SYMBOLIC TRAUMA AND MEDIA DISCOURSE: A SEMIOTIC READING OF ROMANIA’S DIGITAL PRESS ON THE WAR IN UKRAINE

DOI: https://doi.org/10.26758/16.1.39

(1) Faculty of Journalism and Communication Sciences, University of Bucharest, Romania e-mail: danpodaru@fjsc.ro

(2) Faculty of Journalism and Communication Sciences, University of Bucharest, Romania e-mail: antonio.amuza@fjsc.ro, https://orcid.org/0000-0003-2470-3432

Address correspondence to: Antonio AMUZA, University of Bucharest, Faculty of Journalism and Communication Sciences, Bucharest, Romania, Ph.: +40755020197, e-mail: antonio.amuza@fjsc.ro

Abstract

Objectives. This study examines the symbolic and affective dimensions of the war in Ukraine as reflected in Romania’s digital and media discourse. Beyond physical destruction, the conflict is explored as a generator of symbolic trauma and a catalyst for collective imaginaries shaped by fear, solidarity, guilt, and political polarisation. We investigate how these representations contribute to the emotional climate and public perceptions within Romanian society, particularly in the context of regional insecurity and the country’s ambivalent positioning between East and West.

Material and methods. We employ a qualitative research design combining discourse analysis and semiotic decoding. The corpus consists of over 73,000 articles published by Romania’s mainstream media outlets between February 2022 and February 2023. Using semiotic and narrative tools, we decode the affective charge of images, metaphors, and symbolic tropes related to the war, with attention to how different audiences might interpret them in relation to historical memory, political affiliation, and national identity.

Results. The analysis reveals two dominant symbolic axes: the spectacle of Ukrainian resistance and the trauma of civilian suffering, frequently articulated through binary oppositions such as “hero vs. aggressor” or “East vs. West.” National and religious symbols are reinterpreted as carriers of anxiety, resilience, or moral orientation. Clusters of discourse reflect distinct emotional registers, from outrage and grief to admiration and fear. Particularly salient are the ways in which institutional narratives attempt to depoliticise the conflict, while emotionally charged expressions (such as hyperbolic titles or mythologised accounts) tend to dominate attention. These findings suggest that war discourse in Romania is not only shaped by geopolitical agendas but also infused with affective intensity and symbolic contestation.

Conclusions. The war in Ukraine emerges not solely as a geopolitical crisis but as a symbolic event embedded in collective memory and emotional life. Through the Romanian media ecosystem, war becomes an emotionally mediated spectacle that fuels both symbolic trauma and ideological realignment. The study proposes that symbolic and affective communication plays a crucial role in shaping public narratives, and that periods of instability are accompanied by intensified semiotic production, where myth, fear, and identity become entangled.

Keywords: semiotics; cultural trauma; digital discourse; war in Ukraine; resilience; collective identity

Suggested citation (APA):

Podaru, D. N., & Amuza, A. (2026). Symbolic trauma and media discourse: A semiotic reading of Romania’s digital press on the war in Ukraine. Anthropological Researches and Studies, 16, 610–625. https://doi.org/10.26758/16.1.39

Introduction

The war in Ukraine, which erupted in February 2022 following Russia’s full-scale invasion, has triggered not only a geopolitical crisis of continental magnitude but also a profound symbolic disturbance within the European public sphere. Romania, as a country situated on NATO’s eastern flank and sharing a border with Ukraine, has experienced this conflict not only in terms of physical proximity and humanitarian response, but also as a media-saturated symbolic event. The digital reverberations of war (amplified by journalistic coverage, political communication, and affective circulation on social platforms) have redefined how conflict is perceived, narrated, and emotionally internalized by Romanian audiences.

In this context, the current study proposes to explore how the war in Ukraine has been mediated and re-signified through Romanian mainstream media, understood as a space of symbolic and affective production. Drawing on a corpus of over 73,000 news articles published between February 2022 and February 2023, selected from the most prominent Romanian outlets listed by SATI (Studiul de Audiență și Trafic Internet) under BRAT (Biroul Român de Audit Transmedia), the research investigates how meanings of heroism, suffering, moral legitimacy, and collective identity are constructed and negotiated in a time of crisis.

By combining computational topic modeling with qualitative semiotic analysis and sentiment extraction, the article aims to advance the understanding of war discourse as more than a transmission of facts. Rather, the war emerges as a symbolic spectacle, emotionally charged and culturally embedded, shaped by recurring narrative tropes and digital rituals. We argue that Romanian media discourse articulates the war through binary oppositions such as “us vs. them”, “victims vs. aggressors” mobilizing national and religious symbolism in ways that reflect deeper anxieties about security, belonging, and moral order. Furthermore, this study emphasizes the affective and mnemonic dimensions of digital war communication. Building on the theories of Alexander (2004) on cultural trauma, Papacharissi’s (2015) concept of affective publics, and Barthes’ (1972) reading of myth as depoliticized speech, we examine how emotional registers (such as fear, empathy, and anger) structure both collective memory and symbolic identity in Romania. The central research question guiding this work is: How does Romanian mainstream media symbolically frame the war in Ukraine, and what affective and ideological structures emerge through digital mediation? By integrating theoretical insights from semiotics, media studies, and trauma sociology with computational methods applied to a large-scale corpus, the article contributes to current debates on digital war discourse, symbolic politics, and the emotional configuration of European peripheries. In doing so, it also raises critical reflections about the role of journalism and algorithmic visibility in shaping not only what is known about war, but what is felt and remembered.

