Pengembangan Sistem Rekomendasi Anime Berbasis Deep Q-Network (DQN)
Abstract
Sistem rekomendasi berperan penting dalam membantu pengguna menemukan konten sesuai preferensi di tengah banyaknya informasi. Penelitian ini membandingkan dua pendekatan, yaitu Collaborative Filtering (CF) berbasis Singular Value Decomposition (SVD) dan Reinforcement Learning (RL) berbasis Deep Q-Network (DQN). Fokus utama penelitian adalah menilai efektivitas keduanya dalam memberikan rekomendasi anime yang relevan, baik untuk pengguna eksisting maupun pengguna baru (cold-start). Dataset penelitian diambil dari Kaggle, melalui tahap preprocessing berupa pembersihan data, normalisasi fitur, dan encoding genre dengan one-hot. Model CF dilatih menggunakan parameter hasil tuning, sedangkan model RL dibangun dalam lingkungan simulasi dengan fungsi reward berbobot yang menggabungkan rating pengguna, skor global anime, dan kesamaan preferensi genre. Evaluasi dilakukan menggunakan skenario Top-N Recommendation (N = 1, 3, 5, 10, 15, 20) dengan metrik Precision@N, Recall@N, dan F1-Score@N. Item relevan untuk pengguna eksisting ditentukan berdasarkan reward persentil ke-80, sementara untuk pengguna baru ditetapkan pada anime dengan skor global ≥ 9.0. Hasil menunjukkan RL dengan DQN unggul pada masalah cold-start, sedangkan CF lebih baik untuk pengguna dengan riwayat interaksi. Perbandingan ini menyoroti kelebihan dan keterbatasan masing-masing pendekatan, sekaligus memberi panduan dalam memilih strategi rekomendasi sesuai konteks pengguna.
Kata kunci— deep q-network, reinforcement learning, colaborative filtering, sistem rekomendasi anime
References
Roziqiin, N. M., & Faisal, M. (2024). Sistem rekomendasi pemilihan anime menggunakan user-based collaborative filtering. JIPI (Jurnal Ilmiah Penelitian dan Pembelajaran Informatika), 9(1), 1–10. https://doi.org/10.29100/jipi.v9i1.4222.
Rahmawati, S., Nurjanah, D., & Rismala, R. (2018). Analisis dan implementasi pendekatan hybrid untuk sistem rekomendasi pekerjaan dengan metode knowledge-based dan collaborative filtering. Indonesian Journal on Computing (Indo-JC), 3(2), 11–20.
Putri, A. (2016). Cosplay sebagai identitas budaya populer. Repository Universitas Telkom.
Hatami, W. (2018). Popular culture of Japanese anime in the digital age and the impact on nationalism of young Indonesian citizens. Journal of Social Studies (JSS), 14(1), 1–10.
Y. Lei and W. Li, “When collaborative filtering meets reinforcement learning,” arXiv preprint arXiv:1902.00715, 2019.
F. Ricci, L. Rokach, and B. Shapira, Recommender Systems Handbook, 2nd ed. New York, NY, USA: Springer, 2015.
Y. Koren, R. Bell, and C. Volinsky, “Matrix factorization techniques for recommender systems. https://doi.org/10.1109/MC.2009.263.
Zou, L., Zhang, P., & Zhang, D. (2019). Reinforcement Learning to Optimize Long-term User Engagement in Recommender Systems. arXiv preprint arXiv:1902.05570.
Zivic, P., Vazquez, H., & Sánchez, J. (2021). Scaling Sequential Recommendation Models with Transformers. https://doi.org/10.48550/arXiv.2412.07585.
Gao, C., Zheng, Y., Li, N., Li, Y., Qin, Y., Piao, J., Quan, Y., Chang, J., Jin, D., He, X., & Li, Y. (2021). A Survey of Graph Neural Networks for Recommender Systems: Challenges, Methods, and Directions. https://doi.org/10.1145/3568022.
Chen, M., Xu, C., Gatto, V., Jain, D., Kumar, A., & Chi, E. (2019). Off-Policy Actor-Critic for Recommender Systems. https://doi.org/10.1145/3523227.3546758.
Li, C., Xia, L., Ren, X., Ye, Y., Xu, Y., & Huang, C. (2023). Graph Transformer for Recommendation. arXiv preprint arXiv: 2306.02330.
Liang, K., Zhang, G., Guo, J., & Li, W. (2022). An Actor-Critic Hierarchical Reinforcement Learning Model for Course Recommendation. Mathematics, 10(18), 3313.
Nurfauzi, A. I., & Wibowo, A. T. (n.d.). Sistem rekomendasi skincare menggunakan matrix factorization dengan metode non-negative matrix factorization. Fakultas Informatika, Universitas Telkom.
Nadhifah, A. R., & Wibowo, A. T. (n.d.). Sistem pemberi rekomendasi anime menggunakan pendekatan hybrid. Fakultas Informatika, Universitas Telkom.
Y. Lin, Y. Liu, F. Lin, L. Zou, P. Wu, W. Zeng, H. Chen, and C. Miao, “A survey on reinforcement learning for recommender systems,” arXiv preprint arXiv:2109.10665, 2021.
X. Zhao, Y. Zhang, L. Xia, and J. Wu, “Deep reinforcement learning for list-wise recommendations,” *Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (KDD)*, pp. 2769–2777, 2019.
M. M. Afsar, T. Crump, and B. H. Far, “Reinforcement learning based recommender systems: A survey,” arXiv preprint arXiv:2101.06286, 2021.
Roderick, M., MacGlashan, J., & Tellex, S. (2017). Implementing the Deep Q-Network. arXiv preprint arXiv:1711.07478. https://arxiv.org/abs/1711.07478
Zheng, G., Noroozi, V., & Yu, P. S. (2021). Recommendations with negative feedback via pairwise deep reinforcement learning. Proceedings of the 14th ACM International Conference on Web Search and Data Mining (WSDM ’21), 620–628.
Patoulia, A. A., Kiourtis, A., Mavrogiorgou, A., & Kyriazis, D. (2022). “A comparative study of collaborative filtering in product recommendation.” Emerging Science Journal, 7(1), 1–15. https://doi.org/10.28991/ESJ-2023-07-01-01
Zhang, W., Wang, J., & Yu, Y. (2012). “Serendipitous personalized ranking for top-N recommendation”. Proceedings of the 6th ACM Conference on Recommender Systems (RecSys '12), 23–30. https://wnzhang.net/papers/seren-wi.pdf.
Kim, S., Jeon, S., Lee, J., & Kang, J. (2022). Diversely regularized matrix factorization for accurate and aggregately diversified recommendation. arXiv preprint arXiv:2211.01328. https://arxiv.org/abs/2211.01328
Haarnoja, T., Zhou, A., Abbeel, P., & Levine, S. (2018). Soft actor-critic: Off-policy maximum entropy deep reinforcement learning with a stochastic actor. https://arxiv.org/abs/1801.01290
Lv, Z., & Tong, X. (2023). A reinforcement learning list recommendation model fused with graph neural networks. Electronics, 12(18), 3748. https://doi.org/10.3390/electronics12183748.
Chen, H., Dai, X., Cai, H., Zhang, W., Wang, X., Tang, R., Zhang, Y., & Yu, Y. (2018). Large-scale interactive recommendation with tree-structured policy gradient. arXiv preprint arXiv:1811.05869. https://arxiv.org/abs/1811.05869
