A novel approach for Sentimental Analysis and Opinion Mining based on SentiWordNet using web data

Opinion mining is an art of tracking the mood of the public about a particular product or topic from a huge set of opinions or reviews publically available in web. In this work, a novel approach is proposed based on SentiWordNet, which generates count of score words into seven categories such as strong-positive, positive, weak-positive, neutral, weak-negative, negative and strong-negative words for the opinion mining task and evaluated using machine learning algorithms like Naïve Bayes, SVM and Multilayer Perception (MLP). The web data is collected using web crawler applied with different pre-processing techniques which include removal of stop-words from online reviews, then stemming is performed using Porter Stemmer algorithm, and then reviews are tagged using Stanford POS tagger. The proposed approach is experimented on movie and product web domains and obtained higher success rate in terms of accuracy measured by various tools like Kappa statistics with an accuracy of 77.7% and has lower error rates. Weighted average of different accuracy measures like Precision, Recall, TP Rate, F-Measure rate depicts higher efficiency rate and lower FP Rate for Naïve Bayes and MLP models. The experimental results of Ten-Fold cross validation on the training data shows that Naïve Bayes & MLP outperforms the SVM model. Thus, the former are used for the Sentimental Analysis of web data. The results demonstrate that the proposed novel approach has higher efficacy and it can be successfully used in Opinion Mining for the task of decision making by any web user.