Yazar "Özbey, Onur" seçeneğine göre listele

Listeleniyor 1 - 1 / 1
  • [ X ]
    Öğe
    A meta-learning-based approach for recommendation systems
    (Adana Alparslan Türkeş Bilim ve Teknoloji Üniversitesi, 2025) Özbey, Onur; Kılıç, Fatih
    Recommender systems play a key role in mitigating information overload by approximating user preferences in e-commerce, streaming, and social media, among other fields. Traditional factorization and neighborhood-based methods are prone to cold-start and sparsity, while deep learning techniques require large-scale interactions to be able to generalize. To alleviate these limitations, this thesis presents MAMLRec, a meta-learning-based recommendation model, which introduces the Model-Agnostic Meta-Learning (MAML) framework for rapid adaptation with sparse user-item interactions. MAMLRec is evaluated on four MovieLens datasets (100k, 1M, 10M, 20M) with regression metrics and ranking metrics. 10-20% improvements are obtained on small and medium datasets compared to state-of-the-art baselines such as SVD++, with comparable performance on large-scale data. More extended experiments with demographic features (age, gender, occupation, genre) also provide evidence of flexibility, even if state-of-the-art sequential models still achieve better ranking quality than MAMLRec. Overall, the findings indicate meta-learning as a promising solution to personalization under sparse data while acknowledging its scalability constraints for extremely large datasets. By positioning MAMLRec as an adaptive yet general-purpose framework, this dissertation contributes empirically and methodologically to meta-learning for recommender systems.