A Novel Neural Collaborative Filtering Model with Auxiliary Information

In recent years, deep learning has been widely used in personalized recommendation systems. Utilizing deep learning to model user-item complex interaction function is a trend of current recommendation domain. However, most of them only consider the rating information of users and items, ignoring the effect of auxiliary information on the interaction function. In this work, we propose a novel neural network architecture named ANCF, which makes full use of rating and auxiliary information. In contrast to existing deep learning based recommender models that use SDAEs to extract high-level representation of one-hot encoded auxiliary information, ANCF utilizes embedding vector to represent auxiliary information and non-linear neural networks, including attention networks and deep interaction layers, to model user-item interaction function. Extensive experiments on three real-world datasets demonstrate the effectiveness of our proposed ANCF framework.