Theoretical framework

War is not communicated solely through factual reporting but through systems of signs that construct, organize, and legitimize social understandings of conflict. Semiotic theory argues that meaning emerges through culturally shared codes rather than through the direct representation of reality (Barthes, 1972; Eco, 1979; Danesi, 2004). Consequently, media discourse should be understood as a process of signification in which linguistic and visual elements transform military events into culturally intelligible narratives. Recent contributions to the semiotics of war further emphasize that armed conflicts continuously redefine symbolic codes, producing representations that legitimize violence, construct enemy images, and shape collective memory (Jacob, 2021). Rather than treating war as a purely military phenomenon, or propaganda in abstract, this perspective conceptualizes it as a communicative system in which language, images, symbols, and narratives generate meanings that structure public interpretations of conflict. As Jacob (2021) argues, war-related sign systems are central not only to the conduct of war but also to its legitimization, and subsequent commemoration.  Within this framework, media narratives do not simply describe reality but naturalize particular symbolic oppositions. Barthes (1972) conceptualizes myth as a second-order semiological system through which historically contingent meanings become perceived as self-evident, while Berger (1972) demonstrates that every image represents a particular “way of seeing.” Similarly, Danesi (2024) argues that contemporary digital environments increasingly rely on symbolic constructions capable of legitimizing ideological narratives through the strategic manipulation of signs.

These symbolic structures become socially consequential when they frame suffering as a shared cultural trauma. Semiotic representations acquire particular social significance when they become embedded in narratives of collective suffering. From the perspective of cultural sociology, trauma is not the automatic consequence of catastrophic events but the outcome of processes through which societies publicly recognize suffering (Alexander, 2004). Cultural trauma emerges when a community comes to perceive a disruptive event as fundamentally threatening its collective identity, transforming individual suffering into a shared symbolic experience. The media play a central role in this process because they do not merely report violence but actively shape its public intelligibility. Rather than functioning as neutral channels of information, media technologies structure the ways in which conflicts are perceived and interpreted (McLuhan, 1964). Through symbolic representations of suffering, they construct moral relationships between audiences and distant victims (Chouliaraki, 2006) and contribute to transforming catastrophic events into culturally recognized traumas (Alexander, 2004).

The computational analyses employed in this study identify recurrent lexical and thematic structures within the Romanian digital press. However, topic modeling and clustering describe patterns of textual similarity rather than their cultural significance. To explain how these discursive configurations contribute to the public understanding of war, the computational results are interpreted through a semiotic framework grounded in cultural sociology and media studies (Barthes, 1972; Alexander, 2004; Danesi, 2004).

Rather than analysing individual articles in isolation, the semiotic interpretation focuses on representative texts selected from each computationally identified cluster. This approach assumes that statistically coherent clusters also represent coherent symbolic configurations, allowing qualitative interpretation to build upon computational regularities instead of replacing them.

Material and methods

The analysis examines five complementary dimensions of meaning construction. First, symbolic actors identify the dominant social roles attributed to political leaders, institutions, civilians, or military actors (e.g., victim, aggressor, protector, witness). Second, symbolic signs and recurring motifs capture the lexical and visual elements through which conflict is represented, including references to territory, destruction, energy, humanitarian suffering, or international organizations. Third, binary symbolic oppositions explore how media discourse organizes reality through contrasting categories such as democracy versus authoritarianism, civilization versus barbarism, or security versus insecurity. Fourth, affective framing examines how symbolic structures are associated with dominant emotional registers identified through sentiment analysis. Finally, cultural myths synthesize these elements by identifying broader narratives through which the conflict is socially legitimized and rendered culturally meaningful.

Guided by the analytical framework outlined above, the empirical analysis combines computational text mining with qualitative semiotic interpretation in order to examine how symbolic trauma is articulated in Romanian digital news coverage. By integrating computational text analysis with qualitative semiotic interpretation, this framework enables the identification not only of what themes dominate media discourse but also of how these themes contribute to constructing symbolic trauma and public understandings of the war. The approach combines computational and interpretive methods in order to capture both large-scale discursive patterns and the symbolic logics embedded in them. The dataset consists of 73,570 unique news articles published between 24 February 2022 and 23 February 2023, all of which contain the keyword “ucraina” (Ukraine) in the title, body text, or metadata. These articles were extracted via automated web scraping from leading Romanian news outlets, selected based on audience rankings from the Romanian Audit Bureau of Circulations (BRAT) and the SATI platform. Included sources span the ideological spectrum and editorial formats – from broadsheet platforms (Adevărul, Hotnews, G4Media) to television-based portals (Digi24, Antena3, Știrile ProTV) and wire agencies (Mediafax).

Each entry includes the article’s publication date, outlet name, title, full text content, keywords (if available), and a column containing pre-processed words used in text analysis. Table 1 presents the ten most prolific sources in the dataset, with Adevărul.ro, Ziare.com, and HotNews.ro accounting for the majority of entries. On average, an article comprises approximately 372 words, with a median of 351 and a maximum exceeding 2000. The sample spans 26 unique news sources.

