Supervised Opinion Mining of Social Network Data Using a Bag-of-Words Approach on the Cloud

Abstract

Social networking and micro-blogging sites are stores of opinion-bearing content created by human users. Sentiment analysis extracts and measures the sentiment or “attitude” of documents as well as the topics within documents. The attitude may be the person’s judgment (e.g., positive vs. negative) or emotional tone (e.g., objective vs. subjective). In this paper we define a supervised learning solution for sentiment analysis on twitter messages which gives profound accuracy using Naive Bayes Classification. It classifies the tweets either positive or negative. Our dataset consists of tweets containing move reviews retrieved from twitter using certain keywords on a cloud platform. The experiment and analysis has 3 major steps. In the first step, the algorithm used is Naive Bayes which performed Boolean classification on bag of words, resulting in 71 % of accuracy. In the second step, Naive Bayes algorithm is applied on bag of words without stop words which gave 72 % accuracy. In the third step, using the concept of Information gain, high value features are selected using Chi Square, which gave maximum accuracy. The evaluating metrics used in this work are Accuracy, Precision, Recall and F-Measure for both the sentiment classes. The results show that Naive Bayes algorithm with the application of feature selection using the minimum Chi Square value of 3 gave an Accuracy of 89 %, Positive Precision 83.07 %, Positive Recall 97.2 %, Positive F-Measure 89.9 %, Negative Precision 96.6 %, Negative Recall 81.2 % and Negative F-Measure 88.2 %. The significant increase in Positive Recall specifies that the classifier classifies positive words with more probability when compared to the negative words. This paper also describes the preprocessing steps needed in order to achieve high accuracy.