一作SIGIR 2026Oral· Information Retrieval (Full)
CATS: Cluster-Aware Thompson Sampling for Negative Mining in Sequential Recommendation
Giulia Di Teodoro · Federico Siciliano · Nicola Tonellotto · Fabrizio Silvestri
摘要
Modern sequential recommendation systems often rely on negative sampling to efficiently train models over vast item corpora. However, common strategies such as uniform, popularity-based sampling, or hard-negative mining, often yield uninformative negatives, introduce popularity bias, or suffer fr...