Semi-Supervised Knowledge-Grounded Pre-training for Task-Oriented Dialog Systems

We present our models for Track 2 of the SereTOD 2022 challenge, which is the first challenge of building semi-supervised and reinforced TOD systems on a large-scale real-world Chinese TOD dataset MobileCS. We build a knowledge-grounded dialog model, S2KG to formulate dialog history and local KB as input and predict the system response.

This paper has been accepted at the SereTOD 2022 Workshop, EMNLP 2022

System Performance

Our system achieves the first place both in the automatic evaluation and human interaction, especially with higher BLEU (+7.64) and Success (+13.6%) than the second place. The evaluation results for both Track 1 and Track 2, which can be accessed via this this link.

S2KG for Generation

We release our S2KG-base model here. You can use this model for knowledge-grounded dialogue generation follow instructions S2KG.