Table 1

Top 10 Media Sources by Article Count (to see Table 1, please click here)

The scraping was conducted using a custom script that extracted the following variables for each article: source (string), publication date (datetime), title (string), keywords (list), full text (string), and URL (string). The variable structure of the dataset is described in Table 2, including their types and intended analytical usage. These variables allowed for both structural (e.g., frequency, volume) and content-based (e.g., sentiment, topic modeling) analyses.

Table 2

Variable Types in the Dataset (to see Table 2, please click here)

The raw data was indexed and visualised in Elasticsearch to monitor the collection process and verify relevance and consistency. The final dataset was then exported in .csv format and pre-processed in R for textual and statistical analysis. All articles were cleaned and standardised using a pipeline that included lowercasing, punctuation and digit removal, tokenization, and removal of Romanian stopwords. We applied stemming and lemmatisation where necessary, depending on the analysis module. These preprocessing steps enabled the transformation of texts into a document-term matrix, which formed the basis of the subsequent analyses.

We employed a mixed-methods research design. On the quantitative side, we implemented Latent Dirichlet Allocation (LDA) with six topics (k = 6) to identify latent discursive structures. Several alternative topic models (k = 4–8) were initially explored. The six-topic solution was retained because it achieved the best balance between semantic coherence, topic exclusivity, and substantive interpretability. Models with fewer topics tended to merge conceptually distinct narratives, whereas models with more topics generated fragmented and overlapping thematic structures. In parallel, a sentiment and emotion analysis was carried out using the NRC Emotion Lexicon (Mohammad & Turney, 2013), which classifies words into ten categories—eight primary emotions (joy, trust, fear, surprise, sadness, disgust, anger, anticipation) and two sentiment polarities (positive, negative). The lexicon was applied to tokenised texts within each cluster, allowing for both quantitative assessment and temporal tracking of emotional shifts. We aggregated emotional frequencies on a monthly basis, enabling a dynamic mapping of affective responses throughout the year. A semantic co-occurrence network was generated using pairwise co-appearance of terms within documents. This network was visualised using the igraph and ggraph libraries in R, applying a Fruchterman-Reingold layout. We filtered the graph to include only frequent co-occurrences (n ≥ 5) and computed centrality metrics to identify key symbolic anchors in the war discourse. Words such as “children,” “Europe,” “heroes,” and “barbarians” emerged as highly central, indicating a moral and symbolic codification of conflict through binary oppositions.

Complementing the computational pipeline, a qualitative semiotic and narrative analysis was conducted for each cluster. Representative articles – identified through high topic probabilities and lexical density – were manually examined to uncover symbolic codes, mythologised figures, and affective rituals. Representative articles were selected according to three complementary criteria: (1) high topic probability within the respective LDA topic, (2) lexical density and thematic representativeness, and (3) consistency with the dominant cluster profile. The qualitative analysis did not seek to interpret isolated articles as individual cases but to identify recurrent symbolic structures shared across each computationally derived cluster. Following the analytical framework introduced above, each representative text was systematically examined with regard to (a) symbolic actors, (b) symbolic signs and motifs, (c) binary symbolic oppositions, (d) affective framing, and (e) cultural myths. These analytical dimensions were applied consistently across all clusters to enhance interpretive transparency and analytical comparability. All analyses were conducted in R using the packages tidytext, topicmodels, tm, dplyr, widyr, igraph, ggraph, lubridate, and ggplot2.

While the dataset covers a broad range of mainstream sources, it excludes user-generated content and social media, which may capture additional affective or informal discourse.

The study complies with the ethical standards of open-source media research. No personal or sensitive data were collected, and all content originated from publicly available sources. While not subject to institutional ethics review due to the nature of the material, the research followed all principles of methodological transparency, reproducibility, and data protection in line with European GDPR standards.

Results

To explore the latent semantic structure of the news discourse concerning the war in Ukraine, we applied Latent Dirichlet Allocation (LDA) as an unsupervised topic modeling technique. LDA is particularly effective for detecting co-occurring terms and clustering documents around probabilistically inferred themes, without relying on predefined categories. This approach enables the identification of discursive patterns that may not be immediately visible through traditional content analysis, offering a data-driven entry point into the narrative architecture of public communication. The six-topic solution, presented in Table 3, reveals a diversified thematic landscape, with each topic characterised by distinct high-salience terms.

Table 3

Top Terms by Topic (LDA, Cleaned) (to see Table 3, please click here)

