Natural Language Processing on Marketplace Product Review Sentiment Analysis

Abstract

Marketplace is a platform that bridges buyer and seller to transact. In January 2020, Central Bank reported that total transactions in 14 Indonesian marketplaces reached 23.27 trillion rupiahs. At the end of the transaction process, the buyer has the opportunity to give a review of the product that has been purchased. This research aims to sentiment analysis on marketplace product reviews. This research applies Natural Language Processing as a pre-processing of text in the sentiment analysis of marketplace product reviews using a machine learning approach with the Naive Bayes and K-NN algorithms. The test scenario obtained an average Naive Bayes accuracy of 52.4% on the Unigram dataset and an average K-NN accuracy of 79.4% on the Bigram dataset. The utilization of NLP, particularly the word normalizer, can increase the accuracy of 10% for Naive Bayes and 4% for K-NN. Sentiment analysis on marketplace product reviews is useful as an opportunity for product improvement for both seller and competitor, in the other side it is also used as a reference for other users before buying a product.