Data classification method and device for e-commerce platform, equipment and storage medium

Abstract

The invention relates to the technical field of data processing, and discloses a data classification method and device for an e-commerce platform, equipment and a storage medium. The method comprises the following steps: performing graph neural network processing on a user behavior feature matrix and commodity information data to obtain a commodity knowledge graph; performing interaction mode analysis processing on the user behavior feature matrix and the commodity knowledge graph through a time sequence attention mechanism to obtain a time sequence interaction feature vector; performing feature fusion processing through a multi-modal feature fusion network to obtain a fusion feature vector; and carrying out classification processing on the fusion feature vector through a hierarchical attention decision network to obtain an initial classification result, and carrying out data classification optimization processing to obtain target classification data. According to the method, the time sequence attention mechanism and the multi-modal feature fusion network are introduced, and the hierarchical attention decision network and the interpretability rule processor are combined, so that deep fusion of user behaviors and commodity features is realized, and the classification accuracy and interpretability are improved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a data classification method, device, equipment and storage medium for an e-commerce platform. Background Art

[0002] At present, the data classification methods of e-commerce platforms mainly rely on manual labeling and rule matching. Traditional classification methods usually use keyword-based text matching, statistical-based collaborative filtering, and simple machine learning-based classification algorithms. These methods mainly focus on the static attribute characteristics of products, such as product titles, descriptions, categories, and other information for classification. At the same time, some improved methods have begun to introduce user behavior data to assist in product classification by analyzing user clicks, collections, purchases, and other behaviors, and have begun to try to use deep learning models to improve classification results.

[0003] However, the existing e-commerce platform data classification methods have the following shortcomings: first, the use of user behavior data remains at a simple statistical level, failing to fully explore the temporal evolution characteristics contained in the user behavior sequence; second, existing methods often separate product features and user behavior features for processing, lacking an effective feature fusion mechanism; third, the classification results lack interpretability, making it difficult to understand the basis for classification decisions; finally, existing methods fail to handle the hierarchical characteristics of the product category system well, resulting in low classification accuracy. Summary of the invention

[0004] The present application provides a data classification method, apparatus, device and storage medium for an e-commerce platform, which is used to achieve deep fusion of user behavior and product features by introducing a temporal attention mechanism and a multimodal feature fusion network, combined with a hierarchical attention decision network and an interpretable rule processor, thereby improving the accuracy and interpretability of classification.

[0005] In the first aspect, the present application provides a data classification method for an e-commerce platform, and the data classification method for an e-commerce platform includes: performing multi-dimensional feature extraction processing on user interaction behavior data to obtain a user behavior feature matrix; performing graph neural network processing on the user behavior feature matrix and product information data to obtain a product knowledge graph; performing interaction pattern analysis processing on the user behavior feature matrix and the product knowledge graph through a temporal attention mechanism to obtain a temporal interaction feature vector; performing feature fusion processing on the user behavior feature matrix, the product knowledge graph and the temporal interaction feature vector through a multimodal feature fusion network to obtain a fused feature vector; performing classification processing on the fused feature vector through a hierarchical attention decision network to obtain an initial classification result; performing data classification optimization processing on the initial classification result through an interpretable rule processor to obtain target classification data.

[0006] In a second aspect, the present application provides a data classification device for an e-commerce platform, the data classification device for an e-commerce platform comprising:

[0007] The acquisition module is used to perform multi-dimensional feature extraction processing on the user interaction behavior data to obtain a user behavior feature matrix;

[0008] A processing module, used to perform graph neural network processing on the user behavior feature matrix and product information data to obtain a product knowledge graph;

[0009] An analysis module, used to perform interaction pattern analysis on the user behavior feature matrix and the product knowledge graph through a temporal attention mechanism to obtain a temporal interaction feature vector;

[0010] A fusion module, used to perform feature fusion processing on the user behavior feature matrix, the product knowledge graph and the time series interaction feature vector through a multimodal feature fusion network to obtain a fused feature vector;

[0011] A classification module, used for classifying the fused feature vector through a hierarchical attention decision network to obtain an initial classification result;

[0012] The optimization module is used to perform data classification optimization processing on the initial classification results through an interpretable rule processor to obtain target classification data.

[0013] The third aspect of the present application provides a computer device, wherein the memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor and the memory communicate via a bus, and when the machine-readable instructions are executed by the processor, the steps of the above-mentioned data classification method for an e-commerce platform are performed.

[0014] A fourth aspect of the present application provides a computer-readable storage medium, which stores instructions. When the computer-readable storage medium is run on a computer, the computer executes the above-mentioned data classification method for an e-commerce platform.

[0015] In the technical solution provided by this application, by performing multi-dimensional feature extraction processing on user interaction behavior data, the user's search, click and purchase behavior characteristics are fully captured, and the accuracy of user behavior portraits is effectively improved. By processing the user behavior feature matrix and product information data with a graph neural network, a product knowledge graph containing rich semantic information and structural relationships is constructed, providing comprehensive knowledge support for product classification. The interactive mode analysis of the user behavior feature matrix and the product knowledge graph is performed using a temporal attention mechanism, which accurately captures the dynamic change characteristics of user interests and improves the classification system's understanding of user preferences. The multimodal feature fusion network is introduced to fuse the user behavior feature matrix, the product knowledge graph and the temporal interaction feature vector, realizing the effective integration of features from different sources and enhancing the integrity and complementarity of feature representation. Classification processing is performed through a hierarchical attention decision network, and a flexible multi-level classification system is established, which improves the accuracy and hierarchy of the classification results. Finally, an interpretable rule processor is used to optimize data classification, which not only ensures the accuracy of the classification results, but also provides a clear description of the classification basis, enhancing the interpretability and credibility of the classification results. The overall solution significantly improves the accuracy and efficiency of product classification on e-commerce platforms through the organic combination of multiple innovative technical means. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0017] Figure 1 This is a schematic diagram of an embodiment of a data classification method for an e-commerce platform in an embodiment of the present application;

[0018] Figure 2 This is a schematic diagram of an embodiment of a data classification device for an e-commerce platform in an embodiment of the present application;

