Search papers, labs, and topics across Lattice.
This paper introduces the Emotional Inertia-Informed Supervised Contrastive Learning (EII-SCL) module, which enhances multimodal emotion recognition in conversation by incorporating the concept of emotional inertia into the learning process. By constructing inertia-affected samples within temporal windows, the method effectively captures the influence of emotional context shifts, leading to improved performance over existing models. Experimental results on the IEMOCAP and MELD datasets demonstrate that EII-SCL consistently surpasses state-of-the-art approaches in accuracy.
Emotional inertia can significantly enhance emotion recognition accuracy in conversation, leading to a leap in performance over traditional methods.
Multimodal emotion recognition in conversation (MERC) achieves accurate predictions by integrating multimodal and contextual information in dialogues. While current MERC approaches focus on modeling complex contextual dependencies in conversation, they often overlook the impact of contextual emotional inertia in emotion shift, leading to sub-optimal performance. To address this issue, we propose a novel Emotional Inertia-Informed Supervised Contrastive Learning module (EII-SCL) that informs the contrastive objective by constructing inertia-affected samples within temporal windows, effectively leveraging emotional inertia as a prior while enabling seamless integration with existing MERC models without requiring additional data. Extensive experiments on IEMOCAP and MELD show that our approach consistently outperforms state-of-the-art methods.
