Sentiment Classification for Mobile Reviews using KNIME

Abstract

Sentiment analysis is most important part of natural language processing. It’s a big issue to review on a product and technology due to countless new features comes every day in the technology. The sentiment analysis same as the opinion mining, which gives different reviews on text, product, article and news that, can be positive and negative. People use social networking sites like Amazon, flip kart and other social websites. The aim of this paper is web scraping mobile reviews using Helium and classified with 1 gram and combination 1 & 2 gram and analysis them by KNIME. KNIME is open and free tool. It is used for extraction, transforming the text or sentence or loading. In this paper, both supervised as well as unsupervised machine learning algorithms are used: Support Vector machine, Decision Tree, Naive Bayes and Random Forest and compares their overall Accuracy, Cohen’s kappa, precisions, recall values and F-measure. These algorithms are used to classify the reviews as positive and negative .It was seen that in case of combination of 1 & 2 gram of mobile review gave far better results than 1 gram. The experimental results show that the combination of 1 & 2 gram gave better result with the Decision Tree and SVM algorithms giving above 88% accuracies and outperforming than the Random Forest and Naive Bayes algorithms. In this paper, after the comparison of four algorithms, SVM and decision tree gave higher accuracy. So, combine both the algorithms as fusion approach to classify the mobile reviews for highest accuracy than above four algorithms. This approach achieved an accuracy of above 90% in combination of 1 & 2 grams. Thus, it can observe that combination of 1 & 2 gram give better result than 1 gram.