Polarity detection of Kannada documents

Document polarity detection is a part of sentiment analysis where a document is classified as a positive polarity document or a negative polarity document. The applications of polarity detection are content filtering and opinion mining. Content filtering of negative polarity documents is an important application to protect children from negativity and can be used in security filters of organizations. In this paper, dictionary based method using polarity lexicon and machine learning algorithms are applied for polarity detection of Kannada language documents. In dictionary method, a manually created polarity lexicon of 5043 Kannada words is used and compared with machine learning algorithms like Naïve Bayes and Maximum Entropy. It is observed that performance of Naïve Bayes and Maximum Entropy is better than dictionary based method with accuracy of 0.90, 0.93 and 0.78 respectively.