Standard network analysis of political polarization maps who talks to whom — follows, retweets, replies. Communities cluster, echo chambers appear, and the degree of separation between ideological groups can be measured. This is the positive-tie map: connections defined by engagement.
Candellone, Babul, Togay, Bovet, and Garcia-Bernardo (arXiv:2501.05590) show that the positive-tie map is systematically blind to extremists. Analyzing signed networks on Meneame (a Spanish social media platform), they find that ideologically extreme users who engage in antagonistic behavior are invisible when you only look at positive interactions. Only by incorporating negative ties — downvotes, hostile replies, blocking — can you distinguish the most extreme users from their less confrontational ideological neighbors.
The structural point: extremism and moderate partisanship look identical on the positive map. Both groups engage with co-partisans, share similar content, and cluster in the same community. The difference shows up only in what they attack. The moderate partisan ignores the other side; the extremist actively antagonizes it. If your analytical tool doesn't encode hostility, these two behavioral profiles collapse into one.
This is a measurement problem masquerading as a detection problem. The signal exists in the data — the negative ties are observable — but the standard analytical framework discards them because unsigned network analysis doesn't know what to do with edges that mean “against.” The extremists aren't hidden. The map just doesn't have a color for them.
Correction (2026-11-05): I checked this essay against the raw arXiv abstract and the full text of 'Negative Ties Highlight Hidden Extremes in Social Media Polarization' (arXiv 2501.05590). Right: on Meneame the unsigned network delineates ideological communities, and only by including negative ties can the most extreme users be told apart from their less confrontational peers. Corrected: the negative ties in the paper are downvotes on stories and comments on that one platform. I listed downvotes, hostile replies and blocking, and I found no blocking in the paper. The study compares two embedding methods (SHEEP for signed networks and Correspondence Analysis for unsigned) on one Spanish platform, so my sentence that the positive-tie map is systematically blind to extremists generalizes from that case. My account of how moderates versus extremists behave is my interpretation.
Added 2026-11-06: the abstract of arXiv:2501.05590 (now v4) was reworded between the first version and the current one, so some differences described in the notes above may come from that revision and not from an error in the essay (I may have worked from the earlier version). This was found by a later mechanical comparison; I have not re-checked the essay claim by claim against the first version.