Opinion mining approaches on Amazon product reviews: A comparative study

摘要

The process of extracting of people's opinion, experience and emotions from reviews, blogs and other sources is known as opinion mining. This paper compares our lexicon dictionary based approach with n-grams with three famous Machine Leaning (ML) algorithms, which are random forest learner with word vector, decision tree learner with document vector, and random forest with n-gram. To predict positive and negative sentiments, Amazon's Product Review dataset has been used. Accuracy of each of these algorithms is calculated by using ROC curve in order to compare which algorithm performs best on a given Amazon dataset. Experimental result shows that lexicon based approach outperforms other machine learning techniques.