Abstract

Sentiment Analysis is initial step of Emotion Recognition which is one of the driving force in the area of research in Artificial Intelligence. Availability of large Web text and improved tools in Natural Language Processing are attracting researchers to this field. However there is tremendous research scope in Marathi Language spoken mainly in Maharashtra, India which is very ancient and rich in morphology. This paper presents an attempt to develop a machine learning model for sentiment classification of Marathi Corpus . N Gram feature of word and polarity of each word is considered to calculate document polarity. Both Negative score and Positive score are fed to classifiers and performance is compared. The conclusion doesn’t stop on any one of the classifier but SVM and Logistic Regression gives overall good results in all circumstances. Stochastic Gradient Descent gives highest accuracy for both bigram and trigram features when 70:30 split is used.