With the development of the economy and society and the deepening of economic globalization, traditional cross-border e-commerce data management methods are no longer able to meet practical needs. In response to this, research is conducted on the title and attribute data of popular cross-border e-commerce products, and natural language processing models (BERT) and deep autoencoder models are improved to construct a multi-dimensional cross-border e-commerce data visualization platform based on deep learning. The specific innovation includes three aspects. In terms of model construction, research is conducted on optimizing the vocabulary extension layer and multi head attention mechanism of BERT, and combining it with a deep autoencoder with classification constraints to construct a multi-dimensional cross-border e-commerce data classification model (MDCCBDL) based on deep neural networks, achieving deep fusion of title and attribute features. In terms of performance validation, the model achieved an average accuracy of 86.3% and an F1 score of 83.7% on six Amazon datasets, which were 6.2% and 4.7% higher than the baseline model with a single data, respectively; On 9 cross-platform datasets including eBay and Wish, the average accuracy was 87.6%, verifying the generalization ability. The recall rate and average accuracy are 84% and 97%, respectively, which are 7.7% and 3.2% higher than those of deep interest networks. In terms of platform applications, the visualization platform achieved an average response time of 0.79 s and an error rate of 0.35% in 1000 concurrent multi-environment tests. The product selection time was shortened by 43.9% compared to traditional methods, and the success rate increased to over 80%. The research results indicate that the model improves classification accuracy and representation ability through multi-dimensional feature fusion, and the platform has time and efficiency advantages in cross-border e-commerce product selection management, providing a practical solution for intelligent management of industry data.
Jingjing Wu (Fri,) studied this question.