A new causal disentanglement objective for recommendation models improves out-of-distribution generalization under policy-induced distribution shift, with production A/B gains despite offline parity.
Deployed in Taobao display advertising production system since July 2022
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Causal Representation Learning for Generalisable Recommendation
A new causal disentanglement objective for recommendation models improves out-of-distribution generalization under policy-induced distribution shift, with production A/B gains despite offline parity.