Domain Adaptation in Sentiment Classification

Abstract

In this paper we analyse one of the most challenging problems in natural language processing: domain adaptation in sentiment classification. In particular, we look for generic features by making use of linguistic patterns as an alternative to the commonly feature vectors based on ngrams. The experimentation conducted shows how sentiment classification is highly sensitive to the domain from which the training data are extracted. However, the results of the experimentation also show how a model constructed around linguistic patterns is a plausible alternative for sentiment classification over some domains.