We describe an extension to the techniquefor the automatic identification and labelingof sentiment terms described in Turney(2002) and Turney and Littman(2002). Their basic assumption is thatsentiment terms of similar orientationtend to co-occur at the document level.We add a second assumption, namely thatsentiment terms of opposite orientationtend not to co-occur at the sentence level.This additional assumption allows us toidentify sentiment-bearing terms very reliably.We then use these newly identifiedterms in various scenarios for the sentimentclassification of sentences. We showthat our approach outperforms Turney’soriginal approach. Combining our approachwith a Na?ve Bayes bootstrappingmethod yields a further small improvementof classifier performance. We finallycompare our results to precision and recallfigures that can be obtained on the samedata set with labeled data.
展开▼