Website-oriented recommendation based on heat spreading and tag-aware collaborative filtering

Recently, Recommender Systems has been widely applied in helping users find potentially interesting items from the era of big data. However, most of researches on this topic have focused on estimating the direct relationships between users and items, neglecting other available information. In this paper, we discuss about mining webs with information extracted from search and browser logs of users. In particular, we utilize the keywords correlated with corresponding websites by Singular Value Decomposition (SVD) technique to model users features and propose the tag-aware k-nearest neighbor Collaborative Filtering (CF). We then build a hybrid recommendation method to help people accurately find websites by employing Heat Spreading (HeatS) method. Experimental results demonstrate that the hybrid method outperforms baseline algorithms at the global ranking metric.