Emotion Cause Pair Extraction By Multi Task Learning on Enhanced English Dataset

• Published in: Procedia computer science 2023, Vol.218, p.766-777
• Main Authors: Khunteta, Arunima, Singh, Pardeep
• Format: Article
• Language: English
• Subjects: emotion cause extraction; emotion recognition; natural language processing
• ISSN: 1877-0509; 1877-0509
• DOI: 10.1016/j.procs.2023.01.057

Emotion analysis and detection is a step further in sentiment analysis that aims to detect the emotion portrayed. A deeper task in mining opinions and emotions is to know the trigger behind the emotion or its stimuli and that is achieved through Emotion Cause Extraction. However, since it requires annotating sentences, limited research has been carried out in this area. Hence, there is a need to explore emotion cause extraction and expand its dataset in the most used language worldwide - English. This paper aims to extract causes without needing annotation of emotion by the proposed approach of Emotion Cause Pair Extraction on English language. For this, we have constructed a dataset of about 6000 English emotionally annotated sentences spanning across six primary emotions - happy, sad, fear, guilt, shame, disgust. Added to this, we have proposed algorithms to extract the causes of emotions from these sentences. The experimental models of our proposed approach gives encouraging results (a ∼ 9% increase from the state of the art methods) and also uses a larger dataset as compared to existing research on English language. We conclude by sharing the results, and discussing future advancements in this field.