Claim Retrieval in Twitter

Wenjia Ma, Wenhan Chao, Zhunchen Luo, Xin Jiang

Abstract: Controversial topics, especially the new emerging ones are widely discussed and searched in social medias like Twitter. When people are interested in topics and search on Twitter, high quality tweets are expected to appear at the top. Since it is only argumentation that truly reasons things out, we believe that high quality tweets are those with argumentation that consists of claim and evidence. Moreover, claim is the heart of argumentation, we concentrate on claim retrieval in Twitter. Based on a learning-to-rank framework, we integrate Twitter structural information and topic-independent claim-related lexicon to re-rank the relevant tweet list pre-retrieved by BM25 scores. We also automatically construct topic-dependent claim-oriented lexicons to further elevate the retrieval performance. Additionally, our model can be easily adapted to new topics without any manual process or external information, which guarantees the practicability of our model.