Functional words and Romanian stopwords were removed to isolate conceptually meaningful tokens. Across the six topics, keywords such as Ucraina, Rusia, Putin, război, energie, and declarații recur, indicating consistent narrative axes around geopolitical conflict, leadership, and resource politics. The model suggests that Romanian mainstream media did not treat the war as a unitary theme but rather articulated it through multiple, relatively coherent discursive pathways. One of the most salient thematic clusters is Topic 2, dominated by economic and energy-related terms (gaze, energie, milioane, euro, miliarde), pointing to a frame in which the war is discussed through the lens of financial burden and dependency on resources. In contrast, Topic 3 centres on national and identity discourses, including variations of the word Ucraineni and references to Zelenski, implying a focus on victimhood, resilience, and personalised leadership. Topic 6 introduces a geopolitical perspective, with mentions of NATO, UE, and Moldova, reflecting Romania’s regional anxieties and broader Western alignment. Meanwhile, Topic 5 features terms like declarații, forțele, rusești, Putin, suggesting a dominant focus on military developments and political communication. This thematic disaggregation reflects a multiplicity of meaning structures within the Romanian media space, rather than a singular narrative about the war. Each topic appears to sustain its own symbolic logic (military action, economic consequence, international affiliation, or national identity) highlighting the complexity of public discourse in times of crisis. The prominence of declarative verbs (declara, spus, potrivit) across several topics further suggests a reliance on elite sources and formal institutional framing in news construction. To further refine the interpretative lens provided by topic modeling, we applied an unsupervised clustering algorithm (k-means) on the document-topic probability matrix generated by the LDA model. This procedure allowed us to group articles not just by dominant topic, but by their overall thematic composition. The result is a classification of documents into four distinct clusters, each characterized by a different dominant narrative configuration. Each row in Table 4 lists the most representative terms within the dominant topic of a given cluster. The variable mean gamma indicates the average document-topic proportion within the cluster for the associated topic; in other words, how strongly that topic defines the cluster. The beta value corresponds to the per-topic word probability, which reflects the relative importance of each term within that topic. By combining these two dimensions, we can infer not only the dominant theme of each cluster but also the lexical contours that shape its narrative specificity. For instance, Cluster 1 is defined primarily by Topic 6 (mean gamma = 0.39), which includes high-probability terms like ucraina, rusia, nato, and președintele. This suggests a narrative frame focused on institutional actors and geopolitical alignments. Cluster 2, associated with Topic 2 (mean gamma = 0.66), emphasizes economic terminology such as milioane, gaze, energie, and euro, pointing to an economic-infrastructural framing of the war. Cluster 3, dominated by Topic 3 (mean gamma = 0.68), concentrates on ethnonational descriptors (ucraineni, ucrainean, ruse), indicating a narrative centered on ethnic identity and conflict. Finally, Cluster 4, with a high gamma score for Topic 4 (mean gamma = 0.58), mixes affective and personalized language — copii, oameni, război, spune, timp — suggesting a humanitarian and emotional register. This clustering approach thus serves to move beyond surface-level topic identification, enabling a deeper analysis of how certain combinations of topics form distinct narrative regimes in Romanian mainstream media discourse about Ukraine. The computational clusters represent statistically coherent discursive formations rather than semiotic categories in themselves. Accordingly, each cluster was interpreted qualitatively using the analytical framework presented above, focusing on symbolic actors, recurring signs, binary oppositions, affective framing, and cultural myths.

• Cluster 1 – “NATO strengthens its eastern flank as Romania reinforces border security.” The dominant symbolic actors are institutional entities (NATO, the Romanian state, political leaders), constructing the conflict through the language of geopolitical authority rather than individual experience. The recurrent signs (border, alliance, security) produce a binary opposition between stability and external threat. Rather than depicting war as humanitarian suffering, this narrative frames it as a question of institutional legitimacy and collective defence. At the mythic level, the cluster reproduces a narrative of Western protection, in which international institutions become guarantors of moral and political order. Predominantly associated with Topic 6, centers around official rhetoric and geopolitical framing, with high-frequency terms such as ucraina, rusia, nato, declarații and președintele. The high presence of institutional and international terms suggests a narrative built on diplomacy, official discourse, and state-level interpretation of the conflict.

Table 4

Dominant Narratives per Cluster (to see Table 4, please click here)

• Cluster 2 – “Energy prices continue to rise as the war disrupts European gas supplies.” Economic indicators function as symbolic signs of insecurity rather than merely financial variables. Terms such as energy, gas, prices, and billions transform geopolitical conflict into everyday vulnerability. The dominant opposition contrasts economic stability with uncertainty, constructing war as a diffuse threat extending beyond the battlefield. The underlying myth is that of collective sacrifice, where economic hardship becomes part of a broader narrative of European resilience. Driven by Topic 2, is largely economic in nature, structured around terms such as milioane, gaze, euro, energie and miliarde. This suggests a media discourse oriented toward the economic repercussions of the war—especially energy dependence, inflation, and international financial flows related to the conflict.
• Cluster 3, “President Zelensky visits liberated Ukrainian town.” This cluster personalizes the conflict through symbolic leadership. Zelensky functions simultaneously as a political actor and as a moral symbol of resistance. References to liberated territory and civilian populations reinforce the opposition between victim and aggressor, while emotional language encourages identification rather than detached observation. The dominant cultural myth is that of heroic resistance, where individual leadership embodies national resilience; Dominated by Topic 3, brings together a more emotional and personal narrative, focusing on descriptors like ucraineană, zelenski, potrivit, spus and kiev. This cluster reflects storytelling centered on humanization, symbolic leadership, and testimonials, often linked to wartime suffering and heroism.
• Cluster 4, “Children remain trapped as fighting continues.” Civilian actors become the primary symbolic focus, replacing military institutions. Children function as universal signs of innocence, transforming the conflict into a humanitarian tragedy. The recurring opposition between innocence and violence constructs moral clarity while encouraging emotional engagement. Rather than emphasizing military strategy, this narrative produces what Alexander (2004) conceptualizes as cultural trauma by framing suffering as a collective moral experience. Linked to Topic 4, includes lexical markers of temporal continuity (ani, timp, acum), alongside phrases with explanatory or argumentative roles (s-a spus, putin, copii, oameni). The narrative here seems focused on contextualization—retrospective reflections, moral justifications, and broader cultural implications of the war.

