Improving sentiment analysis for twitter data by handling negation rules in the Serbian language

The importance of determining sentiment for short text increases with the rise in the number of comments on social networks. The presence of negation in these texts affects their sentiment, because it has a greater range of action in proportion to the length of the text. In this paper, we examine how the treatment of negation impacts the sentiment of tweets in the Serbian language. The grammatical rules that influence the change of polarity are processed. We performed an analysis of the effect of the negation treatment on the overall process of sentiment analysis. A statistically significant relative improvement was obtained (up to 31.16% or up to 2.65%) when the negation was processed using our rules with the lexicon-based approach or machine learning methods. By applying machine learning methods, an accuracy of 68.84% was achieved on a set of positive, negative and neutral tweets, and an accuracy of as much as 91.13% when applied to the set of positive and negative tweets.