Experiments on Hybrid Corpus-Based Sentiment Lexicon Acquisition

Numerous sentiment analysis applications make usage of a sentiment lexicon. In this paper we present experiments on hybrid sentiment lexicon acquisition. The approach is corpus-based and thus suitable for languages lacking general dictionarybased resources. The approach is a hybrid two-step process that combines semisupervised graph-based algorithms and supervised models. We evaluate the performance on three tasks that capture different aspects of a sentiment lexicon: polarity ranking task, polarity regression task, and sentiment classification task. Extensive evaluation shows that the results are comparable to those of a well-known sentiment lexicon SentiWordNet on the polarity ranking task. On the sentiment classification task, the results are also comparable to SentiWordNet when restricted to monosentimous (all senses carry the same sentiment) words. This is satisfactory, given the absence of explicit semantic relations between words in the corpus.