Lexicon Based Document Level Sentiment Analysis on the Multilingual Dataset

Abstract

Sentiment Analysis is a potential and high-speed emergent research topic as it deals with the mining of social users’ opinions which generates incredibly in fraction of seconds. The hasty growth of online medium such as blogs and various social networking sites are nowadays used by many users to express their perspectives on various topics. Most of us are more comfortable in expressing our views and opinions in native language. Commonly using positive and negative English words might also be used along with the native language. Here the authors propose a method to perform polarity calculation on the multilingual dataset considered for sentiment. The multilingual dataset under consideration consists of a mix of reviews in English and Malayalam. Various language detectors are considered for experimentation and the langid algorithm gives more accurate results. The English and Malayalam lexicons need to be treated differently. Since English is a Universal language, the positive and negative lexicons and the list of stopwords are readily available. But the positive and negative Malayalam lexicons need to be created in house. The proposed methodology treats both types of lexicons differently for sentiment analysis.