A Stylometric Approach for Reliable News Detection Using Machine Learning Methods

The rise of mass media and the Internet has resulted in a sudden surge in the spreading of news swiftly within the current society. Even though communication of information through online news has made people more informed of their surroundings, it has also caused the rise of a misinformed society. The characteristic which assists this cause has been the ineffectiveness of humans in distinguishing truthful and untruthful news. The untruthfulness of a news article can be determined by analyzing the content of the article for any untruthful or unreliable claims. Currently, this task is undertaken by two means. Namely, by manual fact-checking or by automated fake news detection. Even though in previous literature automation of fake news detection has been successful, we hold the stance that assessing the correctness of claims is a complex task that AI alone cannot solve satisfactorily at this point, and it demands human intervention which is too laborious. Therefore, this study identifies a model that can distinguish between reliable and unreliable news articles using stylometric features. The model introduced in the study can act as a filter in the fake news detection pipeline such that only the reliable articles are pushed to the endpoint of the pipeline. A novel dataset is introduced through this research for news reliability detection tasks consisting of news articles directed at Sri Lanka. The study demonstrates that stylometric features can be employed to predict the reliability of news articles. A Stacking Model attained the highest accuracy which stood at 79.05% in a space of 31 stylometric features.