Variations in user state embeddings for CMAB recommenders can improve performance more than changing the bandit algorithm, with no embedding or aggregation strategy dominating across datasets.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.IR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
The Bandit's Blind Spot: The Critical Role of User State Representation in Recommender Systems
Variations in user state embeddings for CMAB recommenders can improve performance more than changing the bandit algorithm, with no embedding or aggregation strategy dominating across datasets.