Why Words Alone Are Not Enough: Error Analysis of Lexicon-based Polarity Classifier for Czech

Lexicon-based classifier is in the long term one of the main and most effective methods of polarity classification used in sentiment analysis, i.e. computational study of opinions, sentiments and emotions expressed in text (see Liu, 2010). Although it achieves relatively good results also for Czech, the classifier still shows some error rate. This paper provides a detailed analysis of such errors caused both by the system and by human reviewers. The identified errors are representatives of the challenges faced by the entire area of opinion mining. Therefore, the analysis is essential for further research in the field and serves as a basis for meaningful improvements of the system.