The polarity shift problem is amajor factor that affects classification performance ofmachinelearning-based sentiment analysis systems. In this paper, we propose a three-stage cascade model to address the polarity shift problem in the context of document-level sentiment classification. We first split each document into a set of subsentences and build a hybrid model that employs rules and statis...
