Polarity Classification of Subjective Words Using Common-Sense Knowledge-Base

Abstract

Semantic orientation of a word indicates whether the word denotes a positive or a negative evaluation. We present an approach to compute semantic orientation of words using machine-interpretable common-sense knowledge. We employ ConceptNet (a large semantic network of commonsense knowledge) for determining the polarity or semantic orientation of a sentiment expressing word. We apply heuristics on certain pre-defined predicates expressing semantic relationship between two concepts for classifying words that have a positive or negative polarity and finding words that have similar polarity. The advantages of the proposed approach are that it does not require any pre-annotated training dataset or manually created seed list. The proposed solution relies on a lexical resource which is created by volunteers on the Internet and not by trained or specialized knowledge engineers. We test our approach on publicly available pre-classified sentiment lexicon and present the results of our experiments and also examine the tradeoffs and limitations of the proposed solution. We conclude that it is possible to determine polarity of words with high accuracy by exploiting a machine-understandable layman’s knowledge and basic facts that ordinary people know about the world.