Predicting the Polarity Strength of Adjectives Using WordNet

Authors

  • Gbolahan Williams University of Warwick
  • Sarabjot Anand University of Warwick

Keywords:

sentiment polarity strength, machine learning, evaluation

Abstract

A key element of any sentiment analysis system is the ability to assign a polarity strength value to words appearing within the documents. In this paper we present a novel approach to polarity strength assignment. The approach is knowledge based in that it uses WordNet to build an adjective graph which is used to measure semantic distance between words of known polarity (reference or seed words) and the target word, which is then used to assign a polarity to the target word. We extend previous work in this area by using a small training data set to learn an optimal predictor of polarity strength and to dampen polarity assigned to non-polar adjectives. We also extend the coverage of previous approaches by exploring additional lexical relations not studied previously. The method has been evaluated on a validation set and shows excellent potential in reducing the assignment of spurious polarity and accurately predicting polarity values for polar adjectives.

Downloads

Published

2009-03-20

How to Cite

Williams, G., & Anand, S. (2009). Predicting the Polarity Strength of Adjectives Using WordNet. Proceedings of the International AAAI Conference on Web and Social Media, 3(1), 346-349. Retrieved from https://ojs.aaai.org/index.php/ICWSM/article/view/13995