Autoregressive semantic ID generation creates tree-induced probability correlations that prevent generative recommenders from capturing simple patterns; Latte adds latent tokens to relax these correlations.
How powerful is graph convolution for recommendation? InCIKM, pages 1619–1629
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
-
Expressiveness Limits of Autoregressive Semantic ID Generation in Generative Recommendation
Autoregressive semantic ID generation creates tree-induced probability correlations that prevent generative recommenders from capturing simple patterns; Latte adds latent tokens to relax these correlations.