Sentiment-property extraction using Korean syntactic features

Abstract

Since Korean sentence structure generally has a predicate expressing a sentiment at the end, it is necessary to find out the correct property the predicate explains in a sentence. This study presents a sentiment-property extraction model that can reflect the features of the Korean syntax to find out a correct sentiment-property pair. The model uses a Korean parser to find out the property word dependent on a possible sentiment word in the parsed sentence and extracts the two words to make a sentiment-property pair when they are likely to form a pair. The test set yielded a precision ratio of 93% and recall ratio of 75%. © 2012 Springer Science+Business Media Dordrecht.