Effectiveness gain of polarity detection through topic domains
Résumé
Most of the work on polarity detection consists in finding out negative or positive words in a document using sentiment lexical resources. Indeed, some versions of such approaches have performed well but most of these approaches rely only on prior polarity of words and do not exploit the contextual polarity of words. Sentiment semantics of a term vary from one domain to another. For example, the word "unpredictable" conveys a positive feeling about a movie plot, but the same word conveys negative feeling in context of operating of a digital camera. In this work, we demonstrate this aspect of sentiment polarity. We use TREC Blog 2006 Data collection with topics of TREC Blog 2006 and 2007 for experimentation. The results of our experiments showed an improvement (95%) on polarity detection. The conclusion is that the context plays a role on the polarity of each word.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...