Sentiment classification of online consumer reviews using word vector representations
Barkha Bansal, Sangeet Srivastava · Procedia Computer Science · 2018
With rapid increase in e-commerce, it is important to develop quick and efficient methods for consumer review mining. One way for mining consumer reviews is sentiment classification with machine learning approach. In this paper we use word2vec model and convert reviews into vector representations for classification. Dataset consists of more than 400,000 consumer reviews in the mobile phone category from Amazon. Our research consists of two parts: First we demonstrate the ability of word2vec to find similar semantic features in the domain of study and then we classify the consumer reviews using CBOW (continuous bag of words) and skip-gram models with different machine learning algorithms: SVM, Naïve Bayes, Logistic Regression and Random Forest using 10-folds cross validation. The experimental results show that Random Forest using CBOW achieves most superior accuracy.