Combat Online Incivility by Designing Detection Systems in Hong Kong

The rise of digital communication has amplified both political engagement and online incivility, threatening democratic discourse. This study investigates how artificial intelligence can effectively detect and mitigate incivility in Hong Kong’s Cantonese-speaking online communities. Focusing on LIHKG, Discuss HK, and Hong Kong Golden Forum between 2019 and 2020, it integrates contextual, linguistic, and cultural insights into AI-driven classification. A customized BiGRU-Transformer model, enhanced through improved sampling, annotation, and feature selection, was developed to identify four problematic behaviors: incivility, intolerance, toxicity, and severe toxicity. By combining computational modeling with qualitative content analysis, the research advances the precision and adaptability of incivility detection. Addressing limitations in existing Western-centric frameworks, this study contributes to computational social science and digital governance by offering a theoretically grounded, culturally sensitive approach to understanding and managing online discourse in non-Western sociopolitical contexts.