Mining for Fake News

Abstract

Fake news is an ever-growing concern in the modern age of the internet. Discerning fake information from the truthful is an important task given the simplicity of sharing information digitally. In this paper, we present a data mining solution to classify articles as real or fake by using bag-of-words (BoW) and sequential mining techniques, and compare reliability for detecting fake news on various datasets. Specifically, our solution first cleans the input news by normalizing words and removing “filler” words. It then uses the BoW or sequential mining techniques to vectorize cleaned data. Afterwards, it trains the classification models based on vectorized data and classifies unseen news as real or fake. Evaluation on real-life data shows the feasibility of our solution to mine and classify fake news.