[0019] Figure 3 It is a schematic diagram of the structure of a computer device in an embodiment of the present application. DETAILED DESCRIPTION

[0020] Embodiments of the present application provide a data classification method, device, equipment and storage medium for an e-commerce platform. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or precedence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than the content illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0021] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the data classification method for an e-commerce platform in the embodiment of the present application includes:

[0022] Step S101, performing multi-dimensional feature extraction processing on user interaction behavior data to obtain a user behavior feature matrix;

[0023] Step S102: Perform graph neural network processing on the user behavior feature matrix and product information data to obtain a product knowledge graph;

[0024] Step S103: Perform interaction pattern analysis on the user behavior feature matrix and the product knowledge graph through a temporal attention mechanism to obtain a temporal interaction feature vector;

[0025] Step S104: performing feature fusion processing on the user behavior feature matrix, the product knowledge graph and the time series interaction feature vector through a multimodal feature fusion network to obtain a fused feature vector;

[0026] Step S105, classifying the fused feature vector through a hierarchical attention decision network to obtain an initial classification result;

[0027] Step S106: Perform data classification optimization processing on the initial classification results through an interpretable rule processor to obtain target classification data.

[0028] It is understandable that the execution subject of the present application may be a data classification device for an e-commerce platform, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0029] Specifically, the user interaction behavior data is processed, and the basic behavior sequence data is formed by reading the user's search word sequence, click sequence and purchase sequence on the e-commerce platform. For each user's behavior sequence, the time dimension characteristics such as the distribution of the visit period, shopping cycle, and length of stay are recorded to construct a time feature vector. The behavior association graph is generated by analyzing the relationship between the commodity categories visited by the user before and after, the correlation between the search word and the purchased commodity, the degree of response to promotional activities and other scene-related behaviors. The node importance of the behavior association graph is calculated by the PageRank algorithm to obtain the user interest weight vector. The behavior sequence data is sliced ​​in time series, and the dynamic interest features are extracted through the long short-term memory network to finally obtain the user behavior feature matrix. Based on the obtained user behavior feature matrix, the product information is further processed. First, the product title, description and specification parameters are segmented, and the keyword sequence is extracted and encoded into a word vector. The co-occurrence relationship and conversion relationship of the commodities in the user behavior feature matrix are analyzed to obtain the commodity association data. The commodity text features and the association data are integrated to construct the commodity node feature vector. The hierarchical relationship is established through the commodity category system, and the semantic relationship between commodities is derived using the graph reasoning method. Use graph neural networks to embed nodes and eventually form a knowledge graph containing complete product information.

[0030] After obtaining the user behavior feature matrix and the product knowledge graph, the interaction pattern between the two is analyzed through the temporal attention mechanism. First, the user behavior is divided into time windows, and the behavior data in different time periods are given attenuation weights. The user behavior is matched with the product knowledge graph to extract the user-product interaction sequence. The attention weights at different time points are calculated through the temporal attention network, and the cross features are extracted and sequence encoded. The encoded sequence is subjected to temporal pattern recognition to obtain the interaction pattern data, which is finally converted into a temporal interaction feature vector. The above three features are processed through a multimodal feature fusion network. First, the user behavior features and the product graph are mapped into feature space, and the mutual attention features between them are calculated. The mutual attention features are aligned with the temporal interaction features, and the feature information is supplemented through the residual connection network. The aligned features are cross-modally interacted and reconstructed, and the autoencoder is used for compression mapping. The relationship between different features is captured through the multi-head attention mechanism, and finally the features are normalized and nonlinearly transformed to obtain the fused feature vector.

[0031] For the fused feature vector, a hierarchical attention decision network is used for classification processing. First, multiple levels are divided according to the product category system, and weights are assigned to nodes at different levels. The feature attention weight of each level is calculated, and an adaptive discrimination threshold is set. The features are extracted hierarchically through a hierarchical feature extraction network, and the correlation between nodes is calculated. The classification path probability graph is constructed and pruned to obtain a set of candidate paths. The candidate paths are scored and multi-path integrated, and finally the initial classification results are obtained through confidence evaluation and soft voting mechanism. Finally, the initial classification results are optimized using an interpretable rule processor. Rules are extracted from the classification results and ranked in importance to generate rule weights. The rules are checked for consistency, and conflicting rules are identified and reconciled. The classification decision path is tracked and the contribution of features to the classification results is analyzed. The classification basis is extracted based on the analysis results, and the classification labels are corrected. Finally, target classification data containing category labels, classification basis and confidence scores are generated.

[0032] For example, an e-commerce platform classifies mobile phone products. By analyzing the behavioral data of user A within one month, it is found that the user has a lot of mobile phone browsing records on weekday evenings (20:00-22:00), with an average single stay time of 3 minutes, and mainly focuses on search terms related to camera functions and battery life performance. The PageRank algorithm calculates that the user's interest weight for camera functions is 0.8, and the interest weight for battery life is 0.7. After processing the product information, it is found that the product description of a certain mobile phone contains keywords such as "100 million pixels" and "5000mAh large battery", and it is positioned as a high-end camera phone category through the graph neural network. Analyzing the interaction sequence of user behavior and product features, it is found that the user's click rate for this type of mobile phone increases over time, from 5% in the early stage to 15% in the later stage. After multimodal feature fusion and hierarchical classification, the mobile phone is classified into the "mobile phone / high-end mobile phone / photography mobile phone" category with a classification confidence of 0.85. The main classification basis is the product's camera configuration features and the user's interest matching degree.

