Multi-modal Recommendation based on Graph Neural Networks

The growth of the internet and AI has caused information overload, making it harder for users to find relevant content. Deep learning-based recommendation systems help by analyzing user-product latent interactions for recommendations. However, these systems have limitations, including their inability to fully capture the nuances of users’ long-term and short-term interests, as well as the persistent problem of data sparsity. This paper aims to improve recommendation systems by using graph neural networks GNNs to leverage more available information. The key contributions include the proposal of a multi-modal GNN recommendation algorithm MHGR, which models capture local user interests. The algorithm outperforms others on several datasets, showing $4.46 \%$ improvement in Recall@20 and $2.42 \%$ in NDCG@20 on the Clothing dataset compared to the second-best algorithm LGMRec.