Turkish Fake News Prediction with Word Embeddings and Transformer-Based Classifiers
摘要
In recent times, the widespread dissemination of fake news through various information-sharing platforms, such as social media and other channels on the internet, has become a significant and growing problem. This issue has underscored the necessity for automatic detection of fake news using artificial intelligence-based solutions, which has emerged as a critical research area, especially when considering the extensive social impact that fake news can have. In this study, we address this need by performing fake news classification through the application of natural language processing and deep learning techniques on a newly created fake news dataset, TurkishNewsCorpus, which has been compiled from Turkish news websites on the internet. The effectiveness of different methods, including TF-IDF and fastText word embeddings-based feedforward neural network models, the fastText classification model, and BERT and DistilBERT transformers, is investigated. Following comprehensive experiments, a Turkish fake news classification model with an impressive 99% F1-score in classification performance was achieved.