The Copenhagen Team Participation in the Check-Worthiness Task of the Competition of Automatic Identification and Verification of Claims in Political Debates of the CLEF-2018 CheckThat! Lab.

Abstract

We predict which claim in a political debate should be prioritized<br/>for fact-checking. A particular challenge is, given a debate, how to<br/>produce a ranked list of its sentences based on their worthiness for fact<br/>checking. We develop a Recurrent Neural Network (RNN) model that<br/>learns a sentence embedding, which is then used to predict the checkworthiness<br/>of a sentence. Our sentence embedding encodes both semantic<br/>and syntactic dependencies using pretrained word2vec word embeddings<br/>as well as part-of-speech tagging and syntactic dependency parsing. This<br/>results in a multi-representation of each word, which we use as input to a<br/>RNN with GRU memory units; the output from each word is aggregated<br/>using attention, followed by a fully connected layer, from which the output<br/>is predicted using a sigmoid function. The overall performance of our<br/>techniques is successful, achieving the overall second best performing run<br/>(MAP: 0.1152) in the competition, as well as the highest overall performance<br/>(MAP: 0.1810) for our contrastive run with a 32% improvement<br/>over the second highest MAP score in the English language category. In<br/>our primary run we combined our sentence embedding with state of the<br/>art check-worthy features, whereas in the contrastive run we considered<br/>our sentence embedding alone<br/>