Based On The Interest And The Protection Of Privacy Online Community Recommendation Technology Research

Online social communities are one of the main ways that users of the internet ob-tain content and interact with each other. Online social communities contain abundant user-generated content, support various kinds of interaction functionalities, and draw the attentions of much more users in recent years. With the explosive growth of users in online social communities, user-generated content have become massive scale. The fast-growing popularity of online social communities and the massive amounts of user-generated content pose a critical need for, and new challenges on content recommender system. Recommendation techniques can predict user interests on unrated items by ana-lyzing user interests in the past. In online social communities, user interests are diverse. Content are generated fast in massive scale. User privacy are easily leaked to the recom-mender server or other third parties in static or dynamic ways. These features of online social communities pose critical challenges for recommendation technique, existing col-laborative filtering recommendation techniques cannot meet users’requirements well in various ways, such as accuracy, privacy or efficiency. Thus, how to expand and optimize existing recommendation techniques to meet the above pressing requirements becomes the frontiers of related research.Faced with the above challenges of recommender systems in online social commu-nities, and combined with recent research achievements in recommender systems, this dissertation makes the following contributions:1. Farseer—an interest-based real-time content recommendation solution is pro-posed to make accurate recommendations to users with diverse interests in real time. Farseer adopts interest groups to accurately identify and organize user interests in on-line social communities. Based on accurate interest groups, an interest-based content recommendation algorithm is proposed to solve the user interest diversification problem. Meanwhile, real-time user context analysis is adopted to capture user online sessions and identify neighbors in real time. Based on real-time user activities, Farseer can update item ratings incrementally, and make recommendations to users on real-time basis. De-tailed experimental results in real online social community demonstrate that Farseer can generate recommendations in real time while achieving better recommendation quality compared with three state-of-the-art collaborative filtering algorithms. 2. Pistis—a privacy-preserving collaborative filtering system for online social com-munities is proposed to make high-quality recommendations without compromising user interest privacy to any parties. Pistis adopts distributed secure multi-party computation (SMPC) to deal with all computations during recommendation process, thus protects user interest privacy from being collected by central server. Moreover, Pistis makes rec-ommendations based on interest groups, which can break the ties between public interests and private interests of users, so that malicious users cannot attack user privacy in on-line social communities. Detailed experimental results in real online social community demonstrate that Pistis can protect user privacy from the central server and other third parties while achieving better recommendation quality compared with state-of-the-art privacy-preserving and non-privacy-preserving collaborative filtering algorithms.3. YANA—an efficient privacy-preserving collaborative filtering system is proposed to address the efficiency issue of privacy-preserving collaborative filtering without com-promising accuracy. YANA is group based-it automatically organizes users into groups with diverse interests such that each user’s private interests can be hidden among a set of users, and the server can only obtain the aggregated information. A number of pseudo users are created for each group, each representing a unique interest and the union of them covers all interests of the group members. The pseudo users communicate with the recommender server on behalf of the real users. The real users can then obtain per-sonalized recommendations based on the server’s recommendations to the pseudo users, without exposing their private data to the server. As SMPCs are only conducted within user groups which are small in size, so that the efficiency issues can be addressed. De-tailed experimental results in real online social communities demonstrate that YANA can effectively protect users’privacy, while achieving high recommendation quality and energy efficiency compared with state-of-the-art privacy-preserving collaborative filtering algorithms.