Recommender System Based on User Evaluations and Cosmetic Ingredients

This study considers the compatibility between users and basic skin care products based primarily on the products’ ingredients. We have developed a product recommender system that is expected to recommend products that provide the desired cosmetic effect for different user groups depending on age and skin type. From the cosmetics review site, based on an analysis of user evaluations, we extracted the names of ingredients that are thought to have the best effect and developed a method to recommend products that include these ingredients as their main ingredients. We propose the ingredient frequency-inverse product frequency (IF-IPF) method to derive ingredients characterizing strong-effect product group. We have defined the scale “the recommended product satisfaction level” to evaluate the effectiveness of our recommendation service. As a result, our system can recommend products with a high degree of serendipity and hidden attraction, among others.