Technology-mediated group toxicity polarization is a major socio-technological issue of our time. For better large-scale monitoring of polarization among social media news content, we quantify the toxicity of news video comments using a Toxicity Polarization Score. For polarizing news videos, our premise is that the comments’ toxicity approximates either an “M” or “U” shaped distribution—that is, there is unevenly balanced toxicity among the comments. We evaluate our premises through a case study using a dataset of ~180,000 YouTube comments on ~3,700 real news videos from an international online news organization. Toward polarization-mitigating information systems, we build a predictive machine learning model to score the toxicity polarization of news content even when its comments are disabled or not available, as it is a current trend among news publishers to disable comments. Findings imply that the most engaging news content is also often the most polarizing, which we associate with increasing research on clickbait content and the detrimental effect of attention-based metrics on the health of online social media communities, especially news communities.