Summary: | To deal with the inherent data sparsity and cold-start problem, many recommender systems try to exploit the textual information for improving prediction accuracy. Due to the significant progress of deep learning techniques, neural network-based content modeling methods have been investigated in recent studies. However, most of these existing methods often assume that users are independent and identically distributed (i.i.d), and the social influence is not considered. However, in the real world, we always turn to our friends for recommendations, and the closer the friendship between the friends, the greater the social impact is. These methods only exploit the reviews from an item perspective and rarely consider the user's reviews to capture the user's interests, but in reality, users often express their preferences by posting different reviews to different items. Based on the above-mentioned observations, we propose a social-enhanced content-aware recommendation method by fusing the social network, item's reviews, and user's reviews in a unified framework. Specifically, to better model the item's reviews, we first introduce the convolutional matrix factorization (ConvMF) as our basic recommendation framework, which utilizes convolutional neural network (CNN) to capture the deeper understanding of the content context. Then, to consider the user's social influence, we integrate the user's social network into ConvMF by a shared user latent factor, which can bridge the user's social interests and user's general preferences in the same latent space. To model the user's reviews, similar to ConvMF, we exploit another CNN to learn a deeper understanding of the user's posted contents. Finally, we conduct experiments on the real-world dataset Yelp to demonstrate the effectiveness of our method. The experimental results indicate that our proposed method can effectively model the social and the review information and outperforms other related methods in terms of root mean squared error (RMSE) and mean absolute error (MAE).
|