Presence or Semantic Information in Sentiment Classification?

This paper analyses the implications in the use of a content vector based on the presence or the semantic information of the features that represent an opinion. In our phrase pattern-based method, we automatically construct semantic lexicons to determine the semantic orientation of each feature, that is, the degree of subjectivity associated with each particular n-gram. Using two different datasets with two different learning models: our unsupervised learning approach as well as the use of Bayesian learning methods, our results show that it is possible to maintain a state-of-the art classification accuracy.