Detecting fake news by inferring evidential information about claims with text summarization

Abstract

This research explores the potential use of text summarization in fake news detection. Text summarization can assist human fact-checkers in quickly process the massive amounts of content that need to be handled. Yet, its potential benefit in fake news detection has not been explored thoroughly. This paper shows how succinct text summarization might boost automated fake news detection. We utilize two kinds of summarization methods: Extractive and Abstractive. We employ state-of-the-art implementations of these two methods to the fake news dataset and show that condensed information can strengthen the predictive performance of existing fake news detection models. Our work also provides a level of explainability through 3-level evidence screening from the sentence, word to document gradual downsizing.