A LEXICON BASED APPROACH FOR SENTIMENT ANALYSIS USING MULTI-CLASS LABELS ON TWITTER DATA: CASE

Abstract

In this paper, we are proposing a pioneering hierarchical lexicon model to construct a contextadaptive sentiment echelon (CASE) for multi-level classification purpose. It is universally accepted
that sentiment lexicon as a handy foregoing understanding is crucial in opinion mining or sentiment
analysis. Nevertheless many of the existing sentiment lexicons were constructed without taking the
notice of the variation of the sentiment polarities of words in dissimilar contexts. For instance, the
word “positive” can be considered as positive when appeared with “feeling” whereas it is negative
when appeared with “HIV”. For that reason, we must identify the occurrence of a word according to
the context where they occur. The key advantage of CASE is in finding the sentiment polarity of each
word in various contexts can be captured adequately. In our research work, we proposed a novel
approach by building the idea on twitter data determining the level of intensity of the tweet by using
fine-grained sentiment analysis. Our model is very useful for finding the polarity strength in terms of
weakly positive, positive, strongly positive, neutral, weakly negative, negative, and strongly negative.
We are using multi-class labels in our method which improves the efficiency of all the parameters