Show Us the Way: Learning to Manage Dialog from Demonstrations
arXiv:2004.08114
Abstract
We present our submission to the End-to-End Multi-Domain Dialog Challenge Track of the Eighth Dialog System Technology Challenge. Our proposed dialog system adopts a pipeline architecture, with distinct components for Natural Language Understanding, Dialog State Tracking, Dialog Management and Natural Language Generation. At the core of our system is a reinforcement learning algorithm which uses Deep Q-learning from Demonstrations to learn a dialog policy with the help of expert examples. We find that demonstrations are essential to training an accurate dialog policy where both state and action spaces are large. Evaluation of our Dialog Management component shows that our approach is effective - beating supervised and reinforcement learning baselines.
arXiv comment8 pages + 2 pages references, 4 figures, 4 tables, accepted to DSTC8 Workshop at AAAI2020
References in corpus (8)
- Prioritized Experience Replay
- Exploration by Random Network Distillation
- A Network-based End-to-End Trainable Task-oriented Dialogue System
- Semantically Conditioned LSTM-based Natural Language Generation for Spoken Dialogue Systems
- BBQ-Networks: Efficient Exploration in Deep Reinforcement Learning for Task-Oriented Dialogue Systems
- Learning Montezuma's Revenge from a Single Demonstration
- ConvLab: Multi-Domain End-to-End Dialog System Platform
- Reward Shaping with Recurrent Neural Networks for Speeding up On-Line Policy Learning in Spoken Dialogue Systems