Sentimental Analysis of Book Reviews using Unsupervised Semantic Orientation and Supervised Machine Learning Approaches

Abstract

Sentimental analysis aims at identifying the opinions of various users. This paper presents my research work on the application of sentimental analysis on book reviews. I have applied both unsupervised (Semantic Orientation - Pointwise Mutual Information - Information Retrieval) and supervised (Support Vector Machine and Naïve Bayes) machine learning approaches on two openly available book review datasets from GoodReads and Amazon. The comparative analysis of the approaches on the datasets indicates that unsupervised approach performs better on GoodReads dataset with an accuracy of 73.23% whereas supervised approach gives better results on Amazon dataset with Naïve Bayes giving the maximum accuracy which ranges from 73.72% to 74.73% in the case of 5-folds and 10-folds respectively.