Hybrid Curriculum Learning for Emotion Recognition in Conversation
The paper introduces a hybrid curriculum learning framework that tackles emotion shift and confusing labels in emotion recognition in conversation by applying nested curriculum stages at both conversation and utterance levels, enabling a progressive easy‑to‑hard training that markedly boosts classic ERC model performance across four public datasets and is already deployed in Alibaba’s entertainment AI brain script health‑check service.
This paper presents a hybrid curriculum learning framework for emotion recognition in conversation (ERC), addressing two key challenges: emotion shift and confusing label. The framework employs nested curriculum learning at both conversation-level and utterance-level to create a progressive learning process from easy to difficult samples. Experimental results on four public datasets demonstrate that classic ERC models significantly improve performance when using this training framework, effectively mitigating the aforementioned issues. The method has been successfully applied in Alibaba's entertainment AI brain (Beidou Star) script health check business.
Youku Technology
Discover top-tier entertainment technology here.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.