Context-aware social media recommendation based on potential group

Data recommendation as a kind of active mode is more meaningful and important than traditional passive search mode in social media environment. The importance of contextual information has also been recognized by researchers and practitioners in many disciplines, including recommendation system, e-commerce, information retrieval, mobile computing and so on. In this paper, we propose a novel approach for context-aware social media recommendation via mining different granularities of potential groups, called Common Preference Group (CPG). Intuitively, CPG mining is to cluster those users who are interested in any topic set with certain context and have similar affection degree for each topic in the set. It means each user could belong to multiple CPG corresponding to different topic sets. The approach absorbs the characteristic of Collaborative Filtering (CF) technique but overcomes its defects, such as cold-start, data sparseness. Moreover, we build the Tag-Feature Semantic-pairs (TFS) to represent the semantic topics implied in media object to improve the accuracy of CPG mining. To evaluate the efficiency and the accuracy of our approach, we use two datasets: De is a simulated dataset and Dp is a real-life corpus collected from Flickr. The experimental results show the superiority of our approach for social media recommendation.