Combining opinion and structural similarity in link recommendations to counter extreme polarization

A team of researchers including Gabriella D. Franco has published a new study on arXiv that tackles one of social media's most persistent problems: algorithmic polarization. The paper, 'Combining opinion and structural similarity in link recommendations to counter extreme polarization,' investigates how the two primary mechanisms behind friend suggestions—triadic closure (structural similarity) and homophily (opinion similarity)—jointly shape online discourse. Their agent-based modeling reveals that algorithms relying heavily on either mechanism alone create polarized outcomes, but in different ways. Pure opinion-based recommendations lead to high variation in opinions within echo chambers, while pure structural recommendations (friend-of-a-friend) create highly fragmented networks that coalesce around a single dominant view.

The key breakthrough is that a hybrid approach can mitigate these extremes. The researchers demonstrate that introducing even a weak dependence on structural similarity into a strongly homophilic (opinion-based) recommendation system prevents the network from splintering into disconnected components. This combined strategy fosters environments where moderate opinions can coexist, even in settings predisposed to polarization. The work provides a mathematical and simulation-based framework that could directly inform the next generation of social media algorithms. Instead of optimizing purely for engagement, platforms could implement this balanced approach to actively design for healthier public discourse and reduce the societal harms of extreme polarization.