Social media's global reach makes understanding content virality crucial across diverse fields. During intense intergroup conflicts, this understanding becomes particularly vital. Prior research, largely conducted in the US, has identified negative emotions, moral outrage, and group identity as key correlates of engagement. This study extends this research to the Ukrainian context, a non-WEIRD setting experiencing severe intergroup conflict, examining the association between content expressing ingroup solidarity and outgroup hostility and engagement before and after the 2022 Russian invasion. Social media algorithms utilize engagement metrics (shares, likes, reactions) to curate content, creating a complex interplay between algorithms and human behavior. While algorithm access is limited, researchers can study linguistic patterns in posts to understand engagement drivers. Previous work showed that outgroup mentions were strong correlates of engagement in US political contexts, highlighting the role of social identity cues. Social Identity Theory (SIT) posits that individuals derive part of their identity from group memberships and make comparisons between ingroups and outgroups, offering insights into intergroup conflict. While SIT has been extensively studied, its application to social media is less explored. Some suggest social media exacerbates polarization by increasing social identity salience, but this effect varies across contexts. Research on media and polarization in non-WEIRD contexts is scarce, with the US dominating this field. In the US, negative outgroup cues seem more impactful than positive ingroup cues. However, SIT theorists argue that ingroup-favoring motivations might be more relevant than outgroup derogation, especially after severe shocks or threats. Studies suggest ingroup solidarity is more relevant than outgroup hostility after significant events. This raises questions about the conditions under which outgroup hostility and ingroup solidarity gain traction on social media. Drawing on Intergroup Emotions Theory (IET), which links group identification with emotions and behaviors, this study aims to investigate if social media posts expressing ingroup solidarity and outgroup hostility gain more engagement after the 2022 invasion of Ukraine. This explores the prediction that ingroup solidarity may be more engaging than outgroup hostility during conflict and addresses the gap in non-WEIRD context research. The historical context of the Russia-Ukraine war, starting with the Euromaidan revolution and the annexation of Crimea, is significant. The 2022 invasion represents a high-profile case of affective polarization culminating in war, fully captured on social media. A large body of research explores Ukrainian ethnic, civic, and linguistic identities and their interaction with pro-Russian identities. After the Euromaidan revolution, Ukrainian identity became more anti-Russian, except in Donbas. Figure 1 illustrates the dramatic decrease in positive attitudes between Ukrainians and Russians after the invasion, signifying intensified intergroup tensions.

The literature review draws upon existing research on social media engagement, social identity theory (SIT), and intergroup emotions theory (IET). Studies primarily from the US context have highlighted the role of negative emotions, moral outrage, and group identity cues in driving social media engagement. However, this research has limitations in generalizability to non-WEIRD contexts and diverse cultural settings. SIT provides a theoretical framework to understand intergroup dynamics and conflict. While SIT's role in shaping online behavior is less understood, studies suggest social media can amplify polarization by making social identity more salient. The interplay of social media algorithms and human behavior complicates this relationship, with the algorithms promoting content that maximizes engagement metrics. The limited access to these algorithms has pushed researchers to focus on linguistic patterns and social cues as proxies to analyze content popularity. The role of ingroup solidarity and outgroup hostility in shaping engagement during intergroup conflicts is a crucial aspect, where some scholars argue for the stronger influence of ingroup-favoring motivations than outgroup derogation. IET extends SIT by incorporating the impact of emotions in the intergroup behavior, emphasizing the interplay and bidirectional influence between emotions, identity, and behavior. While IET doesn't explicitly address whether ingroup solidarity or outgroup hostility is more impactful during specific conflict stages, the theory's proponents call for further investigation into the potential superiority of ingroup positivity in driving conflict.

This study employed a mixed-methods approach, combining quantitative analysis of social media data with qualitative textual analysis to understand the linguistic patterns associated with user engagement. Three studies were conducted. Study 1 replicated previous research in the context of Ukraine before the 2022 invasion. Data were collected from Facebook and Twitter from July 2021 to February 2022 from pro-Ukrainian and pro-Russian news sources. Posts were analyzed based on mentions of ingroup and outgroup identities, along with negative, positive, and moral-emotional language. Mixed-effects linear regressions were used to model the association between these variables and log-transformed engagement (sum of reactions +1). Study 2 examined engagement in pro-Ukrainian news sources after the invasion (February 2022 to September 2022). In addition to the measures used in Study 1, two classifiers—ingroup solidarity and outgroup hostility—were created using machine learning techniques on a labeled subset of posts. Mixed effects linear regressions predicted log-transformed engagement based on these new classifiers, identity mentions, and emotional language, analyzed across 14-day intervals, capturing temporal trends. Study 3 extended the analysis to non-news social media data. A geolocated sample of 148,959 pro-Ukrainian tweets (July 2021 to September 2022) was analyzed using a fine-tuned ROBERTa model to classify posts’ viewpoints, focusing on pro-Ukrainian posts after the invasion. The ingroup solidarity and outgroup hostility classifiers from Study 2 were further fine-tuned on this data. Mixed effects linear regressions modeled engagement, controlling for user verification and follower count. All statistical tests were two-tailed, using Satterthwaite d.f. for p-values and Cohen's d. Robustness checks were conducted using robust mixed effects models and different dictionary variations. The data for Studies 2 and 3 were collected retrospectively. The classifiers were built using manually labeled posts and achieved an accuracy of around 0.87 and an F1-macro of 0.80. Dictionaries were also created for ingroup solidarity and outgroup hostility to provide a complementary approach to the classifier. These were used to count the number of words from each category in posts. The engagement metric was calculated as the sum of all platform reactions, allowing for a direct comparison across social media platforms. To examine trends over time, a sliding window approach was used in Study 2, fitting regressions on 14-day intervals and shifting the window by one day. The use of custom dictionaries and fine-tuned BERT-style models provided a robust approach for capturing the nuanced linguistic expressions of ingroup solidarity and outgroup hostility.

