In this research, the authors address challenges in empathetic dialogue generation.

They highlight that existing approaches often treat emotions as static throughout conversations, leading to inappropriate responses.

They also mention potential conflicts between common-sense knowledge and emotions in generating responses.

To tackle these issues, they propose the Serial Encoding and Emotion-Knowledge interaction (SEEK) method.

1. Dynamic Emotion Handling
2. Emotion-Knowledge Interaction

Overall, experiments on EmpatheticDialogues show that SEEK outperforms strong baseline models in both automatic and manual evaluations.

Powerpoint for this talk

Reference Paper

• COMET: Commonsense Transformers for Automatic Knowledge Graph Construction
• Towards Empathetic Open-domain Conversation Models: A New Benchmark and Dataset
• Knowledge Bridging for Empathetic Dialogue Generation