SSLCL: A Computationally Efficient and Model-Agnostic Supervised Contrastive Learning Framework for Emotion Recognition in Conversations
Supervised Sample-Label Contrastive Learning with Soft Hirschfeld-Gebelein-R{'e}nyi Maximal Correlation (SSLCL) is an efficient and model-agnostic supervised contrastive learning framework for the problem of Emotion Recognition in Conversations (ERC), which eliminates the need for a large batch size and can be seamlessly integrated with existing ERC models without introducing any model-specific assumptions. Extensive experiments on two ERC benchmark datasets, IEMOCAP and MELD, demonstrate the compatibility and superiority of our proposed SSLCL framework compared to existing state-of-the-art supervised contrastive learning (SCL) methods.
The overall framework of SSLCL is illustrated as follows, which is made up of three key components: sample feature extraction, label learning, and sample-label contrastive learning.