[0033] In the embodiment of the present application, by performing multi-dimensional feature extraction processing on the user interaction behavior data, the user's search, click and purchase behavior characteristics are fully captured, and the accuracy of the user behavior portrait is effectively improved. By processing the user behavior feature matrix and the commodity information data with a graph neural network, a commodity knowledge graph containing rich semantic information and structural relationships is constructed, providing comprehensive knowledge support for commodity classification. The interactive mode analysis of the user behavior feature matrix and the commodity knowledge graph is performed using a temporal attention mechanism, which accurately captures the dynamic change characteristics of user interests and improves the depth of understanding of user preferences by the classification system. The multimodal feature fusion network is introduced to fuse the user behavior feature matrix, the commodity knowledge graph and the temporal interaction feature vector, realizing the effective integration of features from different sources and enhancing the integrity and complementarity of feature representation. Classification processing is performed through a hierarchical attention decision network, and a flexible multi-level classification system is established, which improves the accuracy and hierarchy of the classification results. Finally, an interpretable rule processor is used to optimize data classification, which not only ensures the accuracy of the classification results, but also provides a clear description of the classification basis, enhancing the interpretability and credibility of the classification results. The overall solution significantly improves the accuracy and efficiency of product classification on e-commerce platforms through the organic combination of multiple innovative technical means.

[0034] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0035] (1) Perform query record processing on user interaction behavior data, extract the user's search word sequence, click sequence, and purchase sequence to obtain basic behavior sequence data, and perform time feature analysis on the basic behavior sequence data to obtain the user's time feature vector;

[0036] (2) Performing sequence association analysis on the basic behavior sequence data and the user time feature vector to obtain a user behavior association graph, and performing node importance calculation on the user behavior association graph using the PageRank algorithm to obtain a user interest weight vector;

[0037] (3) Performing time-series slicing processing on the user's temporal feature vector and the user's interest weight vector to obtain time-series interest segments, and performing sequence encoding processing on the time-series interest segments to obtain a user interest evolution sequence;

[0038] (4) extracting sequence features of the user interest evolution sequence through a long short-term memory network to obtain a dynamic interest feature vector, and concatenating the dynamic interest feature vector and the user interest weight vector to obtain a combined feature vector;

[0039] (5) Perform dimension normalization on the combined feature vector to obtain a standardized feature vector, and perform matrix transformation on the standardized feature vector to obtain a user behavior feature matrix.

[0040] Specifically, the interactive behavior data generated by users on the e-commerce platform is queried and recorded, and each user's historical search records, click records, and purchase records are read from the database. The search term sequence records the search keywords entered by the user and their timestamps, the click sequence records the ID and duration of the product browsed by the user, and the purchase sequence records the order information of the user's final purchase of the product, forming the basic behavior sequence data. The basic behavior sequence data is analyzed for time characteristics, and features such as the user's active time distribution, shopping cycle, and time intervals of various behaviors are extracted to generate a user time feature vector to characterize the user's time dimension behavior pattern. Taking a certain user as an example, his daily active peak time is 20:00-22:00, and he has an average of one search behavior every 3 days. The average duration of each browsing of the product is 2 minutes. These time characteristics are encoded into the feature vector.

[0041] Further sequence association analysis is performed on the basic behavior sequence data and user time feature vectors to explore the association relationship between user behaviors. The semantic relevance between adjacent search terms, the conversion relationship from search to click, the conversion relationship from click to purchase, etc. are analyzed to construct a user behavior association graph. The importance score of each behavior node is calculated for the constructed association graph through the PageRank algorithm to generate a user interest weight vector. During the algorithm process, nodes directly related to purchase behavior are given a higher initial weight, and the final importance score of each node is calculated through iterative propagation. For example, for the mobile phone category, the user's "camera phone" search term node score is 0.8, indicating that the user has a strong interest in the camera function of the mobile phone.

[0042] Then, the user time feature vector and user interest weight vector are processed by time series slicing, and the complete behavior sequence is segmented according to time windows (such as days, weeks, and months) to obtain multiple time series interest segments. For each time series interest segment, the user's behavior characteristics and interest distribution in the time period are extracted, and encoded into a vector representation of fixed dimensions through a sequence encoder to form a user interest evolution sequence that reflects the trend of user interest changes. For example, a user's interest gradually shifted from initially focusing on the basic functions of the mobile phone to focusing on camera performance, and this evolution trend was encoded in the sequence.

[0043] Then, the user interest evolution sequence is processed using a long short-term memory network, which can capture long-term dependencies in sequence data. Through the network's gating mechanism, the interest features of different periods are selectively retained and forgotten, generating a feature vector that can express the user's dynamic interest changes. The dynamic interest feature vector is spliced ​​with the user interest weight vector obtained previously to obtain a combined feature vector that contains both the current interest distribution and interest evolution information. For example, the user's current interest weight for the photo function is 0.8, and from the historical evolution trend, this interest is steadily increasing. This information is integrated into the combined feature vector. Finally, the combined feature vector is dimensionally normalized, and the eigenvalues ​​of different dimensions are uniformly mapped to the [0,1] interval to eliminate the dimension effect. The normalized feature vector is processed by linear or nonlinear transformation to adjust the feature distribution and generate the final user behavior feature matrix. Each row of the matrix corresponds to a user, and each column corresponds to a feature dimension, which fully describes the user's behavior characteristics and interest preferences. Based on this feature matrix, more accurate product recommendations and personalized services can be performed in the future. For example, for the aforementioned user, according to his feature matrix showing that he pays great attention to camera performance, the system will give priority to recommending mobile phone products with outstanding camera capabilities.

[0044] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0045] (1) Performing text segmentation processing on the product information data, extracting keyword sequences from the product title, description, and specification parameters to obtain a product feature word sequence, and performing word vector encoding processing on the product feature word sequence to obtain a product text feature vector;

[0046] (2) Perform product association analysis on the user behavior feature matrix, extract the co-occurrence relationship and conversion relationship between products, obtain product association data, and perform feature fusion processing on the product association data and product text feature vector to obtain the product node feature vector;

[0047] (3) Perform category-level matching on the feature vectors of the commodity nodes, construct the category relationship of the commodities, obtain the commodity category structure data, and deduce the commodity relationship on the commodity category structure data through the graph reasoning method to obtain the commodity semantic relationship data;

[0048] (4) Graph structure construction is performed on the product semantic relationship data to establish the connection relationship between product nodes to obtain the initial knowledge graph, and node embedding learning is performed on the initial knowledge graph through a graph neural network to obtain the product node embedding vector;

[0049] (5) Perform knowledge fusion processing on the product node embedding vector and product semantic relationship data to obtain the product knowledge graph.

