Comparative Study of Classification Algorithms in Sentiment Analysis

Abstract

In daily life customer reviews or opinions play a very key role. Whenever anyone has to take a decision, they are likely to consider opinions of others. Analyzing the people’s feeling is significant for many applications such as companies, ventures, enterprises trying to find out the review of their goods in the marketplace, predicting socio-economic situations like stock exchange and predicting political elections. Since there is a good quantity of collective data present on the internet automatically therefore, it’s very important to employ techniques that automatically classify them. Sentiment Classification is also called as Opinion Mining. This project addresses sentiment analysis in Hotel reviews that classifies the hotel reviews according to the sentiment expressed by the public into: positive or negative. To classify the hotel reviews i.e. the given sentiment we used natural language tool kit (NLTK) to pre-process the natural language tasks. While pre-processing the text review first we remove stop words, punctuations and regular expressions .We also used n-grams to improve the sentimental analysis. After pre-processing the sentiment we choose a classifier to classify the sentiment. So, in order to selecting the best classifier in this paper we use different classifiers of different models to classify the sentiment and we also give the brief comparison of classifiers in terms of Accuracy, Precision, Recall and F-Measure. With these evaluation metrics we can show the best classifier. Multinomial NaiveBayes is the most accurate classifier to our data set and it gives quite efficient results than other classifiers.