Using Linguistics and Psycholinguistics Features in Machine Learning for Fake News Classification Through Twitter

Abstract

AbstractThe pinnacle of technology today has allowed for fake news to be spread widely and rapidly through social media channels. The huge volumes of fake news have led to the obsoletion of slow manual fact-checking websites where they are unable to keep up with the speed in classifying fake news. The solution to this lies in automated fact-checking applications which can be automated and scaled to suit the large volume of data. There are still limitations to these as readily available datasets lack multi-dimensional information that can be used to increase the accuracy of predictive model performances when detecting fake news. Since the objectives of fake news writers are to sway readers’ emotions in their favor, there is a possibility that fake news articles display distinct linguistic and psycholinguistic features that affect the human brain and emotions in a certain way. Hence, compared to prior studies, this paper investigates and discovers potential features that can be derived from news texts such as their linguistic and psycholinguistic features that are influential in predicting fake news. This paper is also an expansion of our previous work that used attributes derived from Twitter data to classify fake news. Classification is a supervised machine learning technique used to categorize data into pre-determined classes or outcomes. In this paper, a binary classification model is used to classify news articles into either one of the two outcomes: real or fake news. The machine learning model derived from this research achieved an accuracy of 98.63% in detecting fake news.KeywordsMachine learningData miningClassification modelPredictive modelingFake news classificationText analyticsSocial media analytics