[0050] Specifically, the product information data of the e-commerce platform is processed, and the text content such as product title, product detailed description and specification parameters are divided into word units through word segmentation technology. When extracting the keyword sequence, the important information such as product model, brand, core parameters, etc. is retained to form a product feature word sequence. For each feature word, the word vector encoding model is used to calculate its vector representation in the semantic space, and the product text feature vector is summarized. For example, the description text of a certain mobile phone "6.7-inch OLED curved screen, 100 million pixel main camera, 5000mAh large battery" is obtained after word segmentation. The feature word sequence, each word is mapped to a vector of fixed dimension. Based on the user behavior feature matrix, the association relationship between products is analyzed, the frequency of different products being browsed or purchased by the same user is counted, and the co-occurrence probability between product pairs is calculated. At the same time, the conversion path of users from browsing one product to finally purchasing another product is analyzed to obtain the conversion relationship between products. The statistical product association data is fused with the product text feature vector to generate a product node feature vector containing multi-dimensional information. For example, if it is found that after users browse high-end camera phones, there is an 80% probability that they will pay attention to other camera phones in the same price range. This correlation is integrated into the node features.

[0051] Match the product node feature vectors to the category system, map the products to the category hierarchy structure preset by the e-commerce platform, and determine the category path to which each product belongs. Construct formalized product category structure data based on the category affiliation of the products. Use graph reasoning methods to expand the category structure, deduce the implicit semantic relationship between products through transitive rules and similarity rules, and generate product semantic relationship data. For example, classify "mobile phone" into the category path of "mobile phone / smartphone / camera phone", and deduce the similarity relationship with other mobile phones with outstanding camera functions.

[0052] Based on the semantic relationship data of commodities, a graph structure is constructed, with commodities as nodes in the graph and various relationships between commodities as edges to form an initial knowledge graph. The nodes in the graph are represented and learned by graph neural network technology. The network aggregates neighbor node information during message transmission and learns to obtain commodity node embedding vectors containing structural information. For example, for camera phone nodes, their embedding vectors will integrate the feature information of neighbor nodes such as related accessories and similar products. Finally, the commodity node embedding vectors are integrated with the commodity semantic relationship data to construct a complete commodity knowledge graph. This graph contains not only the basic attribute information of commodities, but also rich knowledge such as hierarchical relationships, similar relationships, and association relationships between commodities. For example, in the final knowledge graph, each mobile phone product is connected to its accessories, competitors, applicable scenarios, and other entities to form a complete knowledge network. Based on this knowledge graph, the system can better understand the relationship between commodities and provide knowledge support for subsequent classification tasks. For example, when classifying newly listed mobile phones, the category information and attribute characteristics of similar commodities in the knowledge graph can be referred to.

[0053] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0054] (1) Performing time window division processing on the user behavior feature matrix to obtain user behavior time series segments, and performing time attenuation weighting processing on the user behavior time series segments to obtain time series weight data;

[0055] (2) Perform interactive matching processing on the temporal weight data and the product knowledge graph, extract the user-product interaction sequence, obtain the interaction sequence data, and calculate the temporal features of the interaction sequence data through the temporal attention network to obtain the temporal attention weight;

[0056] (3) Perform cross-feature extraction on the interaction sequence data and temporal attention weights to obtain user-product interaction features, and perform sequence encoding on the user-product interaction features to obtain an interaction encoding vector;

[0057] (4) performing temporal pattern recognition processing on the interaction coding vector to obtain interaction pattern data, and performing feature combination processing on the interaction pattern data and the temporal weight data to obtain a combined interaction feature;

[0058] (5) Perform dimension conversion on the combined interaction features to obtain a time series interaction feature vector.

[0059] Specifically, the user behavior feature matrix is ​​divided into time windows, and the continuous time series is divided into multiple time series segments according to a fixed length (such as hours, days, weeks). A time decay factor is introduced into the user behavior data in each time series segment, so that the behavior data farther away from the current time has a lower weight, and the time series weight data is generated. For example, the user's browsing history for the past 30 days is divided by day, and the behavior weight of the i-th day is set to 0.95. i , making recent behavior receive higher attention.

[0060] The user behavior data with temporal weights are interactively matched with the product knowledge graph, and the user's interaction with different products at different time points is recorded to extract the user-product interaction sequence. These interaction sequence data are processed through the temporal attention network to calculate the importance weights of the interaction behaviors at different time points. For example, when a user browses a mobile phone, a quick browse of 0-2 minutes and a detailed view of 2-5 minutes are compared. The latter gets a higher attention weight, indicating that the user is more interested in the product.

[0061] Based on the interaction sequence data and the calculated temporal attention weights, cross-feature extraction is performed to analyze the association pattern between user behavior and product attributes. The user's operation sequence (such as click, favorite, add to cart, purchase) is cross-combined with the product's attribute characteristics (such as price, category, performance parameters) to generate user-product interaction features. These interaction features are encoded through a sequence encoder to convert the variable-length interaction sequence into a fixed-dimensional interaction encoding vector. For example, it is recorded that the user browsed a number of mobile phones of different prices, and then focused on mid-to-high-end camera phones. This behavior pattern is encoded into the vector. Temporal pattern recognition is performed on the interaction encoding vector to analyze the law of user behavior changing over time. Identify the user's interest transfer trajectory, shopping decision path and other temporal patterns to generate interaction pattern data. The identified interaction pattern data is combined with the previous temporal weight data to obtain a combined interaction feature containing temporal information. For example, if the system finds that the user initially pays attention to the basic functions of the mobile phone, then gradually turns to the camera performance, and recently shows a strong purchase intention, this evolution pattern is retained in the combined feature.

[0062] Finally, the combined interaction features are transformed into a dimension, and the features are mapped to a suitable dimensional space through operations such as dimensionality reduction or feature transformation to generate the final time-series interaction feature vector. This feature vector not only reflects the user's historical behavior trajectory, but also contains the current interest tendency, providing an important basis for subsequent classification tasks. For example, based on the time-series interaction feature vector of a certain user, the user's attention to the camera phone category continues to increase, and the recent browsing time and interaction frequency are both high. This information can be used to predict the user's purchase intention and product category preference.