Study 1 replicated previous findings showing that outgroup mentions were the strongest predictor of engagement before the invasion, consistent across pro-Ukrainian and pro-Russian news sources on Facebook and Twitter, with the exception of pro-Russian Twitter. After the invasion, outgroup mentions and ingroup mentions became less strongly associated with engagement, with effects dropping on both platforms for each additional word referring to the outgroup (Facebook: 16% decrease, Twitter: 23% decrease) and the ingroup (Facebook: 11% decrease, Twitter: 16% decrease), controlling for other variables. Post-invasion, positive and moral-emotional language were associated with engagement, while negative language was not significant. Study 2 showed that before the invasion, ingroup solidarity was associated with more engagement than outgroup hostility. However, after the invasion, ingroup solidarity became the strongest predictor of engagement on Facebook (92% increase) and Twitter (68% increase), whereas outgroup hostility showed minimal effect (1% increase on Facebook, no effect on Twitter). The proportions of posts classified as ingroup solidarity and outgroup hostility increased significantly after the invasion on both platforms. Study 3 replicated the post-invasion findings of Study 2 using a geolocated sample of non-news tweets in Ukraine. Ingroup solidarity predicted greater engagement (14% increase) than outgroup hostility (7% increase) and ingroup/outgroup mentions (4% and 7% increases respectively) after the invasion. These results suggest that the relationship between identity language and engagement is highly context-dependent. The findings are robust to different approaches like using custom dictionaries and BERT-style models.

This study’s findings demonstrate a shift in the factors driving social media engagement in Ukraine around the 2022 invasion. Before the invasion, outgroup mentions, aligning with previous US-based research, were associated with heightened engagement, mirroring patterns observed in highly polarized settings where negativity drives interactions. However, the post-invasion context reveals a different pattern. Ingroup solidarity emerges as the dominant driver of engagement, surpassing the influence of outgroup hostility. This suggests a potential shift in the dynamics of online interaction during periods of acute intergroup conflict. This transition aligns with Intergroup Emotions Theory (IET), where ingroup-focused emotions and behaviors might outweigh outgroup negativity. The observed increase in ingroup solidarity on social media might reflect a consolidation of national identity and a collective mobilization effort in response to the invasion. This phenomenon aligns with “rally ‘round the flag” effects observed in other conflict contexts. However, several alternative explanations are considered, such as algorithmic changes, reluctance to share hostile content, or fear of repercussions. Although the study can not definitively distinguish between these explanations, the consistency of findings across multiple platforms weakens the case for purely algorithmic explanations. The study highlights the importance of contextual factors in shaping social media engagement and the need to further study the complex interaction between human behavior and algorithmic effects. Future research should explore the causal pathways between identity expression, emotional responses, and engagement behaviors. The findings suggest that interventions designed to reduce affective polarization should consider ingroup solidarity as a potential positive force. Further research could explore these effects cross-culturally, comparing different conflict contexts and examining the role of moderation policies in social media platforms.

This study demonstrates a shift in the predictors of social media engagement in Ukraine following the 2022 Russian invasion. Before the invasion, outgroup mentions drove engagement, but post-invasion, ingroup solidarity became the most significant predictor. This finding highlights the context-dependent nature of social media engagement and the potential for ingroup solidarity to outweigh outgroup hostility during intense intergroup conflict. Future research should explore the causal mechanisms underlying this phenomenon and investigate the implications for interventions aimed at mitigating affective polarization. Further research should also compare engagement patterns across different platforms, conflicts, and cultures to test the generality of these findings.

The study has several limitations. Studies 2 and 3 employed an exploratory design, where classifiers and dictionaries were not created before data collection. The correlational nature of the findings does not allow for causal inferences. The reliance on existing engagement metrics as a measure of virality might overlook other aspects of online diffusion. The study does not capture the full extent of social media activity in the Russian context due to platform bans, limiting the comparative analysis. The generalizability of findings might be limited to the specific socio-political context of the Ukraine-Russia war. Further experimental studies are necessary to validate the findings and isolate the underlying psychological processes.