Penentuan Nilai Rating Produk Kecantikan Berdasarkan Analisis Sentimen Berbasis Aspek Menggunakan Latent Dirichlet Allocation dan Adaptive Boosting

Abstract

The significant development of beauty products in the present day has given rise to online platforms dedicated to the beauty industry. The advancement of technology and easier access to mobile devices have turned beauty platforms into valuable sources of information for users seeking knowledge about beauty products. Through these platforms, users can also provide feedback on their experiences with a particular product. One of the commonly used features by beauty product consumers is the review feature. This research aims to analyze sentiment in reviews related to beauty products and design a product rating system based on specific aspects, making it easier for users to make informed decisions before purchasing a product. The data for this study consisted of 4000 product records obtained through web scrapping from the Female Daily website. The research also underwent preprocessing steps, including data cleaning, case folding, stemming, punctual removal, stopwords removal, normalization, and tokenization. The feature extraction used for word weighting in this study was TF-IDF. Topic determination using LDA, resulted in four topics: acne-fighting capability, post-usage effects, purchase evaluations, and product texture. Sentiment classification was carried out using the AdaBoost Classifier. The overall accuracy obtained from the Confusion Matrix evaluation of the model was 78% for all aspects, with the highest accuracy of 85% for the product texture aspect, 80% for post-usage effects, 78% for purchase evaluations, and 70% for acne-fighting capability.