Disinformation Classification Using Transformer based Machine Learning
Mohammed Al-Alshaqi, Danda B. Rawat · 2023
The proliferation of false information via social media has become an increasingly pressing problem in recent years when it comes to identifying authentic facts. Social media is known as the primary source of influencing people. The identification of fake news has been a topic of interest in academia, aiming to report on the questionable authenticity behind shared facts over the internet. The term ‘fake news' refers to unauthentic news that presents baseless facts to create propaganda and influence the perception of the audience. Consequently, an automatic approach to detecting fake news has been suggested as a viable solution to curbing the spread of disinformation. However, developing an automated solution for fake news detection raises concerns due to the high risk of inaccurately classifying data. Therefore, improving classification schemes has been recognized as a relevant solution for accurately interpreting the data. In this regard, our research work focuses on developing a transformer-based fake news classification model. Many machine learning models have been reported to lack a word embedding process. Thus, we employ transformers to extract word embeddings from news articles. The proposed scheme is compared with the baseline Hybrid CNN and RNN techniques using matrix indicators such as the F1-score, precision, recall, and accuracy. Based on the results obtained from our presented classifier model, the proposed methodology outperforms earlier approaches, achieving significantly enhanced accuracy in the detection of fake news.