Influence-Aware Truth Discovery

Today is the age of information explosion. Information for the same entity can come from different sources and errors become inevitable. To reduce the errors made by individual sources, aggregation methods are widely applied to identify the trustworthy information. Among these aggregation methods, truth discovery has been proposed to improve the aggregation results by estimating the source expertise simultaneously. Intuitively, if a piece of information is from a source of high expertise, then it is more trustworthy, and in return, the source that provides trustworthy information has higher expertise. Many truth discovery methods assume that sources make their claims independently, which may not be true in practice. In fact, the influence among sources are ubiquitous, and the claims made by one source may be influenced by others. Although there is some work that considers source correlation, those methods limit the claims to be categorical type, which is not general enough to handle the complicated real world applications. In this paper, we proposed a Influence-Aware Truth Discovery (IATD) model to tackle these challenges. The proposed method utilize an unsupervised Bayesian model and can take the source correlation as prior for influence derivation. To model the influence among sources, we introduce claim expertise as a fusion of the expertise of the source who provides the claim and the expertise of its influencers'. The proposed IATD model can also handle different data types using different distributions in the Bayesian model. Experiments on real-world datasets show that IATD model can significantly improve the performance of the aggregation results compared with the state-of-the-art truth discovery approaches since it considers the inter-source influence as well as different data types. The properties of IATD model are further illustrated using simulated datasets.