LAUSR

Sentiment lexicon, which provides sentiment information for words, plays an important role in sentiment analysis task. Currently, most of sentiment lexicons have only one sentiment polarity for each word and ignore sentimental ambiguity. In this paper, we propose to construct the sentiment lexicon based on context-dependent part-of-speech (POS) chunks, namely CP-chunks, which aims at solving the ambiguity of lexical sentiments. Given that the POS of context has impact on the word polarity and intensity, we take CP-chunks as an unit to do sentiment calculation. Our method is evaluated through the classification task of text sentiment. The experiment results indicate that, in comparison with the existing methods, the applicability of our method is more stable and balanced for both the positive and negative polarities corpora, and the accuracy of our method reaches 82% for the sentiment classification of a domain-specific corpus.