Comparative Analysis of DistilBert, XLNET, RoBERTa & BigBird for Emotion Detection in Conversational Text
Download| Volume 6 Issue 1, 2025 | |
|---|---|
| Author(s): |
Arslan Ahmed Sukkur IBA University, arslanahmed.mscss24@iba-suk.edu.pk Barkha Sukkur IBA Universit, barkhakanjwani.mscsses23@iba-suk.edu.pk Tushar Kanjwani Sukkur IBA University, tusharkanjwani.becsef21@iba-suk.edu.pk Shahzad Nasim Begum Nusrat Bhutto Women University,Sukkur, shahzad.nasim@bnbwu.edu.pk |
| Abstract | Detection of Emotion from Conversational Data is the key to improving human-computer interactivity and under- standing how emotions work in digital communication. This paper investigates the role of four latest sophisticated models based on Transformer architecture includes DistilBERT, XLNet, RoBERTa and Bigbird in achieving high accuracy for emotion recognition from text-based dialogues. This research employs a comprehensive methodology of text cleaning, label encoding, model training on the diverse dataset collected from social media platforms and chat logs for six key emotion classes: anger, fear, joy, love, sadness, and surprise. The model was evaluated using measures of accuracy, precision, recall F1-score and confusion matrices. The findings indicated significant model performance variations, where RoBERTa achieved the highest validation accuracy while BigBird remained robust across different metrics particularly in identifying intricate emotional subtleties. The classification of ’surprise’ is often misperceived with both joy and love across all models, which the analysis also flagged as a consistent challenge. This study highlights both advantages and limitations of these models, while offering new grounds for improved accuracy in affective computing. The findings will have a substantial impact on how to create more empathetic and effective AI-driven communication tools like customer service, mental health therapy, and social media analysis. |
| Keywords | Emotion Detection, Conversational AI, Natu- ral Language Processing (NLP), Transformer Models, Affective Computing, Textual Analysis, Machine Learning, Emotional Dynamics, Sentiment Analysis |
| Year | 2025 |
| Volume | 6 |
| Issue | 1 |
| Type | Research paper, manuscript, article |
| Recognized by | Higher Education Commission of Pakistan, HEC | Category | Journal Name | ILMA Journal of Technology & Software Management | Publisher Name | ILMA University | Jel Classification | -- | DOI | - | ISSN no (E, Electronic) | 2790-590X | ISSN no (P, Print) | 2709-2240 | Country | Pakistan | City | Karachi | Institution Type | University | Journal Type | Open Access | Manuscript Processing | Blind Peer Reviewed | Format | Paper Link | https://ijtsm.ilmauniversity.edu.pk/arc/Vol6/i1/pdf1.pdf | Page | 1-12 |