A systematic review of Hyperpartisan News detection: A Comprehensive Framework for Definition, Detection, and Evaluation

Abstract

Hyperpartisan news detection employs advanced technological methods to identify news articles characterized by extreme bias, reporting information in a highly polarized manner. The spread of such news could lead to detrimental effects in terms of social cohesion and facts-perception. This systematic review delves into the uncharted territory of hyperpartisan news detection, where a conspicuous absence of comprehensive studies has left a void in scholarly discourse. Following the PRISMA methodology, we selected 80 pertinent articles. The paper commences by addressing the inherent challenges arising from disparate and ambiguous definitions of hyperpartisan found in existing literature. Through a meticulous analysis, we propose a refined and comprehensive definition of hyperpartisan news, serving as a foundational framework for subsequent exploration. The review systematically evaluates Non-Deep Learning, Deep Learning (i.e. non-Transformer and Transformer-based methods), and mixed approaches employed in the identification of hyperpartisan content across various studies. By scrutinizing methodologies, algorithms, and key features, we categorize and compare their performances, shedding light on their efficacy and limitations. Furthermore, we conducted an examination of 36 datasets utilized in diverse studies, offering insights into the commonalities and variations in data sources. This inclusive dataset analysis contributes to a nuanced understanding of the challenges faced in the development and validation of hyperpartisan news detection models. This paper not only pioneers the systematic review of hyperpartisan news detection, but also establishes a robust foundation for future research endeavors in this critical domain. We emphasize the necessity of standardized definitions, rigorous evaluation metrics, and shared datasets to advance the field.