A multi-task user satisfaction model with SimCSE-style contrastive learning and domain-intent classification shows small gains on DuerOS, but test-set threshold tuning and input-label leakage weaken the evidence.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Reward-Driven Interaction: Enhancing Proactive Dialogue Agents through User Satisfaction Prediction
A multi-task user satisfaction model with SimCSE-style contrastive learning and domain-intent classification shows small gains on DuerOS, but test-set threshold tuning and input-label leakage weaken the evidence.