Sentiment Analysis in a Resource Scarce Language:Hindi

A common human behavior is to take other’s opinion before taking any decision. With the tremendous availability of documents which express opinions on different issues, the challenge arises to analyze it and produce useful knowledge from it. Many works in the area of Sentiment Analysis is available for English language. From last few years, opinion-rich resources are booming in other languages and hence there is a need to perform Sentiment Analysis in those languages. In this paper, a Sentiment Analysis in Hindi Language (SAHL) is proposed for reviews in movie domain. It performs 1) preprocessing like stopword removal and stemming on the input data, 2) subjectivity analysis on the preprocessed data, to remove objective sentences that are not contributing to opinion of the input data, 3) document level opinion mining for classification of the documents as positive and negative using two different methods: Machine learning technique and Lexicon based classification technique. We have used Naive Bayes Classifier, Support Vector Machine and Maximum Entropy techniques for Machine learning. In Lexicon based classification, adjectives are considered as opinion words and according to the polarity of the adjectives, the documents are classified, 4) negation handling with window size consideration for improving the accuracy of classification. The effectiveness of the proposed approach is confirmed by extensive simulations performed on a large movie dataset.