Hybrid Recommendation of Articles in Scientific Social Networks Using Optimization and Multiview Clustering

Abstract

Several published articles are shared every day through scientific social network, which make it very difficult for researchers to find highly valuable and appropriate articles. To solve this issue, we propose in this paper a novel hybrid recommendation of articles combining an improved version of the content based filtering (CBF) and the collaborative filtering (CF) algorithms. First, the profiles of researchers and articles are built integrating the social tag information into the CBF algorithm. Then, the social friend information was integrated into the CF algorithm. Due to the problem sparsity of the CF, we have considered the singular matrix factorization (SVD) and unified probabilistic matrix factorization (PMF) algorithms. Finally, in order to further improve the performance of the CF, an optimized clustering has been applied using the Kmedoids algorithm and the BAT meta-heuristic. Different hybridization have been proposed: (1) a weighted hybrid algorithm which combines the two improved versions of the CBF and CF algorithms; and (2) the multiview clustering based hybrid algorithm. Experimental results conducted on the CiteULike dataset demonstrate that the proposed approach has significantly improved the recommendation accuracy and outperforms the baselines and exiting methods.