Abstract:Recommender systems are tools that support personalization in terms of supporting navigation, sharing, and discovery of information and help users to find their desired content over the large volume of information. Recently, new research area on context-aware recommendations has emerged to provide the capability of utilizing social contents and exploit related tags and rating information and personalize the search for desired content by considering user's actual situation (contextual information). In this study, we propose an approach for clustering contextually similar information using unsupervised learning approach through K-Medoids clustering and demonstrate the extraction of latent preferences for recommending items under a given contextual cluster and study how such clusters of similar contextual information can be exploited to improve the prediction accuracy of a context-aware recommender systems. To evaluate the performance of our proposed recommendation strategy, the empirical analysis is conducted on the popular LDOS-CoMoDa dataset and we showed that our proposed approach outperforms state-of-the-art algorithms in terms of prediction accuracy of the computed recommendations.