On Modelling for Bias-Aware Sentiment Analysis and Its Impact in Twitter
Ahsan Mahmood, Hikmat Ullah Khan, Muhammad Ramzan
Abstract
Sentiment Analysis (SA) is an active research area for the last ten years. SA is the computational treatment of opinions, sentiments, and subjectivity of text. Twitter is one of the most widely used micro-blog and considered as an important source for computation of sentiment and of data analysis. Therefore, companies all over the world analyze Twitter data using SA and extract knowledge which has potential applications in diverse areas. Although SA is the successful way of finding the people’s opinion, the bias in the tweets affects the results of the SA and reflects inaccurate analysis that may mislead users to take erroneous decisions. The biased tweets are shared by valid, but biased human users as well as the social bots to propagate the biased opinions on certain topics. To counter this, this research study proposes a statistical model to identify such users and social bots who share the biased content in the form of tweets in the Twitter social media. For experiment purpose, we use annotated twitter dataset and argue the results of SA with and without the biased tweets and explored the effects of biased users at micro-level and macro level. The empirical results show that the proposed approach is effective and properly identifies the biased users and bots from other authentic users using sentiment analysis.