Extended correlated principal component analysis with SVM-PUK in opinion mining

Abstract

With the rapid growth of microblogs and online sites, an inordinate number of product reviews are available on the Internet. They not only help in analyzing, but also assist in making informed decisions about product quality. In the proposed work, an extended correlated principal component analysis (ECPCA) is used for dimensionality reduction. A comparative analysis is conducted on movie reviews (DB-1) and Twitter datasets (DB-2 and DB-3) in opinion mining extraction. The performance of naive Bayes, CHIRP, and support vector machine (SVM) with kernel methods such as radial basis function (RBF), polynomial, and Pearson (PUK) are compared and analyzed on the three datasets. The experimental results using ECPCA for selecting relevant features and SVM-PUK as a classifier exhibit better performance on movie reviews and Twitter datasets. The performance of the proposed approach is 99.69 %, 99.4 %, and 99.54 % on the DB-1, DB-2, and DB-3 datasets, respectively, and comparatively outperforms the existing methods.