Information Fusion-Based Deep Neural Attentive Matrix Factorization Recommendation

The emergence of the recommendation system has effectively alleviated information overload problem. However, traditional systems either ignore rich attribute users and items, such as user’s social-demographic features, item’s content etc., facing sparsity problem, or adopt fully connected network to concatenate information, ignoring interaction between information. In this paper, we propose fusion-based deep neural attentive matrix factorization (IFDNAMF) model, which introduces adopts element-wise product different domains learn cross-features when conducting fusion. addition, attention mechanism is utilized distinguish importance on prediction results. IFDNAMF high-order items. Meanwhile, conduct extensive experiments two datasets: MovieLens Book-crossing, demonstrate feasibility effectiveness model.