The clusters thus function as meta-topics—higher-order groupings that reflect how the media platform aggregates and reframes the war discourse into recognizable cognitive schemas. These findings allow us to map the media grammar of meaning as it evolves over time and to relate symbolic and emotional cues to macro-discursive structures such as economic fear, moral outrage, or political mobilization.

Finally, to investigate the affective dimension of media representations, a sentiment analysis was conducted using an extended sentiment lexicon applied to the titles of all 73,000 articles in the sample. To explore the affective structure of media discourse on the war in Ukraine, we conducted a sentiment analysis using the NRC Emotion Lexicon, a widely validated dictionary-based model developed by the National Research Council Canada (Mohammad & Turney, 2013). The model categorizes words into eight basic emotions — anger, fear, anticipation, trust, surprise, sadness, joy, and disgust — as well as into positive and negative sentiment categories. The method employed calculates a sentiment score for each article title, ranging from -5 (extremely negative) to +5 (extremely positive), which was then aggregated monthly to observe the temporal evolution of emotional tone in news coverage. Because the NRC Emotion Lexicon is an English-language resource, texts used for the emotion analysis were automatically translated into English prior to lexicon matching. Translation was performed to enable dictionary-based emotion detection, while acknowledging that lexical translation may affect the interpretation of emotionally nuanced expressions. Consequently, the results are interpreted as relative emotional tendencies across clusters rather than as precise measurements of emotional intensity. Figure 1 illustrates the evolution of average sentiment scores across the analysied period. Overall, the trend reveals a predominantly neutral or mildly negative framing, with occasional downward spikes indicating emotional intensification. These sentiment valleys often correspond with key geopolitical or humanitarian events in the war, such as major offensives, massacres, or energy crises. The lowest sentiment scores occurred in March 2022 and October 2022, which aligns with the Russian siege of Mariupol and missile attacks on Ukrainian infrastructure, respectively. This analysis highlights the capacity of Romanian mainstream media to modulate affective registers in response to unfolding conflict dynamics. Rather than a static emotional frame, sentiment varies in alignment with the perceived gravity of events, pointing to a form of contingent affective mediation. While the titles do not show high polarity extremes, their cumulative fluctuation across months illustrates a dynamic engagement with the emotional salience of war narratives. This layer of analysis complements the previous topic modeling and clustering outputs by offering an interpretive bridge between lexical-semantic content and affective resonance. It suggests that symbolic trauma, far from being uniform or ubiquitous, is selectively activated in Romanian media according to contextual triggers.

Figure 1

Distribution of Sentiment Categories Across News Articles on Ukraine (to see Figure 1, please click here)

To further deepen the interpretive value of the thematic clusters, the results, presented in Figure 2 (below), show marked differences in the emotional tonality associated with each cluster. Cluster 1 (which predominantly frames the war through military and geopolitical strategy) is heavily saturated with fear, trust, and anticipation—emotions associated with uncertainty and institutional reliance. Cluster 2, which includes human stories and accounts of suffering, is dominated by sadness, disgust, and anger, reflecting its strong humanitarian dimension. Cluster 3, focused on international reactions and EU/NATO dynamics, features high levels of trust and positive sentiment, possibly reflecting political optimism or institutional legitimacy. Finally, Cluster 4, which centers on propaganda, conspiracy, and polarizing rhetoric, scores high in anger, disgust, and negative sentiment, consistent with its disinformation-laden and emotionally manipulative nature. These findings confirm that each narrative environment (cluster) is not only topically distinct but also emotionally coded in ways that may influence audience perception, political stance, or engagement patterns.

Figure 2

Comparative Sentiment Analysis Across Clusters (to see Figure 2, please click here)

