Hybrid Deep Matrix Factorization dan Pendekatan Neural Content-Based Filtering untuk Sistem Rekomendasi Film Anime

Suandita, Komang Ryandhi (2025) Hybrid Deep Matrix Factorization dan Pendekatan Neural Content-Based Filtering untuk Sistem Rekomendasi Film Anime. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5002211109-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (2MB) | Request a copy

Abstract

Transformasi digital telah mendorong peningkatan konsumsi konten multimedia, khususnya anime, yakni animasi asal Jepang dengan desain visual khas dan alur cerita unik. Konsumsi anime di Indonesia didominasi oleh generasi muda berusia 15-34 tahun dengan rata-rata waktu tonton mencapai 7,5 jam per minggu tahun 2022. Dengan lebih dari 18.350 judul anime dirilis tahun 2021, pengguna menghadapi information overload yang memerlukan sistem rekomendasi untuk memfasilitasi discovery konten yang sesuai preferensi individual. Sistem rekomendasi konvensional seperti Content-Based Filtering dan Collaborative Filtering yang menggunakan teknik Cosine Similarity dan Matrix Factorization masih terbatas dalam menangani hubungan non-linear dan menghadapi cold start problem pada data dengan interaksi sparse, dimana sebagian besar pengguna hanya memberikan rating pada sedikit anime. Penelitian ini mengembangkan sistem rekomendasi film anime menggunakan Neural Content-Based Filtering yang memanfaatkan preferensi genre pengguna, Deep Matrix Factorization untuk menganalisis pola interaksi historis pengguna-anime dan Hybrid Filtering secara sequential untuk mengatasi masalah sparsity data. Dataset MyAnimeList 2023 yang digunakan mencakup 24.905 anime, 21.545 pengguna dan 2.432.519 rating dengan sparsity 99,5467% dan densitas 0,4533%. Evaluasi menunjukkan Neural Content-Based Filtering memberikan nilai terendah dengan RMSE 0.1749 dan MAE 0.1324, diikuti Hybrid Filtering dengan RMSE 0.2527 dan MAE 0.1829, serta Deep Matrix Factorization dengan RMSE 0.3845 dan MAE 0.3714. Hybrid Filtering berhasil mengurangi sparsity menjadi 98,7436% dan meningkatkan densitas menjadi 1,2564%, serta memperbaiki performa Deep Matrix Factorization. Ketiga model terbukti mampu memberikan rekomendasi Top 10 judul film anime kepada pengguna berdasarkan prediksi rating tertinggi.
====================================================================================================================================
Digital transformation has driven increased consumption of multimedia content, particularly anime, which is Japanese animation with distinctive visual design and unique storylines. Anime consumption in Indonesia is dominated by young people aged 15-34 years with an average viewing time of 7.5 hours per week in 2022. With over 18,350 anime titles released in 2021, users face information overload that requires recommendation systems to facilitate content discovery matching individual preferences. Conventional recommendation systems such as Content-Based Filtering and Collaborative Filtering using Cosine Similarity and Matrix Factorization techniques are still limited in handling non-linear relationships and face cold start problems on sparse interaction data, where most users only rate a few anime. This research develops an anime movie recommendation system using Neural Content-Based Filtering that utilizes user genre preferences, Deep Matrix Factorization to analyze historical user-anime interaction patterns and Hybrid Filtering sequentially to address data sparsity issues. The MyAnimeList 2023 dataset used includes 24.905 anime, 21.545 users and 2.432.519 ratings with 99,5467% sparsity and 0,4533% density. Evaluation shows Neural Content-Based Filtering achieves the lowest values with RMSE 0.1749 and MAE 0.1324, followed by Hybrid Filtering with RMSE 0.2527 and MAE 0.1829 and Deep Matrix Factorization with RMSE 0.3845 and MAE 0.3714. Hybrid Filtering successfully reduced sparsity to 98,7436% and increased density to 1,2564%, as well as improved Deep Matrix Factorization performance. All three models proved capable of providing Top 10 anime title recommendations to users based on highest rating predictions.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Sistem Rekomendasi, Content-Based Filtering, Collaborative Filtering, Neural Network, Deep Matrix Factorization, Pendekatan Hybrid
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QA Mathematics > QA76.9.D343 Data mining. Querying (Computer science) Q Science > QA Mathematics > QA76.9.I58 Recommender systems (Information filtering)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Komang Ryandhi Suandita
Date Deposited:	30 Jul 2025 06:52
Last Modified:	30 Jul 2025 06:52
URI:	http://repository.its.ac.id/id/eprint/123442

Actions (login required)

View Item