A Hybrid Learning approach for Sentiment Classification in Telugu Language

Sentiment or Opinion Analysis is a subset of Natural Language Processing which is employed in wide range of business verticals to disentangle, analyze, distinguish and comprehend the general opinion of user reviews, comments, feedback, news, and so on. Humans generate close to 2.5 quintillion bytes of data each day on internet and business leaders are tasked to derive hidden patterns and meaningful insights from this data to understand human behavior and take shrewd decisions. Since last decade, enormous amounts of text data on internet is generated for Indian languages. Researchers have shown inquisitiveness in deriving and analyzing this data to extract relevant information. To the best of our knowledge, there has been no requisite amount of research in classifying sentiment in Telugu language because of inadequate language resources and also being a regional language. This research paper illustrates a methodical approach which leverages lexicon based approach and machine learning in the field of sentiment analysis to classify the opinions in Telugu language. Firstly, by employing Lexicon based approach - Telugu SentiWordNet we identified the subjective sentences from the Telugu corpus. Secondly, by utilizing machine learning algorithms - SVM, Naïve Bayes and Random Forest we categorized the sentiment in the corpus. Our proposed methodology achieved highest accuracy of 85%.