Leonardo Di Gioia · ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 2026
DOI: 10.1145/3799682.3840223
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).
Efficiency and trustworthiness constitute two of the pivotal desiderata in modern recommender systems architecture. In this sense, increasingly large and complex models can improve recommendation quality, while explainability mechanisms provide understandable and verifiable justifications for their outputs. However, the computational cost of high-capacity models often conflicts with industrial requirements such as low-latency inference and resource-efficient deployment. Knowledge distillation (KD) addresses this trade-off by transferring the behavior of a large teacher model to a smaller, more efficient student. Existing work, however, mainly evaluates whether the student preserves the teacher's ranking performance, without assessing whether it also preserves the evidence underlying its recommendations. Students' recommendations may rely on different user-item evidence, reducing the transparency and reliability of the deployed system. My doctoral research investigates how KD affects recommendation explanations and whether explanations can actively guide distillation. Initial experiments have evaluated teacher-student explanation consistency through different post-hoc methods and comparison protocols, together with recommendation and explanation-quality metrics. They provide an empirical basis for studying how preservation depends on the model, explanation space, and evaluation setting. Building on this analysis, the next stages will investigate explanation-guided distillation strategies to improve teacher-student fidelity, recommendation performance, and explanation reliability, as well as the use of explanations to identify and reduce noisy interactions.
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
ACM International Conference on Information and Knowledge Management (CIKM) 2026 · 0 citations
No comments yet — start the discussion below.