BERT for Evidence Retrieval and Claim Verification

We investigate BERT in an evidence retrieval and claim verification pipeline for the task of evidence-based claim verification. To this end, we propose to use two BERT models, one for retrieving evidence sentences supporting or rejecting claims, and another for verifying claims based on the retrieved evidence sentences. To train the BERT retrieval system, we use pointwise and pairwise loss functions and examine the effect of hard negative mining. Our system achieves a new state of the art recall of 87.1 for retrieving evidence sentences out of the FEVER dataset 50K Wikipedia pages, and scores second in the leaderboard with the FEVER score of 69.7.