[Paper Review] Computational Sarcasm Analysis on Social Media: A Systematic Review

Faria Binte Kader, Nafisa Hossain Nujat+4 more|arXiv (Cornell University)|Sep 13, 2022

Sentiment Analysis and Opinion Mining4 citations

TL;DR

This systematic review synthesizes recent advancements in computational sarcasm detection on social media, focusing on English-language datasets, features, methodologies, and challenges. It evaluates deep learning and transformer-based models, highlights the growing use of contextual and multimodal features, and provides comparative tables of datasets, features, and performance metrics to guide future research in sarcasm detection and sentiment analysis.

ABSTRACT

Sarcasm can be defined as saying or writing the opposite of what one truly wants to express, usually to insult, irritate, or amuse someone. Because of the obscure nature of sarcasm in textual data, detecting it is difficult and of great interest to the sentiment analysis research community. Though the research in sarcasm detection spans more than a decade, some significant advancements have been made recently, including employing unsupervised pre-trained transformers in multimodal environments and integrating context to identify sarcasm. In this study, we aim to provide a brief overview of recent advancements and trends in computational sarcasm research for the English language. We describe relevant datasets, methodologies, trends, issues, challenges, and tasks relating to sarcasm that are beyond detection. Our study provides well-summarized tables of sarcasm datasets, sarcastic features and their extraction methods, and performance analysis of various approaches which can help researchers in related domains understand current state-of-the-art practices in sarcasm detection.

Motivation & Objective

• To provide a comprehensive overview of recent trends and methodologies in computational sarcasm detection on social media.
• To identify and categorize the most relevant datasets, features, and feature extraction techniques used in sarcasm detection research.
• To analyze the performance of various detection approaches, including deep learning and transformer-based models, and compare their effectiveness.
• To explore challenges such as contextual understanding, real-time processing, and lack of multimodal integration in current sarcasm detection systems.
• To extend beyond detection by examining related research areas such as sentiment analysis, opinion mining, and sarcasm in multimodal environments.

Proposed method

• Conducted a systematic literature review of peer-reviewed studies on computational sarcasm detection published between 2006 and 2022.
• Categorized and analyzed datasets based on source (e.g., Twitter, Reddit, TheOnion), annotation method (e.g., hashtags, manual labeling), and size.
• Classified sarcasm detection features into lexical, syntactic, semantic, contextual, and multimodal categories, with detailed extraction techniques for each.
• Evaluated detection methodologies, including rule-based systems, machine learning models, and deep learning architectures such as CNN-LSTM and BERT-based transformers.
• Mapped the evolution of sarcasm detection techniques over time, highlighting the shift from traditional NLP to pre-trained transformer models.
• Provided comparative tables summarizing dataset statistics, feature types and extraction methods, and model performance across different benchmarks.

Experimental results

Research questions

• RQ1 What types of features and datasets are most commonly used in sarcasm detection on social media?
• RQ2 How have methodologies for sarcasm detection evolved over the past decade, particularly with the rise of deep learning and transformers?
• RQ3 What are the primary challenges and limitations currently hindering high-accuracy sarcasm detection in real-world social media settings?
• RQ4 Beyond detection, what other research directions exist in sarcasm analysis, such as contextual understanding or multimodal sarcasm detection?

Key findings

• Transformer-based models, especially BERT and its variants, have become the dominant approach in sarcasm detection, significantly outperforming earlier rule-based and traditional machine learning methods.
• Contextual features such as conversational history, author information, and post sequence have been shown to improve detection accuracy, especially in sequential and dialogue-based sarcasm.
• Multimodal sarcasm detection—integrating text, image, and emoji—has emerged as a promising but underexplored direction, with limited datasets and inconsistent performance across studies.
• Despite advances, real-time sarcasm detection remains a major challenge, with only a few studies (e.g., Bharti et al. 2016) demonstrating scalable processing using Hadoop and MapReduce, achieving a 66% reduction in processing time for 1.4 million tweets.
• The use of pre-trained models fine-tuned on news headlines datasets (e.g., TheOnion) and transferred to social media data has shown potential for improving generalization and performance on low-resource domains.
• A lack of standardized benchmarks, inconsistent annotation practices, and limited availability of high-quality, diverse datasets remain key obstacles to reproducibility and progress in the field.