<![CDATA[The rapid growth of the skincare industry has triggered information overload, complicating consumer decision-making particularly among Generation Z users on e-commerce platforms. Conventional Collaborative Filtering approaches are limited by popularity bias and the cold-start problem, and are unable to account for ingredient-level compatibility with individual skin conditions. Addressing this gap, this study proposes a novel Content-Based Filtering recommendation system that integrates TF-IDF and Cosine Similarity with a Knowledge-Based Normalization layer. This original framework maps informal consumer terminology into standardized dermatological categories, effectively reducing semantic inconsistency in unstructured product descriptions. Data were obtained from the Kaggle public repository (third-party extracted dataset) and underwent a validation process, yielding a final dataset of 91 skincare products. The system was evaluated using Precision@K across five skin-condition scenarios. Results yield an average Precision@5 of 0.80 (80%), with a peak cosine similarity score of 0.3606. The low absolute cosine value is attributable to TF-IDF vector sparsity in short-text descriptions, a characteristic acknowledged in prior literature. Implementation as a web application confirms the system's practical utility in guiding users toward biologically appropriate skincare choices, independent of market-trend bias.]]>
Copyrights © 2026
