Context-aware social recommendation in sharing community

The social recommendation has attracted great attention due to its wide applications in domains such as entertainment, online news broadcasting etc. Since contexts highly affect social user preferences, great efforts have been put into context-aware recommendation in recent years. However, it is still challenging to address the problem of effective and efficient social recommendation due to the huge data volume and extremely complex social contexts. In this thesis, context-aware recommendation approaches are proposed in three different application scenarios: individual recommendation, social group recommendation and online recommendation over streams. In the individual recommendation, existing techniques cannot capture the optimal context information from a large number of available features flexibly for the effective and efficient context-aware social recommendation. To address this issue, we propose a generic framework for the context-aware recommendation in sharing communities, which exploits the characteristics of media content and contexts. First, our proposed novel approach can fully removes the redundancy among context information. It selects the optimal features used in recommendations based on the correlation between a feature and a group of other ones for selecting. Then, our proposed graph-based model called content-context interaction graph (CCIG) is compact and extendible when representing the media by analysing the metadata content and social contexts, and the interaction between attributes. Finally, our proposed hash-based index over Apache Storm for organizing and searching the media database guarantees recommendation is generated in real time. In the social group recommendation, existing techniques mainly focus on small user groups. However, online sharing communities have enabled group activities among thousands of users. Accordingly, recommendation over big groups has become critical. We propose a new framework to accomplish this goal by exploring the group interests and the connections between group users. Compared with the existing work, the novelties of our proposed techniques are as follows. First, our proposed framework can fully mine various group preferences based on the user and item interactions rather than treating the preference of the whole group as unique. Then, our proposed framework improves the recommendation efficiency dramatically. The traditional recommendation approaches, such as collaborative filtering, generate recommendations based on the whole media in the database, resulting in much time cost. To reduce the time cost, our framework collects a comparably compact potential candidate set of media-user pairs, on which the collaborative filtering is performed to generate an interest subgroup-based recommendation list. Third, a novel aggregation function is proposed to integrate the recommended media lists of all interest subgroups as the final group recommendation result. It addresses the dynamic combination challenges in big group recommendation by considering both the subgroup contribution and activeness. In online recommendation over streams, we propose a novel framework for the social recommendation over streaming environments effectively and efficiently. Our proposed novel Bi-Layer Hidden Markov Model (BiHMM) precisely predicts user’s interest over streams by adaptively capturing the behaviors of social users and their interactions with the help of influential official accounts to predict their long-term and short-term interests. Meanwhile, existing techniques are not effective for handling social users with diverse interests. To address this problem, we design a new probabilistic entity matching scheme for effectively identifying the relevance score of a streaming item to a user. Following that, we propose a novel indexing scheme called CPPse-index for improving the efficiency of our solution since existing approaches for recommending items to a particular user are not efficient when applied to a huge number of users over high-speed streams