Hybrid Recommendation Algorithm Based on Latent Factor Model and PersonalRank

Abstract

To promote the development of future social networks, an efficient recommendation algorithm is needed. Due to high-dimensional structures and missing rating information of the user-item matrix caused by sparse data, the direct use of a single recommendation algorithm is inefficient. In order to improve the accuracy of personalized recommendations, we propose an improved hybrid recommendation algorithm based on the latent factor model (LFM) and the PersonalRank algorithm, which adopts a cascade mixing approach with features expanding. First, it fills in a sparse matrix using LFM for the users who did not evaluate for the items. Next, it sets up a graph model and uses the PersonalRank algorithm in the filling matrix. Finally, it computes PR values of each user for items and implements sorting. The efficiency of the hybrid algorithm was verified on the MovieLens dataset. Compared with the PersonalRank algorithm, it can improve the accuracy rate and the recall rate of top-N recommendations.