Inclusion of Semantic and Time-Variant Information Using Matrix Factorization Approach for Implicit Rating of Last.Fm Dataset

摘要

Linked Open Data provide an opportunity to employ openly available metadata for the use of various applications. Recently it has been received great attention from the Recommender System community, to incorporate this semantic data as side information for getting more accurate results. Despite the popularity of these Recommender Systems, it suffers from the problem of data sparsity and high dimensionality. Moreover, the utilization of semantic data becomes a bigger challenge, when user preferences are defined implicitly, instead of in a fixed rating scale. The above-mentioned challenges necessitate us to introduce a comprehensive framework that is able to leverage the additional source of information for boosting the accuracy of the existing systems. In this paper, we have proposed a modified Joint Matrix Factorization approach for incorporating semantic information related to items and tag-based information with an implicit user preference for boosting the accuracy of the overall system. The model adheres to the phenomena of time-variant Recommender System; thus, it also utilizes the time-related information of items. Experimental results show that our method gets more accurate recommendation results with faster converging speed than other existing Matrix Factorization-based approaches.