HyST learns per-slot whether to use full-distribution or candidate-generation methods for dialogue state tracking and reports 24% relative gain over prior SOTA on MultiWOZ-2.0.
An end-to-end trainable neural network model with belief tracking for task-oriented dialog
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2019 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
HyST: A Hybrid Approach for Flexible and Accurate Dialogue State Tracking
HyST learns per-slot whether to use full-distribution or candidate-generation methods for dialogue state tracking and reports 24% relative gain over prior SOTA on MultiWOZ-2.0.