Unlearning in Recommender Systems
The 49th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR 2026) took place from July 20 to 24, 2026, in Melbourne | Naarm, Australia. Organized by the ACM Special Interest Group on Information Retrieval, the conference coverd all aspects of information storage, retrieval, and dissemination.
At SIGIR 2026, BIFOLD researcher Pierre Sicco Lubitzsch, together with Maarten de Rijke (University of Amsterdam) and Sebastian Schelter, presented their paper on unlearning in recommender systems. It is listed among 61 Resource Papers at the conference.
The paper introduces ERASE, a benchmark for machine unlearning in recommender systems. It addresses gaps in existing benchmarks, such as limited task diversity, unrealistic deletion request sizes, and the absence of practical constraints like sequential unlearning and efficiency. ERASE covers three recommender tasks, seven unlearning algorithms, nine datasets, and nine state-of-the-art models, along with more than 600 GB of reusable artifacts, including over 1,000 model checkpoints.
Publication:
ERASE - A Real-World Aligned Benchmark for Unlearning in Recommender Systems
Authors: Pierre Sicco Lubitzsch, Maarten de Rijke, Sebastian Schelter
Link: https://doi.org/10.48550/arXiv.2603.08341