Using Ensemble of Transformers for Bragging Detection in Social Media

Bragging is a strategy that entails people emphasizing their positive traits in order to project a positive self-image. Bragging detection plays a significant role in the improvement of various NLP tasks, including personality analysis, mental health assessment, social media analytics, and user behavior modeling. However, the variety of contexts and subjective nature of bragging, the existence of subtle linguistic clues, the development of trends and methods, and the risk of false positives and negatives make the task more challenging. To address these challenges, we proposed a transformer-based ensemble approach to determine bragging on social media. We used three pretrained transformer models including BERTweet, RoBERTa, and XLMRoBERTa to exploit the multifarious contextual dimensions. Later, we fine-tuned these models to learn task-specific knowledge and employ a feed-forward linear layer to get the prediction. Finally, the majority voting-based ensemble approach is used to determine the final la bel. The experimental results indicate that our proposed approach outperforms the current state-of-the-art method by 0.60% in binary and demonstrates a significant 5.23% improvement in multiclass classification.