Relevance based Bilingual Sarcasm detection using Transformer Model

Abstract

Introduction: The COVID-19 pandemic has underscored the importance of effective communication and understanding of public sentiment across diverse linguistic contexts. In response, this study proposes a novel approach for relevance-based bilingual sarcasm detection using a Transformer Model, specifically BERT (Bidirectional Encoder Representations from Transformers). Leveraging BERT’s bidirectional context representation, our model aims to enhance the accuracy of sarcasm detection in bilingual contexts.

Objectives: Propose a relevance-based approach for bilingual sarcasm detection using Transformer models. Evaluate the effectiveness of our approach on a dataset of bilingual sarcastic and non-sarcastic text. Compare the performance of our model with baseline methods namely, DNN, CNN, LSTM, BiLSTM, ANN, BERT, and BERT-LSTM. Discuss the implications of our findings for improving communication and sentiment analysis in the context of the COVID-19 pandemic and beyond.

Methods: The proposed method incorporates BERT’s contextual embeddings, a relevance checking engine, POS tagging, and a sarcasm classification transformer trained on annotated Twitter data. Data preprocessing includes hashtag-based extraction, manual annotation, and multilingual tokenization. Fine-tuning the BERT model enhances performance in capturing sarcastic language across Hindi and English.

Results: The model demonstrates superior performance compared to traditional approaches with a precision of 0.94, recall of 0.99, and F1-score of 0.94 for sarcastic tweets. Evaluation metrics surpass benchmarks on bilingual sarcasm detection datasets, affirming the robustness of the proposed method.

Conclusions: Integrating contextual relevance and part-of-speech features improves sarcasm detection in bilingual contexts. Our Transformer-based approach shows promise for multilingual sentiment analysis and communication monitoring, particularly during global events like the COVID-19 pandemic.