Discussions

Thematic Structure

The topic modeling analysis applied to the corpus has revealed six dominant thematic dimensions in the Romanian digital press discourse on the war in Ukraine. These topics – ranging from geopolitical diplomacy to military escalation, economic repercussions, humanitarian crisis, domestic political reactions, and symbolic-religious frames – offer a multifaceted representation of the conflict. The Latent Dirichlet Allocation (LDA) method, widely used in computational political communication studies (Blei et al., 2003; Grimmer & Stewart, 2013), enabled the extraction of word co-occurrence patterns that consistently grouped around interpretable topics, highlighting the underlying discursive architecture of the corpus. This structure mirrors previous findings in the computational study of conflict narratives. For instance, in their work on media representations of war, Hamborg et al. (2019) emphasize the tendency of news outlets to alternate between strategic, economic, and humanitarian framings depending on geopolitical developments. In our analysis, the prevalence of strategic and diplomatic language in Topics 1 and 2 echoes the early months of the war, when Romania’s media largely framed the invasion through NATO, EU, and international law lenses. By contrast, Topics 4 and 5 bring forth a more emotionally charged, symbolic vocabulary—such as “martyr,” “refugees,” or “faith”—which gradually gained visibility, suggesting an evolution from geopolitics to affective solidarities. While the LDA model offers a valuable heuristic for identifying these latent themes, it is essential to emphasize its limitations. As observed by DiMaggio et al. (2013), topic models may miss subtleties of rhetorical tone or symbolic inversion, especially in discursive contexts where irony, polarization, or populist stylization are prevalent. As Edelman (1988) posits, the spectacle functions not to inform but to evoke, anchoring public sentiment in emotive storylines rather than analytical complexity. Therefore, the thematic map produced here serves not as a definitive categorization but as a basis for further qualitative interpretation and cluster-based exploration. This thematic framing transforms distant suffering into a local symbolic resource, reinforcing both national identity and political alignment, while also deflecting attention from the intricacies of international relations. The thematic clustering of articles reveals recurring symbolic oppositions (heroism vs. helplessness, East vs. West, resilience vs. fragility) that echo global democratic anxieties. These narratives do not merely report war—they cultivate shared affects that substitute for direct civic engagement, reflecting how publics experience global crises as emotionally charged, mediated events (Papacharissi 2021).

Narrative Clusters (K-means Clustering)

As William Merrin (2020) argues, digital war is not simply an extension of military operations into new media spaces, but a radical transformation of the symbolic and affective logics through which war is perceived, narrated, and experienced. Digital platforms do not merely report on conflict; they structure it as a performative event governed by visibility, virality, and symbolic overcoding. This is particularly evident in the Ukrainian conflict, where the battlefield is paralleled by a mediatized front of spectacle and moral polarity. Merrin points out that “war is now simultaneously a kinetic, informational, and symbolic event” (p. 5), where meaning is co-produced by media, publics, and algorithmic infrastructures. The analysis of narrative clusters through k-means clustering reveals a consistent symbolic differentiation in how the war in Ukraine was framed across Romanian mainstream media. The six resulting clusters are not only linguistically coherent but also affectively distinct, each organizing specific semiotic fields: the military-political battlefield, victimization and trauma, diplomacy and statecraft, NATO/EU geopolitics, Russian aggression, and symbolic national responses. Such clustering corresponds to the broader theoretical understanding of digital discourse as fragmented, affect-laden, and ideologically embedded (Papacharissi, 2015; Alexander, 2004).

The spectacle of suffering in digital environments operates as a paradox: it creates immediacy while simultaneously distancing the viewer. As Chouliaraki (2006) observes, contemporary media often mediate suffering through “an aesthetics of distance,” turning pain into consumable images and framing it within moral discourses of action, guilt, or indifference. This logic is particularly evident in the Romanian digital discourse on the war in Ukraine, where the repetition of graphic imagery or symbolic representations of victimhood (e.g., children, ruins, flags) contributes not only to affective mobilization but also to the normalization of atrocity.

The clustering model was not designed merely to condense large data into readable categories but to expose the latent thematic divisions structuring public interpretation. Each cluster contains dominant tokens and phrases that suggest different anchoring points for public emotion and political resonance. One way to interpret this thematic clustering is through the lens of cultural trauma theory. As Alexander et al. (2004) argue, collective suffering becomes symbolically meaningful when it is narratively framed as a rupture in the moral order and inscribed into broader civilizational or national identities. In our analysis, Romanian media representations of the war in Ukraine often follow this logic—portraying the conflict not merely as geopolitical aggression, but as a civilizational collapse, a betrayal of shared values, or even a spiritual trial. These framings contribute to the transformation of suffering into symbolic trauma, embedded within narratives of European destiny or moral resistance. In contrast, the cluster centered on NATO/EU and regional geopolitics reflects a technocratic-discursive frame, more closely aligned with rational-institutionalist reporting (Entman, 2003). Furthermore, the emergence of these clusters supports the theoretical model of affective publics (Papacharissi, 2015), in which algorithmically mediated attention flows create semantic enclosures. These narrative clusters do not exist in isolation but compete for dominance within the symbolic field, offering different forms of legitimacy to political action or inaction. This also aligns with post-structuralist views on discourse formation, whereby meaning emerges not through coherence, but through opposition and repetition (Laclau 2005, Mouffe 2016). This suggests that the digital mediation of war discourse may reinforce a passive spectatorship, structured around visual familiarity and symbolic ritualization rather than critical reflection or agency. Building on Young’s (2023) account of how identity-driven cognition and emotionally appealing media content shape our receptiveness to misinformation, the symbolic framing of the Ukraine war in Romanian digital discourse can be seen as a process of affective polarization. Rather than purely informing audiences, much of the media coverage functions to reaffirm existing identity narratives, reinforcing a sense of moral superiority, victimhood, or ideological alignment. As Young explains, individuals are more likely to accept simplified, mythologized accounts of events when these narratives align with their social identities and emotional predispositions.

