<p>We investigated online discourse on dehumanization and violence among polarized and non-apathetic neutral individuals, the latter defined as those who actively engage in political debates without being politically polarized. We employed text mining and machine learning to analyze over 39 million user-generated comments from two online forums—<i>Lihkg</i> (popular among anti-government users) and <i>HKDiscuss</i> (popular among pro-government users)—during the 2019 social unrest in Hong Kong. On <i>Lihkg</i>, non-apathetic neutral individuals expressed stronger dehumanizing sentiments compared to anti-government users. On <i>HKDiscuss</i>, pro-government users exhibited stronger dehumanizing tendencies compared to both non-apathetic neutral and anti-government individuals. Furthermore, non-apathetic neutral individuals on <i>Lihkg</i>, as well as non-apathetic neutral and anti-government-learning neutral individuals on <i>HKDiscuss</i>, were more likely to endorse violence compared with other groups. These findings suggest that non-apathetic neutrality can intensify online political polarization and conflict. Our results enhance the understanding of how online political polarization contributes to dehumanization and violence, underscoring the importance of further investigating non-apathetic neutrality in online discourse.</p>