Hybrid Rating Prediction using Graph Regularization from Network Structure

To understand users’ preference and business performance in terms of customer interest, rating prediction systems are wildly deployed on many social websites. These systems are usually based on the users past history of reviews and similar users. Such traditional recommendation systems suffer from two problems. The first is the cold start problem. In many cases, the system has no knowledge about the interests of a new user. Second, the data sparsity problem, some users may have a few business ratings. However, a noticeable trend in social life is that people sharing characteristics and attributes tend to stay friends so that knowledge about the person’s neighborhood structure demonstrates his preference in certain degree. The traditional recommendation approaches fail to take network structure of users into consideration. To address the aforementioned three problems, we proposed a hybrid method based on network structure and collaborative filtering approach. Specifically, a graph regularization is learned by random walk algorithm which discovers the underlying network structure and the matrix factorization based collaborative filtering is used to predict user ratings. The experiments conducted on Yelp userbusiness reviews dataset demonstrate the advantages of our proposed method beyond baselines and validate the contribution of its two components. Keywords—Social Network, Data Mining, Model, Algorithm, Recommendation System, Hybrid method