Abstract

We describe an extension to the technique for the automatic identification and labeling of sentiment terms described in Turney (2002) and Turney and Littman (2002). Their basic assumption is that sentiment terms of similar orientation tend to co-occur at the document level. We add a second assumption, namely that sentiment terms of opposite orientation tend not to co-occur at the sentence level. This additional assumption allows us to identify sentiment-bearing terms very reliably. We then use these newly identified terms in various scenarios for the sentiment classification of sentences. We show that our approach outperforms Turney’s original approach. Combining our approach with a Naive Bayes bootstrapping method yields a further small improvement of classifier performance. We finally compare our results to precision and recall figures that can be obtained on the same data set with labeled data. 1

Citations