A Transformer-based approach for Fake News detection using Time Series Analysis

Fake news is a growing problem in the digital age, spreading misinformation and affecting public opinion. Existing fake news detection is based on style analysis or news generator's behavior analysis, the former fails if the content is generated using existing news corpus while the latter requires a very specific data set. In this research, we aim to address the issue of fake news detection by employing deep learning-based time series analysis (TSA). We propose a TSA method to gauge the authenticity of the news based on previously available news content of a similar genre. We employed pre-trained models for encoding including GloVe and BERT and transformer-based sequence-to-sequence (Seq2Seq) for TSA. The results demonstrate 98% accuracy of pre-trained models, such as GloVe and BERT, over traditional encoding approaches having accuracy between 77% and 93%. Our study also compares the effectiveness of various deep learning methods, including Long-Short Term Memory (LSTM), Gated Recurrent Unit (GRU), LSTM with attention, GRU with attention, and Transformers with 8 and 16 multi-heads. The results show that Transformers with 8 and 16 multi-heads achieve 98% and 97% accuracy respectively as compared to LSTM (87%) and GRU (88%). Our work is useful for future research in TSA-based fake news detection and proposes to use GloVe and BERT-based encoding and multi-head transform architecture.