Networked Hatred: Co-evolution of Dehumanising Discourse and Channel Structure of Russian and Ukrainian Telegram During the 2022 Invasion

This article investigates how dehumanising discourse emerged and spread within Russian and Ukrainian Telegram channels as military hostilities unfolded at the onset (24 December 2021-24 April 2022) of the Russia’s 2022 full-scale invasion of Ukraine, from a networked approach. The study aligns with the Workshop’s focus on hostile communication by: (i) isolating a specific hostile style, outgroup dehumanisation (ii) analysing it longitudinally and comparatively in two non-Western media systems, and (iii) contributing to understanding of how platform (Telegram) features shape diffusion and reception. It also speaks to elite communication: preliminary patterns indicate more strategic, top-down deployment of dehumanising language by state-affiliated actors (including politicians) in Russia, compared with more reactive, grassroots use in Ukrainian channels. Dehumanisation is understood as perceiving or treating an outgroup as less than human. It includes denying agency and communality, deindividuation, extreme negative evaluation and delegitimising beliefs (e.g., seeing the outgroup as not sharing prosocial values), and construing fundamental difference consistent with psychological distancing. Intergroup dehumanisation hinders empathy, undermines prosocial behaviour, and contributes to aggression; in wartime, it can foster support for violence and extreme outgroup-targeted policies. Because social media both reflects and shapes intergroup dynamics, it offers a powerful lens on such language. We first define and operationalise outgroup dehumanisation drawing on psychological theory and linguistic markers. We then assemble a manually labelled dataset (n = 225) and utilise large language models to classify posts from leading news and political Telegram channels in Russia and Ukraine as dehumanising or non-dehumanising (n = 238,190). Building on our prior work showing that dehumanisation strengthens ingroup cohesion under conflict, and that this relation differs for aggressor and defender, we examine how its prevalence and reception evolved through the invasion’s onset. Two complementary models link discourse to networked dynamics. Model 1 uses a generalized linear mixed model to estimate time-series dynamics of public reception (views, reactions, forwards) as a function of discourse type, channel affiliation, and structural position. Model 2 treats channels as the unit of analysis and applies a Stochastic Actor-Oriented Model to the forwarding network to study co-evolution between ties (who forwards whom) and channel attributes (including each channel’s share of dehumanising content). Guided by a relational framework that views publics as dynamic networked spaces (Emirbayer & Sheller, 1998), we test whether shifts in narratives reshape interaction patterns and whether evolving connections, in turn, condition which frames gain prominence. By doing so, the study addresses two research questions: RQ1: How was dehumanising discourse received and amplified within the Russian and Ukrainian public spheres across stages of the invasion? RQ2: How did dehumanising frames co-evolve with the topology of the Telegram network over time? While full estimation is pending, preliminary analyses indicate growing centrality of dehumanising language in both countries, earlier emergence in Russia, increasing audience responsiveness to dehumanising posts, and feedback between discursive escalation and advantageous network positions. The approach illuminates structural logics behind the diffusion of hostile rhetoric and the network drivers shaping its escalation in wartime public spheres.