Sentiment Classification using Swarm Intelligence in Big Data

Abstract

This thesis also develops a PSO based new feature selection approach by newlinemodifying the representation scheme of particles which is able to generate the desired newlinenumber of high-quality features from a large set of features. The proposed approach has newlinebeen tested on the large dataset of text reviews and results show that proposed PSO newlinebased feature selection approach is yielding higher classification accuracies in all the newlineconsidered classifiers and efficiently deal with high-dimensional feature space. To newlineimprove the performance of sentiment classification, this thesis explores various newlineimportant features of unstructured data and also proposes two new features which are newlinehelpful in finding sentiment of contrastive sentences more accurately. The experiment newlineconducted on TripAdvisor dataset reveals the significant improvement in classification newlineaccuracy after incorporating our proposed features.