Sentiment Analysis on Beauty Product Reviews using LSTM Method

Abstract

A review is an opinion that contains value on the job or event being reviewed. Many sites provide reviews of products or goods in the modern era to users, such as the femaledaily.com site, which provides a platform for users to review products purchased. The sentiments contained in these reviews are valuable information for business owners. Thanks to product reviews, business owners get insights and data related to the products they sell to improve their products' quality. However, getting opinion information from an unstructured review text is quite difficult. This study aims to classify these reviews as "positive" or "negative". The model proposed for classification is LSTM. Long Short-Term Memory (LSTM) was used in the previously trained model to classify this review. The model designed for the model focuses on preprocessing reviews as follows: data cleansing, case folding, normalization, tokenization, stopword, and stemming. Once classified, this review is visualized as a graph. The best-case scenario results with an accuracy of 95,10% on the sentiment towards the price aspect.