the training sample set based on user preferences, including: for each user behavior data, taking the first behavior with corresponding product feedback as the first category; traversing each behavior in turn, and calculating the preference change rate between the current behavior and the previous behavior based on the product information corresponding to the user behavior; If the preference change rate is greater than the threshold, the current behavior and the previous behavior are of different types; otherwise, the current behavior and the previous behavior are of the same type.

3. The interactive system for an e-commerce platform according to claim 2, characterized in that: The preference change rate includes the price change rate and category change rate ; Calculating the preference change rate between the current behavior and the previous behavior based on the product information corresponding to the user behavior, including: calculating the statistical characteristics of the preference corresponding to each product information of the user behavior; The rate of change in preference between the current action and the previous action is calculated using the following formula: Price Change Rate Category change rate in, and Respectively represent the commodity prices corresponding to the current behavior and the previous behavior, and Respectively represent the product categories corresponding to the current behavior and the previous behavior.

4. The interactive system for an e-commerce platform according to claim 1, characterized in that: The number of input channels of the multi-channel neural network model is the same as the number of behavior types; A product recommendation model is obtained by training a multi-channel neural network model based on each user behavior data in a sample set, each type of product information in the behavior data, and feedback corresponding to the user behavior, including: inputting behaviors of the same type in each user behavior data into the same input channel to train the multi-channel neural network model.

5. The interactive system for an e-commerce platform according to claim 1, characterized in that: The loss of the multi-channel neural network model is calculated using the following formula: in, The model predicts the The product recommendation matrix corresponding to each behavior type, Indicates The gold standard product recommendation matrix corresponding to each behavior type, represents the number of behavior categories, and represents the weight parameter, represents the regularization parameter, Represents the total sample loss.

6. The interactive system for an e-commerce platform according to claim 1, characterized in that: Generating a training sample set based on the user behavior data and the corresponding product information includes: resampling the user behavior data and the corresponding product information to standard voxels using a linear interpolation method; Normalize the voxel values ​​of user behavior data; The normalized user behavior data is segmented to obtain a user behavior mask, the minimum circumscribed cuboid of the user behavior mask is calculated, the user behavior data and the corresponding product information in the minimum circumscribed cuboid are extracted, and a training sample set is generated.

7. The interactive system for an e-commerce platform according to claim 1, characterized in that: The multi-channel neural network model includes a multi-channel convolution layer, a downsampling convolution module and an upsampling convolution module connected in sequence; The multi-channel convolution layer is used to extract a multi-channel feature image from multiple input channels by using convolution; the downsampling convolution module is used to extract features of different levels from the multi-channel feature image, and the upsampling convolution module is used to upsample the extracted features and output the recommendation results; The downsampling convolution module includes a plurality of downsampling convolution units, each of which includes a three-dimensional convolution layer, a LeakyReLU layer, a batch normalization layer, and a maximum pooling layer connected in sequence; The upsampling convolution module includes multiple upsampling convolution units, each of which includes a three-dimensional convolution layer, a LeakyReLU layer, a batch normalization layer and a transposed convolution layer connected in sequence.

8. An interactive device for an e-commerce platform, applied to an interactive system for an e-commerce platform as claimed in claims 1-7, characterized in that: Includes the following modules: A sample set generation module, used to obtain user behavior data and corresponding commodity information, and generate a training sample set based on the user behavior data and the corresponding commodity information; A behavior classification module is used to classify the user behavior data in the training sample set based on user preferences to obtain the type of each user behavior data; A classification model training module is used to train a classification neural network model based on each type of commodity information in each user behavior data in the sample set to obtain a user behavior classification model; A recommendation model training module is used to train a multi-channel neural network model based on each user behavior data in the sample set, each type of product information in the behavior data, and feedback corresponding to the user behavior to obtain a product recommendation model; An automatic interaction module, used for inputting the behavior data of the user to be interacted into the user behavior classification model to obtain the type of the user behavior to be interacted; According to the type of the user behavior to be interacted, the user behavior data to be interacted is input into the product recommendation model to obtain product recommendations for the user behavior.

9. The interactive device for an e-commerce platform according to claim 8, characterized in that: The behavior classification module classifies the user behavior data in the training sample set by using the following steps: for each user behavior data, the first behavior with corresponding product feedback is taken as the first category; each behavior is traversed in turn, and the preference change rate between the current behavior and the previous behavior is calculated based on the product information corresponding to the user behavior; If the preference change rate is greater than the threshold, the current behavior and the previous behavior are of different types; otherwise, the current behavior and the previous behavior are of the same type.

10. The interactive device for an e-commerce platform according to claim 9, characterized in that: The preference change rate includes the price change rate and category change rate ; Calculating the preference change rate between the current behavior and the previous behavior based on the product information corresponding to the user behavior, including: calculating the statistical characteristics of the preference corresponding to each product information of the user behavior; The rate of change in preference between the current action and the previous action is calculated using the following formula: Price Change Rate Category change rate in, and Respectively represent the commodity prices corresponding to the current behavior and the previous behavior, and Respectively represent the product categories corresponding to the current behavior and the previous behavior.

Images