Bangaranga on Facebook during Eurovision week: The Outrage That Wasn't or How Bulgaria's Eurovision Moment Defied Its Own Critics

This brief analytical report examines the most visible Bulgarian-language Facebook posts about DARA’s Eurovision participation and victory, with particular attention to how the public attention around “Bangaranga” was used during the final week to circulate support, criticism, and broader moralized or propagandistic claims.

The analysis combines three datasets downloaded on 17 May 2026 (in the morning after the final) with the date range set to the last 7 days: a manually scraped public-content dataset for the search string “бангаранга”; a downloadable Meta Content Library dataset for the same search string; and a Boolean subset for “бангаранга AND сатанизъм” since both datasets demonstrated the existence of such a narrative.

In the Meta Content Library interface, the search string “бангаранга” (the name of the song borrowed from Jamaican Patois) returned an estimated 5,700 total results in public content, while the downloadable subset compressed this universe to around 1,000 estimated results because downloadable public data are restricted to posts from Pages with 15,000 or more likes or followers and from profiles that are verified or have 25,000 or more followers. The downloadable dataset is therefore not equivalent to the full public-content result set but a threshold-based subset of it, and for Bulgarian-language research these eligibility thresholds are arguably too high because they exclude a substantial share of locally relevant visible discourse.

Data and method

The manually scraped top-100 sample contains 100 posts from 2026-05-10 to 2026-05-17 and 76 distinct producers. The Boolean subset contains 41 posts from 2026-05-14 to 2026-05-17 and 40 distinct producers. The downloadable top-100 sample presented in the first part of this report contains 100 posts from 2026-05-10 to 2026-05-17 and 67 distinct producers. (See Appendix 2 for a brief analytical report presenting the main metrics and findings of the full dataset). The first two are recreated as simplified research tables with six variables: views, reactions, producer, date, post text, and content type. The downloadable subet retains richer metadata, including timestamps, owner identifiers, link metadata, and detailed engagement counts.

Figure 1. Daily posting volume across the three datasets.

To compare the two top-100 samples, posts were matched on normalized text after lowercasing, whitespace normalization, and removal of trivial string noise. The round-2 coding scheme distinguishes six stance categories: supportive direct, supportive indirect, critical direct, critical indirect, neutral informational, and context-dependent. A second variable records whether a post engages a moral-panic frame and, if so, whether it endorses or contests that frame.

Dataset profile

Figure 2. Harmonized content types in the two top-100 samples.

Important: An LLM-assisted exploratory coding procedure involving numerous iterations and comparisons was utiliesed rather than a full manual annotation protocol with intercoder reliability testing.

Figure 3. Overlap between the two top-100 samples, matched on normalized post text.

Main findings

Matching on normalized post text identifies 32 shared posts, 63 posts unique to the scraped sample, and 63 posts unique to the downloadable sample. The manually scraped set is more producer-diverse, with 76 distinct producers versus 67 in the downloadable export. The scraped sample has median engagement of 103 100 views and 2 300 reactions, compared with 48 404 views and 958 reactions in the downloadable sample.

In the scraped top-100 sample, supportive posts outnumber critical ones by 41 to 25 (ratio 1.64:1). In the downloadable top-100 sample, the balance is 42 to 18 (ratio 2.33:1). In the Boolean subset, the balance reverses to 10 positive and 27 negative posts (ratio 0.37:1). A substantial share of visible posts in the two broad top-100 samples are best described as neutral informational posts rather than clear endorsement or condemnation.

The Boolean subset differs because it concentrates posts that explicitly invoke or contest a moral-panic frame. Because this dataset is curated around the controversy, it contains a high density of explicit accusations of "satanism," "occultism," and "moral decay." Posts here often describe the performance as a "national shame" or "demonic obsession," frequently linking the song to broader conspiracy theories about "spiritual downfall."

The positive posts in it are often lengthy, sophisticated counter-arguments defending the artist against the accusations. Users in this dataset defend the song by providing historical context (comparing the panic to past reactions to Rock 'n' Roll), critiquing the "narrow-mindedness" of the accusers, and explaining the actual meaning of the word "Bangaranga" (Jamaican slang for chaos/riot) to debunk the "satanic" claims.

Figure 4. Stance categories in the scraped and downloadable top-100 samples.

The moral-panic dimension is the strongest point of divergence between the Boolean subset and the two broad top-100 samples. Posts in the Boolean subset frequently use language around satanism, the devil, occultism, immorality, vulgarity, youth corruption, and cultural decay. Yet the presence of this vocabulary does not automatically signal endorsement.

Figure 5. Stance categories in the Boolean subset retrieved with “бангаранга AND сатанизъм”.

A visible counter-current rejects the accusation, mocks it as hysteria, and reframes the controversy as a backlash against youth culture or non-traditional aesthetics. Taken together, the datasets show that the most visible Bulgarian-language Facebook posts about “Bangaranga” during the Eurovision final week combined support, critique, and moralized controversy. 

Figure 6. Moral-panic framing across datasets: no frame, endorsement, contestation, and mention.

Producer concentration

Producer concentration differs across the datasets. The manually scraped top-100 sample is more dispersed across producers, while the downloadable sample is somewhat more concentrated. The Boolean subset is smaller and more fragmented, with many producers appearing only once. The ranked producer charts below list the twelve most frequent producers in each dataset.

