Cross Domain Emotion Recognition using Few Shot Knowledge Transfer
This work addresses emotion recognition for text analysis in domains with limited labeled data, but it is incremental as it applies existing few-shot methods to new datasets.
The paper tackled the challenge of emotion recognition from text by transferring knowledge from the GoEmotions Reddit dataset to the SemEval tweets corpus using few-shot techniques, achieving results comparable to supervised baselines with minimal supervision.
Emotion recognition from text is a challenging task due to diverse emotion taxonomies, lack of reliable labeled data in different domains, and highly subjective annotation standards. Few-shot and zero-shot techniques can generalize across unseen emotions by projecting the documents and emotion labels onto a shared embedding space. In this work, we explore the task of few-shot emotion recognition by transferring the knowledge gained from supervision on the GoEmotions Reddit dataset to the SemEval tweets corpus, using different emotion representation methods. The results show that knowledge transfer using external knowledge bases and fine-tuned encoders perform comparably as supervised baselines, requiring minimal supervision from the task dataset.