ADN-BERT: Attention-Based Deep Network Model Using BERT for Sarcasm Classification

Abstract

Sarcasm classification has gained popularity among researchers due to its complexity and implicit form of context representation. The most challenging aspect of sarcasm identification is to understand the exact context behind the statement. Therefore, a context-based model helps solve this critical task in the research domain. The importance of sarcasm detection and classification is primarily helpful for society to avoid misinterpreting statements or reviews that could affect one’s mental condition and perception. Unfortunately, to increase the retention level of audience toward media news, often the media incorporate sarcasm in their news headlines. However, people find it difficult to detect sarcasm in news headlines, resulting in them having a false impression of the news and spreading it to their surroundings. Therefore, it has become a significant concern among the society to develop a sense of understanding of the hidden context behind the statement before making any type of impression and judgment. In this paper, we have focused on this problem statement and developed a novel technique called an Attention-based deep network model using Bidirectional Encoder Representations from Transformers (BERT) i.e., ADN-BERT to classify news headlines into sarcastic and non-sarcastic categories. The News Headlines dataset is collected from Kaggle, and we have incorporated text augmentation to increase the size of the dataset to yield better results and accuracy. The evaluation metrics used in our study include accuracy, recall, precision, and F1-score. Our proposed model (ADN-BERT) has outperformed the recent state-of-the-art techniques with 94.1% accuracy, 95.2% recall, 94.5% precision, and 94.8% F1-score.Please check and approve the edits made in the chapter title and running title.Edited. ApprovedPlease check and confirm if the author names and initials are correct.Yes, correct.