[0063] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0064] (1) Perform feature space mapping on the user behavior feature matrix to obtain the user behavior projection vector, and perform graph structure embedding on the product knowledge graph to obtain the product graph vector;

[0065] (2) Perform mutual attention calculation on the user behavior projection vector and the product graph vector to obtain the interactive attention feature, and perform feature alignment on the interactive attention feature and the temporal interactive feature vector to obtain the aligned feature matrix;

[0066] (3) Performing feature supplementation processing on the aligned feature matrix through a residual connection network to obtain supplementary feature data, and performing cross-modal feature extraction on the supplementary feature data to obtain modal interaction features;

[0067] (4) Performing feature reconstruction processing on the modal interaction features to obtain a reconstructed feature vector, and compressing and mapping the reconstructed feature vector through an autoencoder to obtain compressed feature data;

[0068] (5) performing multi-head attention calculation processing on the compressed feature data to obtain a multi-head feature vector, and performing feature normalization processing on the multi-head feature vector to obtain normalized feature data;

[0069] (6) performing feature concatenation processing on the normalized feature data and the modal interaction features to obtain a concatenated feature vector, and performing nonlinear transformation on the concatenated feature vector through a multi-layer perceptron to obtain transformed feature data;

[0070] (7) Perform feature selection processing on the transformed feature data to obtain a fused feature vector.

[0071] Specifically, the user behavior feature matrix is ​​spatially mapped, and the user behavior data is projected into a new feature space through linear or nonlinear transformation to generate a user behavior projection vector. At the same time, the product knowledge graph is structured and embedded, and the node and edge relationships in the graph are encoded as vector representations to obtain a product graph vector. For example, the user's browsing, collection, purchase and other behavioral features are mapped to a 128-dimensional vector space, and the attributes and relationship information of the product are also mapped to the vector space of the same dimension. The mutual attention score is calculated for the user behavior projection vector and the product graph vector, and the correlation strength between the two types of features is analyzed to generate the interactive attention feature. The interactive attention feature is aligned with the temporal interactive feature vector in the feature dimension to ensure that features from different sources can be compared and fused in the same feature space to form an aligned feature matrix. For example, a user's behavior vector for the camera phone category and the graph vector of the product in this category calculate a high attention score, indicating that the user behavior is highly correlated with the product features.

[0072] The aligned feature matrix is ​​supplemented by the residual connection network, retaining the original feature information while adding new feature representation to obtain supplementary feature data. Cross-modal feature extraction is performed on the supplementary feature data to analyze the complementary relationship between features of different modalities (user behavior, product attributes, and time series relationships) and generate modal interaction features. For example, the user's purchase behavior characteristics are crossed with the product's price range characteristics to discover the user's preference pattern for mid- to high-end priced products.

[0073] The modal interaction features are reconstructed, and reconstructed feature vectors are generated through feature transformation and combination. The reconstructed feature vectors are compressed using an autoencoder to reduce the feature dimension while retaining key information to obtain compressed feature data. For example, the 512-dimensional interaction features are compressed to 256 dimensions through an autoencoder to retain the most representative feature information. The multi-head attention mechanism is used to process the compressed feature data, calculate the attention weights from multiple feature subspaces, and generate a multi-head feature vector. The multi-head feature vectors are normalized, and the feature values ​​are scaled to a uniform numerical range to obtain normalized feature data. For example, 8 attention heads are used to focus on different feature dimensions at the same time, and each head generates a 64-dimensional feature representation.

[0074] The normalized feature data is concatenated with the previous modal interaction features to form a more comprehensive feature representation and obtain a concatenated feature vector. The concatenated feature vector is nonlinearly transformed through a multi-layer perceptron to enhance the expressive power of the features and generate transformed feature data. If a three-layer perceptron is used, the hidden layers are 512, 256, and 128 neurons, respectively, and the activation function is ReLU. Finally, feature selection is performed on the transformed feature data, and the most representative feature dimensions are selected according to the feature importance score to generate the final fused feature vector. This feature vector integrates user behavior, product knowledge, and time series information to provide comprehensive feature support for subsequent classification tasks. For example, the fused feature vector of a user shows that he has a high degree of attention to features such as camera functions, screen quality, and battery life in the mobile phone category. This information can be used to predict specific product categories that users may be interested in.

[0075] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0076] (1) Perform multi-level category division processing on the fused feature vector to obtain category hierarchical structure data, and perform node weight allocation processing on the category hierarchical structure data to obtain hierarchical node weights;

[0077] (2) Performing hierarchical attention calculation on the fused feature vector and hierarchical node weights to obtain hierarchical feature weights, and performing adaptive threshold adjustment on the hierarchical feature weights to obtain hierarchical discrimination thresholds;

[0078] (3) Performing hierarchical feature extraction on the fused feature vector through a hierarchical feature extraction network to obtain a hierarchical feature matrix, and performing node correlation calculation on the hierarchical feature matrix to obtain node correlation data;

[0079] (4) Calculate the path probability of the node association data to obtain a classified path probability graph, and perform path pruning on the classified path probability graph to obtain a set of candidate paths;

[0080] (5) Performing path scoring processing on the candidate path set and the hierarchical discrimination threshold to obtain path score data, and performing multi-path integration processing on the path score data to obtain integrated classification data;

[0081] (6) Perform confidence assessment on the integrated classification data to obtain classification confidence, and assign labels to the classification confidence through a soft voting mechanism to obtain the initial classification result.

