A transformer pretrained with a distributionally robust loss on user behavior logs is claimed to beat ten baselines on next-behavior, few-shot, generation, and cross-domain tasks, with fitted scaling exponents alpha=0.51, beta=0.23.
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
REJECT 1representative citing papers
citing papers explorer
-
BehaveGPT: A Foundation Model for Large-scale User Behavior Modeling
A transformer pretrained with a distributionally robust loss on user behavior logs is claimed to beat ten baselines on next-behavior, few-shot, generation, and cross-domain tasks, with fitted scaling exponents alpha=0.51, beta=0.23.