Dynamic SentiPhraseNet to Support Sentiment Analysis in Telugu

In the scarcity of Telugu, annotated dataset makes sentiment analysis task challenging for researchers in the recent times. A rapid growth was seen in development of annotated datasets in Telugu for sentiment analysis. SentiWordNet (SWNet) is one of them where they mapped a sentiment score to every word. However, we found that there is a limitation in unigram words of SWNet. As several unigram words are ambiguous and it is unable to contribute in sentiment analysis task without the context of word in the given sentence. To resolve such limitations of SWNet, this article proposed SentiPhraseNet (SPNet). SPNet is a collection of sentiment phrases (such as bigram and trigram to get the context of word in the given sentence). Additionally, the proposed approach is compared with existing other approaches as well. The SPNet is also extended for dynamic support which learns the unknown phrases automatically while testing the sentiment of Telugu sentences with SPNet. The proposed approach is outperformed the other existing approaches and attains an accuracy of 90.9% after testing five sets of Telugu sentences in five trials.