BigIR at CLEF 2018: Detection and verification of check-worthy political claims

Abstract

With the enormous amount of misinformation spread over the Internet, manual fact-checking is no longer feasible to prevent its negative impact. There is an urgent need for automated systems that can make fact-checking process faster and effectively detect the veracity of claims. In this paper, we present our participation in the two tasks of CLEF-2018 CheckThat! Lab. To rank claims based on their check-worthiness (Task 1), we propose a learning-to-rank approach with features extracted by natural language processing such as named entity recognition and sentiment analysis. For veracity prediction (Task 2), we propose using an external Web search engine to retrieve potentially-relevant Web pages and extract features from relevant segments of those pages to predict the veracity. In the official evaluation, our best performing runs for Task 1 are ranked 4th (out of f 6 runs from 8 teams) and 1st (out of 5 runs from 2 teams) over the English and Arabic datasets respectively, while our best performing run for Task 2 is ranked 6th (out of 10 runs from 5 teams) over the English datasets.