ABSTRACT: The interest around automated fact-checking has increased as misinformation has become a major problem online. A typical pipeline for an automated fact-checking system consists of four steps: (1) detecting check-worthy claims, (2) retrieving relevant documents, (3) selecting most relevant snippets for the claim and (4) predicting the veracity of the claim. In this talk, I will talk about the use of state-of-the-art deep neural networks such as LSTMs and Transformer architectures for these steps. Specifically, how deep hierarchical attention networks can be used for predicting the veracity of the claims and how to use the attention weights to extract the evidence for the claims. In addition, I will also talk about how to do check-worthy claim detection using Transformer models. Using several benchmarks from political debates and manual fact checking websites such as Politifact and Snopes, we show that these models outperform strong baselines. I will also summarize the state-of-the-art research within the areas of automated fact-checking and conclude with a set of challenges and problems remaining in this area.
Details to be announced