Multi-domain Dialog State Tracking using Recurrent Neural Networks

TL;DR

This paper introduces a recurrent neural network-based approach for multi-domain dialog state tracking, enabling models to operate across various domains and improve performance using out-of-domain data for initialization.

Contribution

The paper presents a novel training procedure that leverages out-of-domain data to initialize belief tracking models for new domains, enhancing cross-domain performance.

Findings

Models trained on multiple domains outperform domain-specific models.

Out-of-domain data improves belief tracking performance in new domains.

The proposed method reduces the need for extensive in-domain data.

Abstract

Dialog state tracking is a key component of many modern dialog systems, most of which are designed with a single, well-defined domain in mind. This paper shows that dialog data drawn from different dialog domains can be used to train a general belief tracking model which can operate across all of these domains, exhibiting superior performance to each of the domain-specific models. We propose a training procedure which uses out-of-domain data to initialise belief tracking models for entirely new domains. This procedure leads to improvements in belief tracking performance regardless of the amount of in-domain data available for training the model.