[0082] Specifically, the fused feature vector is divided into multiple levels of categories, and a hierarchical classification structure is constructed according to the commodity category system preset by the e-commerce platform. From the top-level categories to the bottom-level detailed categories, a complete category hierarchy data is formed. Each node in the category hierarchy is assigned a weight according to its importance and discrimination, and the hierarchical node weight is generated. For example, the hierarchical structure of the mobile phone category can be divided into hierarchical paths such as "mobile phone / smart phone / camera phone", and each layer of nodes obtains different weight values ​​according to the number of commodities it contains and the feature discrimination. The fused feature vector and the hierarchical node weight are used for hierarchical attention calculation, and the importance of the features at different hierarchical nodes is analyzed to obtain the hierarchical feature weight. An adaptive discrimination threshold is set for the feature weights of different levels, and the accuracy of classification is controlled by dynamically adjusting the threshold value to generate a hierarchical discrimination threshold. For example, in the mobile phone classification, the discrimination threshold of the first-level category "mobile phone" is set to 0.6, the threshold of the second-level category "smart phone" is set to 0.7, and the threshold of the third-level category "camera phone" is set to 0.8, ensuring that the accuracy of classification is improved layer by layer. Use a hierarchical feature extraction network to process the fused feature vector, extract feature representations at different levels, and form a hierarchical feature matrix. Calculate the correlation between nodes at different levels, analyze the similarity and dependency of node features, and generate node association data. For example, analyze the feature correlation between the "camera phone" node and other mobile phone category nodes to find a set of nodes with significant correlation.

[0083] Calculate the path probability of node association data and construct a classification path probability graph from the root node to the leaf node. Remove the classification paths with low probability through path pruning operation, retain the most likely classification path, and form a candidate path set. For example, in the classification process of a certain product, the possible paths include "mobile phone / smart phone / camera phone" and "mobile phone / smart phone / game phone". The paths with higher confidence are retained through probability calculation and pruning. Combine the candidate path set and the hierarchical discrimination threshold to score each classification path and generate path score data. Integrate multiple possible classification paths, and synthesize the classification results of multiple paths through weighted average or voting to obtain integrated classification data. For example, the score of a mobile phone product in the "camera phone" path is 0.85, and the score in the "game phone" path is 0.65. The final classification direction is determined through integration processing. Confidence evaluation is performed on the integrated classification data, and the reliability score of the classification result is calculated to generate classification confidence. A soft voting mechanism is used to assign labels to classification results with higher confidence, and the voting results of multiple classification paths are comprehensively considered to obtain the initial classification results. For example, for a certain mobile phone product, after confidence evaluation and soft voting, it was finally classified into the "camera phone" category with a confidence level of 0.85, which meets the preset discrimination threshold requirements.

[0084] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0085] (1) Performing rule extraction processing on the initial classification results to obtain a classification rule set, and sorting the classification rule set by rule importance to obtain rule weight data;

[0086] (2) Performing rule matching processing on the initial classification results and rule weight data to obtain a rule coverage matrix, and performing consistency check processing on the rule coverage matrix to obtain rule consistency data;

[0087] (3) performing conflict rule identification processing on the rule consistency data to obtain rule conflict data, and performing rule reconciliation processing on the rule conflict data to obtain a reconciliation rule set;

[0088] (4) tracing the decision path of the reconciliation rule set to obtain the classification decision path, and performing feature contribution analysis on the classification decision path to obtain feature explanation data;

[0089] (5) extracting the classification basis of the feature interpretation data to obtain a classification basis vector, and performing label correction processing on the classification basis vector and the initial classification result to obtain a corrected classification label;

[0090] (6) Performing category structure mapping processing on the modified classification labels to obtain target classification data, wherein the target classification data includes the category label of the product, the classification basis description, and the confidence score.

[0091] Specifically, the rules are extracted from the initial classification results, and the key judgment conditions and feature thresholds in the classification decision process are analyzed to form a classification rule set. Each rule in the rule set is evaluated for importance based on its coverage, accuracy, and discrimination, and rule weight data is generated. For example, for mobile phone classification, the rule "more than 50 million pixels and a photo score greater than 85 points are classified as camera phones" is extracted, and weights are assigned according to the discrimination effect of the rule. The initial classification results are matched with the rule weight data, and each classification result is checked to see if it meets the corresponding rule conditions, and a rule coverage matrix is ​​generated. The rule coverage matrix is ​​checked for consistency, and the coordination and complementarity between different rules are analyzed to obtain rule consistency data. If a mobile phone meets multiple classification rules at the same time, it is necessary to evaluate whether there are logical conflicts between these rules.

[0092] Analyze the rule consistency data, identify the rule combinations that may conflict, and generate rule conflict data. Solve the conflicts between rules through rule reconciliation. Possible solutions include rule merging, priority adjustment, or adding supplementary conditions, and finally obtain a coordinated and consistent set of reconciled rules. For example, when a mobile phone meets the classification rules of both camera phones and gaming phones, reconcile the conflicts by analyzing the weights of core parameters to determine the most suitable classification. Track the decision path of the reconciled rule set, restore the key decision points and judgment basis in the classification process, and form a complete classification decision path. Analyze the influence of each feature in the decision path on the final classification result, calculate the feature contribution, and generate feature explanation data. For example, the analysis found that the main basis for a mobile phone to be classified as a camera phone is its camera configuration and image processing capabilities.

[0093] Based on the feature interpretation data, the key classification basis is extracted, and the qualitative and quantitative judgment conditions are integrated into the classification basis vector. The initial classification results are reviewed and revised in combination with the classification basis vector to ensure the accuracy and interpretability of the classification results, and the revised classification labels are obtained. For example, through analysis, it is found that although a certain mobile phone has excellent camera functions, its main marketing focus and user group positioning are more inclined to game performance, so the classification labels are adjusted accordingly. Finally, the revised classification labels are mapped to the category structure system of the e-commerce platform to generate target classification data containing multi-dimensional information. The classification data not only contains the specific category labels of the goods, but also comes with detailed classification basis descriptions and confidence scores of the classification results, providing a basis for subsequent product management and user services. For example, the final classification data of a certain mobile phone shows that the category label is "mobile phone / smartphone / gaming phone", and the classification basis includes core features such as "Snapdragon 8 series processor, 16GB large memory, 144Hz high refresh rate screen", and the confidence score is 0.92, indicating that the classification results have high reliability.

