Evidence extraction to validate medical claims in fake news detection

Fact-checking of online health information has become necessary due to the increasing usage of internet by people searching for medical advice. There is a plethora of false information available to the public, which can put people in harm’s way. In order to aid the factchecking process, recent research has leveraged the advancements made in NLP and deep learning techniques. Majority of the existing technology relies on the existence of labelled data, which is very limited. In this work we explored an unsupervised approach to identifying evidence sentences, which is the key task in claims verification process. We show by performing experiments on a publicly available dataset that our method achieves performance comparable to that of state-of-the-art supervised techniques. We also show how our proposed method can be adapted incases where labelled data is available.