Evaluating the Effectiveness of Hashtags as Predictors of the Sentiment of Tweets

Recently, there has been growing research interest in the sentiment analysis of tweets. However, there is still a need to examine the contribution of Twitter-specific features to this task. One such feature is hashtags, which are user-defined topics. In our study, we compare the performance of sentiment and non-sentiment hashtags in classifying tweets as positive or negative. By combining subjective words from different lexical resources, we achieve accuracy scores of 83.58 % and 83.83 % in identifying sentiment hashtags and non-sentiment hashtags, respectively. Furthermore, our accuracy scores surpass those scores obtained using models that apply a single lexical resource. We apply derived properties of sentiment and non-sentiment hashtags, including their sentiment polarity to classify tweets. Our best classification models achieve accuracy scores of 81.14 % and 86.07 % using sentiment hashtags and non-sentiment hashtags, respectively. Additionally, our models perform comparably to supervised machine learning algorithms, and outperform a scoring algorithm developed in a previous study.