Sentiment Analysis

While this study does not examine user-generated reactions or social media activity, Romanian mainstream media themselves can be understood as affective infrastructures—sites where sentiment is curated, repeated, and ritualised. In this sense, the textual architecture of news articles—headlines, metaphors, and narrative tropes—functions as a conduit for affective attunement, inviting specific emotional responses even in the absence of interactivity. By applying Papacharissi’s (2015) insight to media content as stimulus, not just platform, we reframe mainstream journalism as an active agent in shaping emotional publics: publics not necessarily expressive, but affectively conditioned. The sentiment analysis provided a quantitative and affective reading of the corpus, revealing strong polarity variation across both topics and time. Using the NRC sentiment model, which assigns weighted emotional valence to individual words based on a lexicon of affective meanings, we computed the mean sentiment score per article and then aggregated these values monthly and by cluster. The resulting scale ranged from –5 (strongly negative) to +5 (strongly positive), allowing us to trace affective intensities associated with different discursive frames. As Sara Ahmed (2014) argues, emotions are not merely personal states but performative acts that circulate between bodies, objects, and signs, shaping social bonds and boundaries. In digital contexts, such as those observed in the Romanian media discourse on Ukraine, emotions like fear, grief, or pride do not remain confined to individual experience but become embedded in collective affective economies. Findings show that clusters focused on human suffering (e.g., civilian victims, humanitarian crisis) were significantly more negative than those centered on institutional responses or strategic diplomacy. This result supports existing literature suggesting that negative affective framing increases salience and emotional engagement in news consumption (Hameleers, 2020). At the same time, sentiment analysis revealed not just differences between clusters, but also within them, showing variation across months. This temporal fluctuation is consistent with the theory of issue-attention cycles (Downs, 1972) and suggests that public emotional engagement may be event-driven rather than structurally stable.

Interestingly, the comparative sentiment per cluster indicated that the NATO-EU frame retained a mostly neutral tone throughout the period, which may reflect institutional communication strategies aimed at minimizing panic or escalation. By contrast, spikes in negative sentiment coincided with key events such as the bombing of Mariupol or the discovery of mass graves in Bucha, moments that acted as affective tipping points (Redlawsk et al., 2010) in public discourse.

This dual-layered affective mapping—across topic and time—provides empirical grounding for theoretical perspectives on affective polarization (Iyengar et al., 2019) and the emotionalization of political discourse in times of crisis (Wahl-Jorgensen, 2019). It also opens the path for future predictive modeling, by examining whether sentiment can be anticipated based on topic or calendar variables—a possibility explored through regression modeling in the following section.

The symbolic and affective dynamics of the Ukraine war within Romania’s digital discourse align with broader transformations in the global media ecology. As Hemming and Sørensen (2022) argue in Global Currents, contemporary media do not merely transmit information but structure affective participation across borders, especially during crises. The concept of “media ecologies of crisis” describes how platforms shape perception and emotional response, amplifying certain narratives while occluding others. Within this framework, the Romanian digital sphere emerges not as a peripheral consumer but as an active node in the transnational circulation of symbolic frames. The mythologisation of Ukrainian resistance and the visual scripting of suffering reflect precisely what the authors describe as “a global choreography of mediated solidarity and antagonism” (p. 77).

Conclusions

This study has shown that war discourse, even when mediated through ostensibly neutral journalistic forms, is deeply structured by symbolic narratives and affective framings. By applying a combination of topic modeling, k-means clustering, and sentiment analysis on over 73,000 mainstream media articles from Romania, we revealed not only the thematic scaffolding of war coverage but also the emotional intensities and moral evaluations that accompany it. The topic modeling suggested a consistent presence of six dominant thematic fields, each articulating different dimensions of the conflict—from geopolitics and diplomacy to violence, trauma, and moral blame. The subsequent clustering of articles reinforced the idea that these topics are not discrete but embedded in larger narrative structures that carry symbolic and ideological weight. These clusters—centered around trauma, strategic governance, international alliances, and aggression—demonstrated how the media organize meaning around war, creating affective entry points for public interpretation and political positioning. Sentiment analysis confirmed that not all topics are affectively equivalent. Clusters related to human loss and civilian suffering produced consistently negative sentiment scores, highlighting the role of emotional saturation in driving narrative visibility. Conversely, institutionally framed articles (on NATO, diplomacy, or energy policy) displayed neutral or buffered sentiment, signaling a discursive attempt at depoliticization. This dichotomy reflects broader patterns of affective governance, where emotion becomes both a mode of engagement and a tool of political control (Redlawsk et al., 2010; Papacharissi, 2015). Together, these results underline the importance of integrating computational methods with critical theory to unpack how crises are represented, mediated, and emotionally coded in the public sphere. In the Romanian case, the war in Ukraine has become not only a geopolitical event but a symbolic stage for narrating trauma, negotiating collective identity, and calibrating national moral hierarchies. In the Romanian digital discourse around the war in Ukraine, symbolic trauma and heroism are not merely responses to real-time events but are embedded in culturally internalized narratives that frame suffering, loss, and identity through historically resonant scripts. As Assmann (2011) argues, cultural memory provides “a fixed point, a reference to which the present can orient itself,” especially in periods of crisis (p. 36). This framework helps explain how national symbols and traumatic imagery are reactivated and resemanticized in digital spaces, giving coherence to fragmented experiences and enabling a form of mediated resilience. In the same time, Routledge Handbook of Humanitarian Communication (Chouliaraki et al., 2021) frames humanitarian discourse as a terrain of moral imagination, where suffering is rendered visible not merely to inform but to mobilize publics. Future research should, therefore explore how these discursive formations influence public opinion, policy debates, and the legitimacy of political action—particularly in post-socialist contexts marked by fragile trust in institutions and a volatile media ecosystem.

