Sentiment-based text segmentation

Abstract

In this paper, we present a text segmentation system based on the sentiments expressed in the text. The system takes as input plain text (product review for instance) and uses two different resources for tagging the sentiment words: a sentiment words dictionary and SentiWordNet. Once the sentiment words are identified, the initial text is annotated with segmentation markers when polarity shifts. The system also outputs the counts of positive and negative sentiment words found in text and optionally annotates them with their valence.