Similarity-Guided Multimedia Recommendation in Heterogeneous Information Network

With the rapid growth in multimedia information, problems on how to discover the individual interests of users and recommend them with the proper goods have become increasingly difficult. Traditional recommendation algorithms simply utilize user rating logs for recommendations, but ignore lots of useful information which can be expressed as a Heterogeneous Information Network. In this paper, we propose a similarity measure, PW- PathSim, to calculate the relevance between two entities of the semi-symmetric weighted meta paths. Then a similarity regularization based recommendation algorithm is proposed to integrate the similarity of users and items with matrix factorization for recommendations. Furthermore, we compare the PWMFP algorithm with several benchmarks including FunkSVD, HeteFM and DSR. Experimental results with Douban dataset show that it outperforms other HIN-based algorithms in terms of recommendation accuracy.