Competing interests

The authors declare no competing interests.

Ethics Committee Approval

Ethical approval was not required for this study because it is based exclusively on previously published literature.

Acknowledgements

The author gratefully acknowledges MAD Intelligence for providing access to the dataset extracted from Romanian digital media platforms during the studied period. Their technical support in web scraping and data collection was instrumental in enabling the present analysis.

References

1. Ahmed, S. (2014). The cultural politics of emotion (2nd ed.). Edinburgh University Press.
2. Alexander, J. C. (2004). Toward a theory of cultural trauma. In J. C. Alexander, R. Eyerman, B. Giesen, N. J. Smelser, & P. Sztompka (Eds.), Cultural trauma and collective identity (pp. 1–30). University of California Press.
3. Assmann, J. (2011). Cultural memory and early civilization: Writing, remembrance, and political imagination. Cambridge University Press.
4. Barthes, R. (1972). Mythologies (A. Lavers, Trans.). Hill and Wang. (Original work published 1957)
5. Berger, J. (1972). Ways of Seeing. Penguin.
6. Blei, D. M., Ng, A. Y., & Jordan, M. I. (2003). Latent Dirichlet Allocation. Journal of Machine Learning Research, 3, 993–1022.
7. Chouliaraki, L. (2006). The Spectatorship of Suffering. SAGE Publications.
8. Chouliaraki, L., Vestergaard, A., Orain, M., & Andén-Papadopoulos, K. (Eds.). (2021). Routledge Handbook of Humanitarian Communication. Routledge.
9. Danesi, M. (2004). Messages, signs, and meanings: A basic textbook in semiotics and communication theory (3rd ed.). Canadian Scholars’ Press.
10. Danesi, M. (2024). Semiotics with a conscience. Bloomsbury Academic.
11. Downs, A. (1972). Up and down with ecology: The “issue-attention cycle.” The Public Interest, 28, 38–50.
12. Eco, U. (1979). A Theory of Semiotics. Indiana University Press.
13. Edelman, M. (1988). Constructing the political spectacle. University of Chicago Press.
14. Enli, G. (2015). Mediated authenticity: How the media constructs reality. Peter Lang.
15. Entman, R. M. (2003). Projections of power: Framing news, public opinion, and U.S. foreign policy. University of Chicago Press.
16. Grimmer, J., & Stewart, B. M. (2013). Text as Data: The Promise and Pitfalls of Automatic Content Analysis Methods for Political Texts. Political Analysis, 21(3), 267–297. https://doi.org/10.1093/pan/mps028
17. DiMaggio, P., Nag, M., & Blei, D. (2013). Exploiting affinities between topic modeling and the sociological perspective on culture: Application to newspaper coverage of U.S. government arts funding. Poetics, 41(6), 570–606. https://doi.org/10.1016/j.poetic.2013.08.004
18. Hameleers, M. (2020). Populist Disinformation: Exploring Intersections between Online Populism and Disinformation in the US and the Netherlands. Politics and Governance, 8(1), 146–157. https://doi.org/10.17645/pag.v8i1.2478
19. Hamborg, F., Meuschke, N., & Gipp, B. (2019). Automated identification of media bias in news articles: an interdisciplinary literature review. International Journal on Digital Libraries, 20(4), 391–415. https://doi.org/10.1007/s00799-018-0261-y
20. Hemming, S., & Sørensen, I. E. (Eds.). (2022). Global currents: Media and technology now. Intellect Books.
21. Iyengar, S., Lelkes, Y., Levendusky, M., Malhotra, N., & Westwood, S. J. (2019). The origins and consequences of affective polarization in the United States. Annual Review of Political Science, 22, 129–146. https://doi.org/10.1146/annurev-polisci-051117-073034
22. Jacob, F. (Ed.). (2021). War and semiotics: Signs, communication systems, and the preparation, legitimization, and commemoration of collective mass violence. Routledge.
23. Laclau, E. (2005). On populist reason. Verso.
24. McLuhan, M. (2001). Understanding media: The extensions of man. Routledge. (Original work published 1964).
25. Merrin, W. (2020). Digital war: A critical introduction. Routledge.
26. Moffitt, B. (2016). The global rise of populism: Performance, political style, and representation. Stanford University Press.
27. Mohammad, S. M., & Turney, P. D. (2013). Crowdsourcing a word–emotion association lexicon. Computational Intelligence, 29(3), 436–465. https://doi.org/10.1111/j.1467-8640.2012.00460.x
28. Papacharissi, Z. (2015). Affective publics: Sentiment, technology, and politics. Oxford University Press.
29. Redlawsk, D. P., Civettini, A. J. W., & Emmerson, K. M. (2010). The affective tipping point: Do motivated reasoners ever “get it”? Political Psychology, 31(4), 563–593. https://doi.org/10.1111/j.1467-9221.2010.00772.x
30. Young, D. G. (2023). Wrong: How Media, Politics, and Identity Drive Our Appetite for Misinformation. JHU Press.
31. Wahl-Jorgensen, K. (2020). Emotions, media and politics. Polity Press.