The abundance of skincare product options and online information often hinders consumers in making informed decisions. Although previous studies have introduced recommender systems to support decision-making, personalization issues remain, notably the inadequate use of user reviews, which offer essential information about individual feelings and preferences. Therefore, we introduce a hybrid recommender system for skincare products that combines content-based filtering (CBF), item-based collaborative filtering (IBCF), and sentiment analysis using the VADER algorithm. This method extracts sentiment scores from user reviews and converts them into discrete ratings, replacing the original ratings with these new values. We use the grid search method to find the optimized weights between CBF and IBCF contributions. The evaluation indicates that this hybrid model of sentiment-based ratings outperforms the conventional method, achieving NDCG@10, MRR@10, and Hit Rate scores of 0.977, 0.9682, and 0.9998, respectively. The results suggest that incorporating sentiment analysis into hybrid recommender systems can enhance the accuracy and relevance of recommendations, while also providing users with a more personalized experience.
Keywords—skincare, hybrid recommender system, content-based filtering, item-based collaborative filtering, sentiment analysis