Afaan Oromo Fake News Detection on Social Media: - Using Deep Learning Approach

Abstract Due to the rapid growth of the internet in recent years, social media has made it easier to create and share information via computer-mediated technologies. As a result of the social media revolution, people's communication and acquaintances have changed. The majority of people nowadays use social media to find and consume news rather than traditional news agencies. On the other hand, while online social media has grown in importance as a source of information and a method of connecting people together, detecting fake information remains a difficult problem to address. The majority of previous works presented appropriate machine learning models of fake news detections are concerned with specific group of language such as English. Despite the fact, the problem is not constricted to language. As a result, we implemented deep learning models on the Afaan Oromo language news dataset extracted from twitter and Facebook pages, to detect fake news on social media. In this study we annotated dataset of 1838 news text for train and testing purposes. Three popular deep learning methods used in the experiments: LSTM, GRU, and Bi-LSTM, to determine the best performing models, and each of the model performance is evaluated with various parameters. Then, when it came to recognizing fake news in news content written in the Afaan Oromo language, bidirectional-LSTM beat the other algorithms, namely unidirectional LSTM and GRU. The model is capable of making predictions on our dataset, with an f-1 score of 90%. Finally, python flask API server with spyder IDE was used for the web-based prototype development of the trained Bi-LSTM models for prediction of news.