Community and trust-aware fake media detection

Abstract

Nowadays, it becomes increasingly difficult to find reliable multimedia content in the Web 2.0. Open decentralized networks (on the Web) are populated with lots of unauthenticated agents providing fake multimedia. Conventional automatic detection and authentication approaches lack scalability and the ability to capture media semantics by means of forgery. Using them in online scenarios is computationally expensive. Thus, our aim was to develop a trust-aware community approach to facilitate fake media detection. In this paper, we present our approach and highlight four important outcomes. First, a Media Quality Profile (MQP) is proposed for multimedia evaluation and semantic classification with one substantial part on estimating media authenticity based on trust-aware community ratings. Second, we employ the concept of serious gaming in our collaborative fake media detection approach overcoming the cold-start problem and providing sufficient data powering our Media Quality Profile. Third, we identify the notion of confidence, trust, distrust and their dynamics as necessary refinements of existing trust models. Finally, we improve the precision of trust-aware aggregated media authenticity ratings by introducing a trust inference algorithm for yet unknown sources uploading and rating media.