[0094] The above describes the data classification method used in the e-commerce platform in the embodiment of the present application. The following describes the data classification device used in the e-commerce platform in the embodiment of the present application. Figure 2 In one embodiment of the present application, a data classification device for an e-commerce platform includes:

[0095] The acquisition module 201 is used to perform multi-dimensional feature extraction processing on the user interaction behavior data to obtain a user behavior feature matrix;

[0096] Processing module 202, used to perform graph neural network processing on the user behavior feature matrix and product information data to obtain a product knowledge graph;

[0097] An analysis module 203 is used to perform interaction pattern analysis on the user behavior feature matrix and the product knowledge graph through a temporal attention mechanism to obtain a temporal interaction feature vector;

[0098] A fusion module 204 is used to perform feature fusion processing on the user behavior feature matrix, the product knowledge graph and the time series interaction feature vector through a multimodal feature fusion network to obtain a fused feature vector;

[0099] A classification module 205 is used to classify the fused feature vector through a hierarchical attention decision network to obtain an initial classification result;

[0100] The optimization module 206 is used to perform data classification optimization processing on the initial classification results through an interpretable rule processor to obtain target classification data.

[0101] Through the collaboration of the above components, the user interaction behavior data is processed by multi-dimensional feature extraction, and the user's search, click and purchase behavior characteristics are fully captured, which effectively improves the accuracy of user behavior portraits. By processing the user behavior feature matrix and product information data with graph neural network, a product knowledge graph containing rich semantic information and structural relationships is constructed, providing comprehensive knowledge support for product classification. The temporal attention mechanism is used to analyze the interaction mode of the user behavior feature matrix and the product knowledge graph, accurately capturing the dynamic change characteristics of user interests and improving the classification system's understanding of user preferences. The multimodal feature fusion network is introduced to fuse the user behavior feature matrix, product knowledge graph and temporal interaction feature vector, realizing the effective integration of features from different sources and enhancing the integrity and complementarity of feature representation. Classification is performed through the hierarchical attention decision network, and a flexible multi-level classification system is established, which improves the accuracy and hierarchy of the classification results. Finally, the interpretable rule processor is used to optimize data classification, which not only ensures the accuracy of the classification results, but also provides a clear description of the classification basis, enhancing the interpretability and credibility of the classification results. The overall solution significantly improves the accuracy and efficiency of product classification on e-commerce platforms through the organic combination of multiple innovative technical means.

[0102] Based on the same technical concept, the embodiment of the present application also provides a computer device. Figure 3 As shown, it is a schematic diagram of the structure of a computer device 300 provided in an embodiment of the present application, including a processor 301, a memory 302, and a bus 303. Among them, the memory 302 is used to store execution instructions, including a memory 3021 and an external memory 3022; the memory 3021 here is also called an internal memory, which is used to temporarily store the operation data in the processor 301, and the data exchanged with the external memory 3022 such as a hard disk. The processor 301 exchanges data with the external memory 3022 through the memory 3021. When the computer device 300 is running, the processor 301 communicates with the memory 302 through the bus 303.

[0103] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the data classification method for an e-commerce platform.

[0104] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0105] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0106] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A data classification method for an e-commerce platform, characterized in that: The data classification method for the e-commerce platform includes: Perform multi-dimensional feature extraction on user interaction behavior data to obtain a user behavior feature matrix; Performing graph neural network processing on the user behavior feature matrix and product information data to obtain a product knowledge graph; Performing interaction pattern analysis on the user behavior feature matrix and the product knowledge graph through a temporal attention mechanism to obtain a temporal interaction feature vector; Performing feature fusion processing on the user behavior feature matrix, the product knowledge graph and the time series interaction feature vector through a multimodal feature fusion network to obtain a fused feature vector; Classifying the fused feature vector through a hierarchical attention decision network to obtain an initial classification result; The initial classification results are subjected to data classification optimization processing through an interpretable rule processor to obtain target classification data.

2. The data classification method for an e-commerce platform according to claim 1, characterized in that: The multi-dimensional feature extraction process is performed on the user interaction behavior data to obtain a user behavior feature matrix, including: Perform query record processing on the user interaction behavior data, extract the user's search word sequence, click sequence and purchase sequence to obtain basic behavior sequence data, and perform time feature analysis processing on the basic behavior sequence data to obtain a user time feature vector; Performing sequence association analysis on the basic behavior sequence data and the user time feature vector to obtain a user behavior association graph, and performing node importance calculation on the user behavior association graph using a PageRank algorithm to obtain a user interest weight vector; Performing time-series slicing processing on the user time feature vector and the user interest weight vector to obtain time-series interest segments, and performing sequence encoding processing on the time-series interest segments to obtain a user interest evolution sequence; Performing sequence feature extraction processing on the user interest evolution sequence through a long short-term memory network to obtain a dynamic interest feature vector, and performing feature splicing processing on the dynamic interest feature vector and the user interest weight vector to obtain a combined feature vector; The combined feature vector is dimensionally normalized to obtain a standardized feature vector, and the standardized feature vector is matrix transformed to obtain the user behavior feature matrix.

3. The data classification method for an e-commerce platform according to claim 1, characterized in that: The performing graph neural network processing on the user behavior feature matrix and the product information data to obtain a product knowledge graph includes: Performing text segmentation processing on the product information data, extracting keyword sequences from the product title, description and specification parameters to obtain a product feature word sequence, and performing word vector encoding processing on the product feature word sequence to obtain a product text feature vector; Performing product association analysis on the user behavior feature matrix, extracting co-occurrence relationships and conversion relationships between products, obtaining product association data, and performing feature fusion processing on the product association data and the product text feature vector to obtain a product node feature vector; Performing category-level matching on the commodity node feature vectors, constructing commodity category relationships, obtaining commodity category structure data, and performing commodity relationship deduction on the commodity category structure data through a graph reasoning method, obtaining commodity semantic relationship data; Performing graph structure construction processing on the commodity semantic relationship data, establishing connection relationships between commodity nodes, obtaining an initial knowledge graph, and performing node embedding learning on the initial knowledge graph through a graph neural network to obtain commodity node embedding vectors; The product node embedding vector and the product semantic relationship data are subjected to knowledge fusion processing to obtain the product knowledge graph.

