60 - FEVER: a large-scale dataset for Fact Extraction and VERification, with James Thorne
NLP Highlights
00:00
The Baseline System for Refuted Claims
Our baseline system is a pipeline of two components. We first retrieve the right evidence, and then we do a classification as to whether the evidence supports the refutes the claim. So by default, and not enough, not enough evidence, claim, if we don't find any evidence, we could mark that as not enough evidence. But in reality, with our retrieval system, we're finding noisy information from Wikipedia, which we think is right, but doesn't fully support a refuted claim. so we've had to negatively sample evidence in that case and train a three way classifier.
Transcript
Play full episode
