Bidirectional LSTM Based on POS tags and CNN Architecture for Fake News Detection

Abstract

Fake news generally on social media spreads very quickly and this brings many serious consequences. Traditional lexico-syntactic based features have limited success to detect fake news. Majority of fake news detection techniques are tested on small dataset containing limited training examples. In this work, we evaluate our architecture on Liar-Liar dataset which contain 12836 short news from different sources including social media. The proposed architecture incorporates POS (part of speech) tags information of news article through Bidirectional LSTM and speaker profile information through Convolutional Neural Network. The results show that the resulting hybrid architecture significantly improves detection performance of Fake news on Liar Dataset.