Deep neural approach to Fake-News identification

摘要

The complexities of fake news detection cannot be overcome solely with Natural Language Processing. Even a human being finds it difficult to decide the authenticity of an article without further fact checking. Hence a Deep Learning model entirely based on NLP is bound to have huge limitations. In order to address this shortcoming, the proposed system additionally includes a live data stage mining which provides secondary features. These features include source domains of the article, author names etc.. Since these features mimic the process of fact-checking to an extent, the model is expected to outperform existing models that are solely based on NLP. We seek to compare the results from models with and without secondary mined features. LSTM and FF Neural Networks are explored. Additionally, effectiveness of different word vector representations in relation to this problem are also investigated.