Progressive Fusion of LLM-Augmented Multi-Modal Data for Enhanced Product Recommendations

Recommender systems in E-commerce often face challenges such as data sparsity and cold-start problems, which hinder their ability to provide accurate and personalized recommendations. To address these issues, we propose Progressive Fusion Muti-modal Graph Convolution Networks (PFMGCN), a novel framework tailored for product recommendations. In PFMGCN, we first employ a Large Language Model (LLM) to generate more product attribute data for enriching contextual information. A knowledge graph (KG) is then constructed with augmented nodes, and their embeddings are extracted via pretrained models. A cross-attention module is used to dynamically fuse textual and visual features, capturing fine-grained product details. Progressive graph convolutional networks integrated with multi-modal cross-attention are then applied to extract deep features and deliver accurate recommendations. Extensive experiments on real-world E-commerce datasets demonstrate that our approach significantly outperforms state-of-the-art methods, achieving superior accuracy in product recommendation tasks.