Truth Discovery by Claim and Source Embedding

摘要: Information gathered from multiple sources on the Web often exhibits conflicts. This phenomenon motivates need of truth discovery , which aims to automatically find true claim among conflicting claims. Existing methods are mainly based iterative updates, optimization or probabilistic models. Although these have shown their own effectiveness, they a common limitation. These do not model relationships between each pair source and target such that well capture underlying interactions in data. In this paper, we propose new for discovery, learning representations claims targets. Our first constructs heterogenous network including source-claim, source-source truth-claim relationships. It then embeds into low dimensional space trustworthy close. way, can be conveniently performed embedding space. Moreover, our implemented both semi-supervised un-supervised manners deal with label scarcity problem practical discovery. Experiments three real-world datasets demonstrate outperforms existing state-of-the-art