Özet:

Social media platforms become indispensable channels to discover the latest news by the Internet users. Millions of news is broken first, spread faster, and reach larger communities on these platforms in a very short time compared to traditional media organs. However, in contrast to traditional media, social media platforms lack of security in terms of control mechanisms to verify the reliability and accuracy of the disseminated news. This brings the need for automatic fake news detection systems for these platforms to prevent or reduce spread of false information. In this paper, we study the problem of fake news detection on social media for two languages, both of them having distinct linguistic features: Turkish and English. In this regard, we create the first real-world public dataset of Turkish fake and real news tweets, named SOSYalan, to the best of our knowledge. For English language, we carry out experiments with two benchmark datasets, BuzzFeed and ISOT. We develop deep learning based fake news detection systems for both of Turkish and English languages based on convolutional neural networks (CNN), and recurrent neural networks-long short term memory (RNN-LSTM) approaches with Word2vec embedding model. We evaluate the developed systems in terms of accuracy, precision, recall, F1-score, true-negative rate, false-positive rate, and false-negative rate metrics. The results demonstrate that the developed systems for English language produce higher accuracy rates compared to the most of the existing state-of-the-art studies. Additionally, the results confirm the superiority of our systems developed for Turkish language in comparison to very few studies conducted in this area.

Anahtar Kelimeler: