A robust method to extract polar words from unstructured text

Abstract

In the last decade, sentiment analysis becomes popular by helping to quantify user’s opinion. A very important step to implement sentiment analysis is to extract polar words from the target text. It is easy to achieve if the text is clean and structured like news contents. By contrast, if the text is dirty and unstructured, particularly for the social data such as tweets and product reviews, polar words extraction becomes very hard. This problem may be much more serious for some Asian languages like Korean. In order to extract high-quality polar words from the unstructured text in the Korean language, this paper presents a robust method by detecting and expanding the variations of polar word roots. The experimental results show that the proposed method can extract more polar words than the basic extraction method and meanwhile reserve a very high precision. © Research India Publications.