Deep Collaborative Filtering Combined With High-Level Feature Generation On Latent Factor Model

Recommender System becomes indispensable in the era of information explosion nowadays. Former researchers have noticed the important role of high-level feature playing on semantic factor cases. However, in more common scenes where semantic features cannot be reached, research involving high-level feature on latent factor models is lacking. Analogizing to the idea of the convolutional neural network in image processing, we proposed a Weighted Feature Interaction Network to generate high-level features from the low-level latent factors. An intuitive interpretation is also given to help understand. Then it is integrated into a Deep Collaborative Filtering Model. The results on two real-world datasets show that weighted feature interaction network works and our Deep Collaborative Filtering Model outperforms some conventional and state-of-the-art models. Our work improves the feature representation and recommendation performance on Latent Factor Model.