Sentence Level Sentiment Analysis in Malayalam

Abstract

In this paper, we propose a sentence level sentiment analysis for Malayalam language. This is to analyze the sentiment of Malayalam sentence. In the existing system sentiment of a sentence in English is done by reversing the sentence or split the sentence as words sequence. When the sentence is split into words sequence, polarity inconsistency occurs. We address this issue by segmenting the sentence as unigrams and bigrams then predicts sentence level polarity based on the segmentation results. Here we introduce a training phase and prediction phase. We extend the polarity framework (positive-negative) into three classes (positive-negative-neutral) by considering neutral reviews into consideration. Finally, we develop a corpus based method to get the polarity of words. We conduct experiments for sentiment analysis on different review datasets. Experimental results show that: (1) our method will perform well on different datasets of film reviews; (2) when we consider the neutral polarity then gets more accurate result than the binary polarity (positive-negative).