Analysis of Customer Opinion Using Machine Learning and NLP Techniques

Abstract

The amount of information readily accessible has increased dramatically due to the popularization and evolution of the Internet. It has become the main platform to express opinions for a large section of the population. These opinions can be shared in the form of reviews, in online discussion forums, or through social media platforms like Twitter, Facebook etc. Having easy and free access to all these opinions opens the door to opinion mining and sentiment analysis. Sentiment Analysis primarily deals with the identification of the reviewers’ intent towards a product, service or topic, that is either positive, negative or neutral, automatically through computational techniques. These techniques are gaining popularity drastically due to their effect on any given business. Hence, in this paper, we have explored various algorithms towards both classification, as well as data preprocessing and feature extraction, and compared their results in terms of FAR, FRR and accuracy measure. The simulation results conclude that SVM classifier, when applied on a dataset of features extracted by the N-gram model, results in the highest accuracy of 95.23% for the given dataset.

Keywords: Bag of Words, Decision Tree, K-Nearest Neighbor, Naïve Bayes, N-gram, Random Forest, ROC, Supervised learning, Support Vector Machine, Term Frequency-Inverse Document Frequency