Enhancing sarcasm detection on social media: A comprehensive study using LLMs and BERT with multi-headed attention on...

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Computational Linguistics

Background:

• Sarcasm detection in natural language processing (NLP) is challenging due to subtle contextual cues in social media.
• Transformer-based models offer potential for improved sarcasm detection.
• The Self-Annotated Reddit Corpus (SARC) provides a large dataset for this task.

Purpose of the Study:

• To leverage transformer models (BERT, GPT-3, Claude-2, Llama-2) for enhanced sarcasm detection.
• To investigate the impact of multi-head attention mechanisms on model performance.
• To compare the effectiveness of fine-tuned transformer models on the SARC dataset.

Main Methods:

• Utilized transformer models including BERT, GPT-3, Claude-2, and Llama-2.
• Employed multi-head attention mechanisms to capture contextual relationships.
• Fine-tuned BERT, GPT-3, and Llama-2 models for comparative analysis on the SARC dataset.

Main Results:

• The fine-tuned BERT-based model achieved state-of-the-art performance.
• Achieved precision, recall, F1 score, and accuracy of 0.918, 0.917, 0.917, and 0.917, respectively.
• Demonstrated superiority over other evaluated models through statistical validation and error analysis.

Conclusions:

• Fine-tuned BERT with multi-head attention is highly effective for sarcasm detection.
• Highlights the importance of fine-tuning and attention mechanisms in NLP tasks.
• The proposed method offers a robust solution for identifying sarcasm in social media text.