Supervised approaches have proved their significance in sentiment analysis task, but they are limited to the languages, which have sufficient amount of annotated corpus. Hindi is a language, which is spoken by 4.70% of the world population, but it lacks a sufficient amount of annotated corpus for natural language processing tasks such as Sentiment Analysis (SA). With the increase in demand and availability of Hindi review websites, an accurate sentiment analyzer for Hindi has become a need. In this paper, we present a bootstrap approach to extract senti words from Hindi-WordNet. The approach is designed such that it minimizes the extraction of the words with the wrong polarity orientation, which is a crucial task, because a word can have positive and negative senses at the same time. The resultant set of 8061 polar words, we call it Hindi senti lexicon, is used for sentiment analysis in Hindi. We get an average accuracy of 87% for sentiment analysis in the movie and product domain.
展开▼
