부산대학교 도서관

Huge abundance e-commerce websites and online reviews have become crucial these days. These reviews help customers in making decisions but one must go through huge pile of reviews in many sites. We have summarized the reviews into STARS on a scale of 1-5 which are easy to perceive. So, for a given customer review, we predict the star rating of the review. Proposed approach in this research work first pre-process review data and then train different classifiers like Multinomial Naïve Bayes, Bigram Multinomial Naïve Bayes, Trigram Multinomial Naïve Bayes, Bigram-Trigram Multinomial Naïve Bayes, Random Forest. Finally, trained models predict star rating of a review. Comparing the performance of these classifiers, it is observed that Random Forest is better than the other classifiers in terms of accuracy. However, Bigram-Trigram Multinomial Naïve Bayes is on par with the results of classifier like Random Forest as well as has far less computational time.