Word2Vec-Based Facial Serum Recommender System

The beauty industry, especially skincare, has grown rapidly with a large selection of products with various compositions for various skin types. Through the variety of ingredients in skincare products, it becomes a hurdle for consumers to choose the right product that suits their needs and skin conditions. Many previous studies have recommended skin care products based on the user's skin condition, but have not fully focused on each composition contained in the product. This research proposes a skin care product recommender system, especially facial serum based on content-based filtering using average Word2Vec and cosine similarity to provide more personalized and relevant recommendations. The system works by analyzing the composition of active ingredients in products and calculating the similarity between products based on the semantic relationships found by the Word2Vec model. The use of Word2Vec to understand the semantic relationships between active ingredient compositions in facial serums represents a novel approach in this domain. Using a dataset of skincare products and user reviews, this study shows that the proposed approach is able to provide highly relevant product recommendations. It achieves a Normalized Discounted Cumulative Gain (NDCG) score of 0.979 and a Hit Rate of 0.998. These results indicate that the Word2Vec and cosine similarity methods are effective in content-based recommender systems. It can capture semantic relationships between ingredients and calculate compositional similarity with products that have been favored. This approach has the potential to assist consumers in making more informed skincare decisions.