Mining Chinese polarity shifters

Abstract

In sentiment analysis, polarity shifting means to shift the polarity of a sentiment clue. Compared with other natural language processing (NLP) tasks, to extract polarity shifters (polarity shifting patterns) in corpora is a challenging one, since the polarity shifters sometimes are very subtle, which often invalidates fully automatic approaches. In this paper, aiming to extract polarity shifters that invert or attenuate polarity, we use a semi-automatic approach based on pattern mining. The approach can greatly reduce the human annotating cost and cover as many polarity shifters as possible. We tested this approach on domain corpora, and encouraging experimental results are reported.