A Sentimental Education: Sentiment Analysis Using Subjectivity Summarization Based on Minimum Cuts

This paper addresses sentiment analysis, the task of identifying the viewpoint expressed in a text span, illustrated by classifying a movie review as thumbs up or thumbs down. Rather than processing an entire document, the authors propose applying standard text-categorization techniques only to the subjective portions of the text. They identify and extract those subjective portions by formulating the problem as finding minimum cuts in graphs.

Because minimum cuts in graphs can be computed efficiently, the formulation greatly facilitates incorporating cross-sentence contextual constraints when determining which text is subjective. By concentrating polarity classification on the subjective content alone, this novel machine-learning method offers a principled way to combine subjectivity extraction with sentiment categorization.