Viability of Sentiment Analysis for Troll Detection on Twitter : A Comparative Study Between the Naive Bayes and Maximum Entropy Algorithms

Abstract: In this study, we investigated whether sentiment analysis could prove to be a viable tool for troll detection on Twitter. The reason why sentiment analysis was analyzed as a possible tool was because of earlier work recognizing it as a feature that could be interesting to examine. By performing two different sentiment analysis methods, Naive Bayes and Maximum Entropy, an idea could be gathered of how well these approaches perform and whether they are viable for troll detection. The data set used was a set of 3000 tweets under the hashtag #BlackLivesMatter. Sentiment analysis was performed on the data set with both the Naive Bayes and Maximum Entropy approaches. The data was then clustered and visually presented in graphs. The results showed that sentiment analysis was not viable as a metric alone. However, together with other metrics it could prove to be useful. Ultimately, for k-means clustering, Maximum Entropy seemed to be the preferable sentiment analysis approach when looking at specific users whereas Naive Bayes performed better when researching individual tweets. As for finding trolls, a general conclusion on the viability of the algorithms could not be drawn, however Maximum Entropy was concluded to be preferable in this specific study.