Emotion and Sentiment Analysis in Dialogue: A Multimodal Strategy Employing the BERT Model
Mahesh Parmar, Akhilesh Tiwari · 2024
In the constantly changing field of natural language processing (NLP), understanding the intricacies of human emotion and sentiment in dialogues has become vital for applications ranging from virtual assistants to social interaction platforms. The task of classifying emotions and sentiments within dialogues has become both challenging and crucial. Human emotions, marked by their diverse intensities, are intricately influenced by independent and context-dependent factors within ongoing conversations. The need for multi-label emotion detection in conversations is paramount, allowing systems to grasp the nuanced emotional expressions of interacting users. Therefore, this study utilizes the Multimodal Emotion Lines Dataset, an enhanced version of Emotion Lines that includes over 13,000 utterances from 1,433 dialogue taken from the television series Friends, which is used in this work. We proposed the BERT model, renowned for its contextualized language representations, to capture the nuances of emotion and sentiment within the dynamic context of dialogues. The model's bidirectional architecture enables a comprehensive understanding of conversational context, allowing for more accurate and contextually aware analyses. The outcomes of our proposed approach exhibit exceptional performance compared to other base models, underscoring the efficacy of our methodology.