Electronic Theses and Dissertation

Skincare merupakan produk perawatan kulit wajah yang saat ini sangat populer di indonesia. produk-produk skincare terbaru terus bermunculan dalam berbagai kategori dan merek yang beragam. namun, keberagaman ini seringkali membuat masyarakat kesulitan memilih produk skincare yang sesuai dengan preferensi mereka. dalam penelitian ini, dikembangkan sebuah sistem rekomendasi produk skincare menggunakan kerangka kerja deep reinforcement learning (drl) dengan algoritma deep determenistic policy gradient(ddpg). data yang digunakan merupakan data rating pengguna terhadap produk-produk skincare yang didapatkan melalui proses crawling dan scraping pada website review produk kecantikan yaitu femaledaily. evaluasi terhadap sistem ini menunjukkan nilai precision@10 = 0.78 dan ndcg@10 sebesar 0.77 pada pelatihan 100 episode, nilai precision@10 = 0.88 dan ndcg@10 sebesar 0.87 pada pelatihan 500 episode serta nilai precision@10 = 0.90 dan ndcg@10 sebesar 0.89 pada pelatihan 1000 episode. hasil ini menunjukkan bahwa kerangka kerja drl dapat diterapkan ke dalam sistem rekomendasi skincare berbasis rating. kata kunci : sistem rekomendasi, skincare, deep reinforcement learning

Skincare products, which are used for facial skin care are currently very popular in Indonesia. New skincare products continue to emerge in various categories and brands. However, this diversity often makes it difficult for consumers to choose skincare products that match their preferences. This research develops a skincare product recommendation system using the Deep Reinforcement Learning (DRL) framework with the Deep Deterministic Policy Gradient (DDPG) algorithm. The data used consists of user ratings for skincare products collected through crawling and scraping processes from the beauty product review website FemaleDaily. The evaluation of the system shows a precision@10 score of 0.78 and an NDCG@10 score of 0.77 after training for 100 episodes, a precision@10 score of 0.88 and an NDCG@10 score of 0.87 after 500 episodes, and a precision@10 score of 0.90 and an NDCG@10 score of 0.89 after 1000 episodes. These results show that the DRL framework can be applied to a rating-based skincare recommendation system. Keyword : Recommendation Systems, Skincare, Deep Reinforcement Learning