Abstract
This work deals with the influence of several methods on sentiment analysis. An introduction to previous research is given. Supervised, unsupervised and ensemble classification methods are introduced and evaluated on three datasets. The impact of fuzzy matching, stemming, stop word removal, negation detection and word-sense disambiguation by part-of-speech tagging is examined.
The results indicate that stop words and negations can provide information on sentiment while other methods seem to be of less influence. Especially the removal of stop words can lead to both decline and improvements in classification performance. The positive or negative effect is not only influenced by the classification method but also the applied text domain.
Additional to these investigations, we try to apply sentiment analysis to a new text domain, the BibSonomy reviews on publications and websites.
Users
Please
log in to take part in the discussion (add own reviews or comments).