Credibility Assessment of User Generated health information of the Bengali language in microblogging sites employing NLP techniques

Abstract

The use of the internet, particularly social networking platforms have been pandemic in recent years for searching, publishing and engaging about health related information, thus making way for the spread of misleading, superfluous or fake health news. Social networks have a powerful influence on individuals in making health decisions. Fake health news spread like a virus because clickbaity content is easier to digest rather than consuming dense scientific, observational material. Bangladesh has also been impacted by such fabricated, sensationalized, evidenceless health news, written in Bangla language on social networks. Although there has been numerous work on tackling fake health news in the English language, to our knowledge, no prior work has been done related to online health information analysis and their authenticity in Bangladesh, making our work a first of its kind. We created a novel dataset of Bangla health news, collected from Twitter. Taking textual features into account, the data was fed to a fake news detection system, which employed state of the art machine learning classifiers and neural network models to predict a label in a fixed category, thus achieving a maximum accuracy of 91% using CNN with fasttext embeddings.