Marialuisa Pisicchio, Alberto Carlo Maria Mancino, Leonardo Di Gioia, Daniele Malitesta, Fedelucio Narducci, Tommaso Di Noia · ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2026
DOI: 10.1145/3799682.3840179
Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
Knowledge Distillation (KD) improves the efficiency of recommender systems by transferring knowledge from high-capacity teacher models to lightweight student architectures. Despite its growing popularity, KD research for recommendation remains fragmented, making reproducibility and fair comparison across distillation strategies difficult. We present RecDistillery, the first open-source, model-agnostic framework for standardizing teacher–student KD workflows in recommender systems. The framework provides a modular architecture unifying data preparation, teacher integration, student training, distillation, and evaluation. It supports both native PyTorch-based recommenders and externally trained models, enabling heterogeneous teacher–student distillation within a unified pipeline. RecDistillery facilitates both the reproduction of existing KD methods and the rapid prototyping of new approaches through reusable losses, data loaders, metrics, and configuration modules. We demonstrate its capabilities through representative case studies involving multiple recommendation frameworks and externally imported teachers. The framework code, documentation, and demo examples are publicly available at https://github.com/sisinflab/RecDistillery.
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
No comments yet — start the discussion below.