Two-tower models are identifiable without swaps when feature distributions overlap across positions, and strong logging policies degrade them only through model misspecification.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.IR 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Unidentified and Confounded? Understanding Two-Tower Models for Unbiased Learning to Rank (Extended Abstract)
Two-tower models are identifiable without swaps when feature distributions overlap across positions, and strong logging policies degrade them only through model misspecification.