SISTEM REKOMENDASI DRAMA KOREA BERDASARKAN PREDIKSI RATING DAN KEMIRIPAN SINOPSIS MENGGUNAKAN SINGULAR VALUE DECOMPOSITION DAN TERM FREQUENCY-INVERSE DOCUMENT FREQUENCY

KOREAN DRAMA RECOMMENDATION SYSTEM BASED ON RATING PREDICTION AND SYNOPSIS SIMILARITY USING SINGULAR VALUE DECOMPOSITION AND TERM FREQUENCY-INVERSE DOCUMENT FREQUENCY

Pengaruh budaya Korea semakin menyebar luas di Indonesia, menyebabkan banyak drama korea (drakor) baru dirilis. Hal ini membuat pecinta drakor bingung dalam memilih drakor yang ingin mereka tonton. Sehingga dibutuhkan sistem rekomendasi drakor berdasarkan prediksi rating dan kemiripan sinopsis. Pecinta drakor mayoritas memilih drakor yang populer, dan rating menjadi salah satu pengaruh sebuah drakor dikatakan populer. Atribut sinopsis untuk memudahkan pecinta drakor mengetahui alur cerita drakor. Singular Value Decomposition digunakan untuk membuat model prediksi rating. Term Frequency-Inverse Document Frequency digunakan untuk memberikan bobot kata pada sinopsis dan deskripsi singkat, kemudian dihitung kemiripannya dengan Cosine Similarity. Model prediksi rating dilatih dan diuji menggunakan lima skenario pembagian data (50:50, 60:40, 70:30, 80:20, 90:10), menghasilkan RMSE terendah pada skenario 90:10 sebesar 1.493423. Kemiripan sinopsis dicari dengan lima skenario deskripsi singkat yang berbeda, dan dihasilkan nilai kemiripan tertinggi yaitu 53%. Sistem rekomendasi yang dibuat bisa memberikan rekomendasi drakor yang sesuai dan bisa diurutkan berdasarkan kemiripan sinopsis dan prediksi rating.

The influence of Korean culture is increasingly widespread in Indonesia, causing many new Korean dramas (drakor) to be released. This makes drakor lovers confused in choosing the drakor they want to watch. So a drakor recommendation system is needed based on rating predictions and synopsis similarity. The majority of drakor lovers choose popular drakor, and ratings are one of the influences of a drakor being said to be popular. Synopsis attributes make it easier for drakor lovers to know the storyline of the drakor. Singular Value Decomposition is used to create a rating prediction model. Term Frequency-Inverse Document Frequency is used to give word weights to the synopsis and short descriptions, then the similarity is calculated using Cosine Similarity. The rating prediction model was trained and tested using five data sharing scenarios (50:50, 60:40, 70:30, 80:20, 90:10), producing the lowest RMSE in the 90:10 scenario of 1.493423. Synopsis similarity was sought with five different short description scenarios, and the highest similarity value was 53%. The recommendation system created can provide appropriate Korean drama recommendations and can be sorted based on the similarity of the synopsis and rating predictions.