Related Experiment Video
Updated: May 31, 2025
05:51
Exploring the Use of Isolated Expressions and Film Clips to Evaluate Emotion Recognition by People with Traumatic Brain Injury
Published on: May 15, 2016
9.0K
Low-resource MobileBERT for emotion recognition in imbalanced text datasets mitigating challenges with limited
Muhammad Hussain1, Caikou Chen1, Sami S Albouq2
1College of Information Engineering Yangzhou University, Yangzhou, PR China.
Plos One
|January 24, 2025
Summary
This study introduces Focal Weighted Loss (FWL) with adversarial training to improve emotion recognition in conversation (ERC). The novel approach enhances empathetic AI interactions without needing large datasets or computational resources.
Area of Science:
- Artificial Intelligence
- Natural Language Processing
- Computational Linguistics
Background:
- Emotion recognition in conversation (ERC) is crucial for empathetic AI.
- Challenges in ERC stem from weak emotion-semantics correlation and context dependency.
- Existing methods often require substantial computational resources.
Purpose of the Study:
- To propose a novel loss function, Focal Weighted Loss (FWL), for improved ERC.
- To address imbalanced emotion classification and reduce reliance on large batch sizes or computational power.
- To enhance the development of human-like dialogue systems.
Main Methods:
- Developed Focal Weighted Loss (FWL) combined with adversarial training.
- Integrated FWL with the compact MobileBERT language model.
- Evaluated the approach on four benchmark datasets: MELD, EmoryNLP, DailyDialog, and IEMOCAP.
Main Results:
- Demonstrated competitive performance across all four benchmark datasets.
- Validated the effectiveness of FWL with adversarial training in handling imbalanced emotion classification.
- Showcased potential for high performance under limited resource constraints.
Conclusions:
- The proposed FWL with adversarial training effectively improves emotion recognition in dialogue systems.
- This method offers a computationally efficient solution for empathetic AI development.
- The approach enables more human-like interactions on digital platforms, even with resource limitations.
