A Survey of Challenges, Methods, and Trends in Sentiment Analysis and Sarcasm Detection

Document Type : Original Article

Authors

Computer Science Department, Faculty of Computers and Artificial Intelligence Helwan University, Cairo, Egypt

Abstract

This survey paper explores the intersection of Natural Language Processing (NLP), particularly sentiment analysis (SA) and sarcasm detection (SD), within machine learning. It emphasizes the challenges inherent in detecting sarcasm due to its implicit nature and contextual dependencies. The paper comprehensively examines the field, conducting a comparative analysis of various studies ranging from traditional machine learning to advanced deep learning methods. Identified challenges include sarcasm’s implicit nature, contextual impact, difficulties with short-text datasets, and the challenge of detecting both sarcasm and sentiment within a multilingual context, especially in low-resource languages. The survey discusses innovative solutions proposed by researchers, spanning classical machine learning to advanced deep learning approaches, such as aspect-based sentiment analysis and advanced transformer models. The methodology, which utilized bidirectional encoder representation transformers (BERT) for aspect-based sentiment analysis, achieves exceptional F1 scores on Twitter and Reddit datasets. Artificial neural networks demonstrate flawless accuracy in detecting sarcasm within emoticons and text. The usage of a modified switch transformer attains an impressive accuracy on the ArSarcasm dataset, highlighting the efficacy of its dynamic switching mechanism. Additionally, incorporating bidirectional LSTM results in significant performance improvements in sentiment and sarcasm classification tasks, evidenced by notable F1 scores.

Keywords

Main Subjects

Files

Share

How to cite

Statistics