Building SentiPhraseNet for Sentiment Analysis in Telugu

Sentiment analysis of Indian languages is a challenging task due to rich morphology and little availability of the annotated datasets. For languages like Hindi, Telugu, Tamil, Bengali, Malayalam, etc., SentiWordNets (SWNets) were developed to tag the sentiment of each word. In this article, we observed that some unigram words of the existing Telugu SWNet are classified as ambiguous and are not sufficient to analyze the sentiment. In such situations, bigram and trigram phrases can be used to resolve the problem of ambiguity in sentiment prediction. Therefore, we proposed an algorithm to build the Telugu SentiPhraseNet (SPNet) for the sentiment analysis in Telugu. To build SPNet, we have collected the data from various sources namely, Telugu e-Newspapers, Twitter and NLTK Indian Telugu data which resolves the problems with existing SWNet. With the proposed SPNet, we have performed the sentiment analysis and it is compared with SWNet, various existing Machine Learning approaches namely, Support Vector Machine (SVM), Logistic Regression (LR), Naive Bayes (NB), Multilayer Perceptron Neural Network (MLPNN), Decision Tree (DT) and Random Forest (RF). The performance of the proposed system outperformed the other existing approaches and attains an accuracy of 85.6%.