4. The data classification method for an e-commerce platform according to claim 1, characterized in that: The interactive mode analysis and processing of the user behavior feature matrix and the commodity knowledge graph by using a temporal attention mechanism to obtain a temporal interaction feature vector includes: Performing time window division processing on the user behavior feature matrix to obtain user behavior time series segments, and performing time attenuation weighting processing on the user behavior time series segments to obtain time series weight data; Performing interactive matching processing on the temporal weight data and the commodity knowledge graph, extracting the user-commodity interaction sequence to obtain the interaction sequence data, and performing temporal feature calculation on the interaction sequence data through a temporal attention network to obtain the temporal attention weight; Performing cross-feature extraction on the interaction sequence data and the temporal attention weight to obtain user-product interaction features, and performing sequence encoding processing on the user-product interaction features to obtain an interaction encoding vector; Performing time series pattern recognition processing on the interaction coding vector to obtain interaction pattern data, and performing feature combination processing on the interaction pattern data and the time series weight data to obtain combined interaction features; The combined interaction feature is dimensionally transformed to obtain the time series interaction feature vector.

5. The data classification method for an e-commerce platform according to claim 1, characterized in that: The step of performing feature fusion processing on the user behavior feature matrix, the commodity knowledge graph and the time series interaction feature vector through a multimodal feature fusion network to obtain a fused feature vector includes: Performing feature space mapping processing on the user behavior feature matrix to obtain a user behavior projection vector, and performing graph structure embedding processing on the product knowledge graph to obtain a product graph vector; Performing mutual attention calculation processing on the user behavior projection vector and the product graph vector to obtain an interactive attention feature, and performing feature alignment processing on the interactive attention feature and the temporal interactive feature vector to obtain an alignment feature matrix; Performing feature supplementation processing on the aligned feature matrix through a residual connection network to obtain supplementary feature data, and performing cross-modal feature extraction on the supplementary feature data to obtain modal interaction features; Performing feature reconstruction processing on the modal interaction feature to obtain a reconstructed feature vector, and compressing and mapping the reconstructed feature vector through an autoencoder to obtain compressed feature data; Performing multi-head attention calculation processing on the compressed feature data to obtain a multi-head feature vector, and performing feature normalization processing on the multi-head feature vector to obtain normalized feature data; Performing feature splicing processing on the normalized feature data and the modal interaction feature to obtain a spliced ​​feature vector, and performing nonlinear transformation on the spliced ​​feature vector through a multi-layer perceptron to obtain transformed feature data; Performing feature selection processing on the transformed feature data to obtain the fused feature vector.

6. The data classification method for an e-commerce platform according to claim 1, characterized in that: The fusion feature vector is classified by a hierarchical attention decision network to obtain an initial classification result, including: Performing multi-level category division processing on the fused feature vector to obtain category hierarchy data, and performing node weight allocation processing on the category hierarchy data to obtain hierarchy node weights; Performing hierarchical attention calculation processing on the fused feature vector and the hierarchical node weights to obtain hierarchical feature weights, and performing adaptive threshold adjustment processing on the hierarchical feature weights to obtain hierarchical discrimination thresholds; Performing hierarchical feature extraction on the fused feature vector through a hierarchical feature extraction network to obtain a hierarchical feature matrix, and performing node correlation calculation processing on the hierarchical feature matrix to obtain node correlation data; Performing path probability calculation processing on the node association data to obtain a classified path probability graph, and performing path pruning processing on the classified path probability graph to obtain a candidate path set; Performing path scoring processing on the candidate path set and the hierarchical discrimination threshold to obtain path score data, and performing multi-path integration processing on the path score data to obtain integrated classification data; The integrated classification data is subjected to confidence evaluation processing to obtain classification confidence, and labels are assigned to the classification confidence through a soft voting mechanism to obtain the initial classification result.

7. The data classification method for an e-commerce platform according to claim 1, characterized in that: The initial classification result is subjected to data classification optimization processing by an interpretable rule processor to obtain target classification data, including: Performing rule extraction processing on the initial classification result to obtain a classification rule set, and sorting the classification rule set by rule importance to obtain rule weight data; Performing rule matching processing on the initial classification result and the rule weight data to obtain a rule coverage matrix, and performing consistency check processing on the rule coverage matrix to obtain rule consistency data; Performing conflict rule identification processing on the rule consistency data to obtain rule conflict data, and performing rule reconciliation processing on the rule conflict data to obtain a reconciliation rule set; Performing decision path tracing processing on the reconciliation rule set to obtain a classification decision path, and performing feature contribution analysis processing on the classification decision path to obtain feature explanation data; Performing classification basis extraction processing on the feature interpretation data to obtain a classification basis vector, and performing label correction processing on the classification basis vector and the initial classification result to obtain a corrected classification label; The modified classification labels are subjected to category structure mapping processing to obtain the target classification data, wherein the target classification data includes the category labels, classification basis descriptions and confidence scores of the commodities.

8. A data classification device for an e-commerce platform, used to implement the data classification method for an e-commerce platform as described in any one of claims 1 to 7, characterized in that: The data classification device for the e-commerce platform includes: The acquisition module is used to perform multi-dimensional feature extraction processing on the user interaction behavior data to obtain a user behavior feature matrix; A processing module, used to perform graph neural network processing on the user behavior feature matrix and product information data to obtain a product knowledge graph; An analysis module, used to perform interaction pattern analysis on the user behavior feature matrix and the product knowledge graph through a temporal attention mechanism to obtain a temporal interaction feature vector; A fusion module, used to perform feature fusion processing on the user behavior feature matrix, the product knowledge graph and the time series interaction feature vector through a multimodal feature fusion network to obtain a fused feature vector; A classification module, used for classifying the fused feature vector through a hierarchical attention decision network to obtain an initial classification result; The optimization module is used to perform data classification optimization processing on the initial classification results through an interpretable rule processor to obtain target classification data.

9. A computer device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor and the memory communicate via the bus. When the machine-readable instructions are executed by the processor, the steps of the data classification method for an e-commerce platform as described in any one of claims 1 to 7 are performed.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the data classification method for an e-commerce platform as described in any one of claims 1 to 7 is implemented.

Images