A Hybrid DistilBERT-BiGRU Model for Enhanced Misinformation Detection: Leveraging Transformer-Based Pretraining Language Model

Misinformation spreads quickly and poses a serious threat to society since it has the power to influence elections, change the public's views, and incite disturbances. The capacity to recognize fake information in a variety of languages is crucial for the survival of an informed and trustworthy information ecology. The article proposes a transformer-based pretrained model for the detection of fake news in English utilizing Distilled Bidirectional Encoder Representations from Transformers (DistilBERT) a distilled version of the BERT model with Gated Recurrent Unit (GRU). A sizable dataset of English news stories pertaining to politics is collected that includes examples of both real and fake news. The text preprocessing technique is employed using NLTK library. Further, using this dataset, the DistilBERT model is optimized to grasp the semantic links and contextual data discovered in the English language by fine-tuning the model. The input text is encoded into contextualized embeddings. After that, one or more BiGRU layers are applied to the DistilBERT output. To capture the dependencies between neighbouring tokens in the text, these Bidirectional GRU layers are used to process the sequential information. As demonstrated by the experimental results, the proposed model has achieved 97.26% accuracy rate in identifying English fake news.