Abstract
Privacy intrusion has become a major bottleneck for current trust-aware social sensing, since online social media allows anybody to largely disclose their personal information due to the proliferation of the Internet of Things (IoT). State-of-the-art social sensing still suffers from severe privacy threats since it collects users' personal data and disclosure behaviors, which could raise user privacy concerns due to data integration for personalization. In this paper, we propose a trust-aware model, called the User and Item Similarity Model with Trust in Diverse Kinds (UISTD), to enhance the personalization of social sensing while reducing users' privacy concerns. UISTD utilizes user-to-user similarities and item-to-item similarities to generate multiple kinds of personalized items with common tags. UISTD also applies a modified k-means clustering algorithm to select the core users among trust relationships, and the core users' preferences and disclosure behaviors will be regarded as the predicted disclosure pattern. The experimental results on three real-world data sets demonstrate that target users are more likely to: (1) follow the core users' interests on diverse kinds of items and disclosure behaviors, thereby outperforming the compared methods; and (2) disclose more information with lower intrusion awareness and privacy concern.