Design and Research of Cross-Border E-Commerce Short Video Recommendation System Based on Multi-Modal Fusion Transformer Model

This study designed a cross-border e-commerce short video recommendation system based on Transformer's multimodal analysis model. When mining associations, the model not only focuses on the relationships between modalities, but also improves semantic context by addressing contextual correlations within and between modalities. At the same time, the model uses a cross modal multi head attention mechanism for multi-level association mining, and constructs an association network interwoven with latitude and longitude. In the process of exploring the essential correlation between patterns and subjective emotional fluctuations, the potential context between patterns has been realized. Fully explore correlations and then more accurately identify the truth contained in the original data. In addition, this study proposes a self supervised single modal label generation method. When multimodal labels are known, it does not require complex deep networks and only relies on the mapping relationship between multimodal representations and labels to generate a single modal label. Modal labeling can achieve phased automatic labeling of single modal labels, and quantify the mapping relationship between modal representations and labels from the representation space to generate weak single modal labels. The study also achieved multimodal collaborative learning in the context of limited differential information acquisition due to incomplete labeling, fully utilizing multimodal information. The experimental results on classic datasets in the field of multimodal analysis show that it outperforms the baseline model in terms of accuracy and F1 score, reaching 98.76% and 97.89%, respectively.

@article{Hu2024,
  title     = {Design and Research of Cross-Border E-Commerce Short Video Recommendation System Based on Multi-Modal Fusion Transformer Model},
  journal   = {International Journal of Advanced Computer Science and Applications},
  volume    = {15},
  number    = {8},
  year      = {2024},
  publisher = {The Science and Information Organization},
  author    = {Yiran Hu},
  doi       = {10.14569/IJACSA.2024.0150840},
  url       = {https://doi.org/10.14569/IJACSA.2024.0150840}
}