Appendix 1. Coding dictionary

The coding dictionary below summarizes the final variables used. The stance variable captures the direction and explicitness of evaluation. The moral-panic variable captures whether the post invokes a frame that links the performance to satanism, immorality, vulgarity, corruption, degradation, or related symbolic threat.

Appendix 2. A brief summary of the downloadable dataset

The analysis of the downloadable public dataset of 911 posts reveals that morally and religiously inflected oppositional content, while substantively present and widely circulated, failed to produce a commensurate affective signal in platform-mediated emotional expression. The findings contribute to ongoing discussions on the reliability of textual content analysis as a proxy for public sentiment, and on the structural conditions under which moral panic framing fails to achieve emotional contagion on social media platforms.

The Bulgarian Facebook response to DARA's Eurovision qualification is characterised, in aggregate, by positive affective dominance — a finding that holds robustly across reaction type distributions and is not materially disrupted by the circulation of critical content, however rhetorically elaborate or widely viewed. The case offers a useful empirical reference point for discussions of emotional contagion, moral panic dynamics, and the relationship between textual and affective measures of public opinion in digital environments.

1. Data and Method

The dataset was retrieved via Meta's Content Library transparency tool and comprises 911 Bulgarian-language public Facebook posts published mainly between 12 and 15 May 2026. Post-level metadata includes content type, authorship attributes (account name, type, and identifier), creation and modification timestamps, attached multimedia, and platform-generated engagement statistics: like, love, haha, angry, care, wow, and sad reaction counts; comment and share counts; and cumulative view figures. Three posts were identified as duplicates on the basis of shared post identifiers, yielding 908 analytically unique records.

Content types are distributed as follows: photos (523 posts, 57.4%), videos (111, 12.2%), links (100, 11.0%), albums (74, 8.1%), text-only status updates (61, 6.7%), reshares (37, 4.1%), and unclassified entries (5, 0.5%). All posts are coded lang: bg; none carry branded content flags.

2. Findings

2.1 Aggregate Engagement

The corpus accumulated a total of 12,736,860 views across the four-day observation window, reflecting the high salience of the Eurovision qualification event within the Bulgarian digital public sphere. 

Table 1. Aggregate Reaction Counts

The distribution is heavily skewed toward positive valence, with Like and Love reactions jointly accounting for 96.7% of all measured affective expression.

2.2 Textual Content: Presence and Character of Oppositional Discourse

Qualitative inspection of post text reveals a substantive minority of content framing DARA's performance and song through explicitly critical, morally inflected, or religiously grounded registers. These posts variously characterise "Бангаранга" as aesthetically deficient, spiritually harmful, or ideologically objectionable. Recurring argumentative strategies include: invocation of Orthodox Christian moral frameworks; pseudo-etymological analysis attributing occult significance to the song's title through reference to West African animist traditions and Japanese popular media; and rhetorical alignment of the song with broader narratives of cultural decline and value erosion.

Several such posts are of considerable length and textual elaboration, and were published by accounts with non-trivial follower bases. One post, authored by a public figure (content creator with a YouTube channel channel that has just over 1,200 subscribers and a Facebook page with over 18,000 followers), accumulated approximately 778,000 views — a reach comparable to high-performing supportive content — while generating 7,771 reactions with a notably elevated haha share, and only 180 angry reactions.

2.3 The Affective Disjunction

The central analytical observation is the absence of a measurable correspondence between the volume and rhetorical intensity of oppositional textual content and the aggregate emotional response of the audience as expressed through platform reaction mechanisms. Despite the circulation of morally charged critique at meaningful scale, the corpus-wide Angry reaction count — the platform mechanism most directly associated with negative affective arousal — constitutes 0.3% of total reactions.

This disjunction cannot be adequately explained by the mere numerical dominance of supportive posts. Even accounting for the asymmetric distribution of high-reach accounts (notably, Lili Ivanova's endorsement post achieved 1.14 million views and 44,936 reactions), the oppositional content's view-to-angry-reaction conversion rate is strikingly low. The data suggest that exposure to critical framing did not reliably produce, or did not reliably translate into, the expression of negative affect through available platform affordances.

3. Discussion

These findings are consistent with a growing body of evidence suggesting that textual outrage and emotionally mobilised audiences constitute distinct — and only partially overlapping — phenomena. Platform reaction mechanisms, operating at low cognitive cost and high temporal immediacy relative to comment production, may function as a more ecologically valid measure of ambient sentiment than textual discourse, which is subject to motivated reasoning, selection effects among producers, and amplification dynamics independent of audience endorsement.

The structural conditions of this case are also relevant. The Eurovision context provided a nationally salient, temporally bounded focal event around which pre-existing affective dispositions — pride, collective identity, anticipatory excitement — were already organised. Under such conditions, oppositional moral framing faces a high threshold for emotional reframing, particularly when the dominant narrative is reinforced by high-credibility, high-reach accounts within the same information environment.

The near-total absence of Care reactions (0 recorded) is a secondary but notable finding, suggesting that neither the empathetic register typically associated with personal distress nor the solidarity frame was activated at any measurable level in this corpus.

4. Limitations

The dataset is restricted to public posts retrievable via Meta's Content Library for researchers and does not capture private group content, direct messaging, or cross-platform discourse. Reaction counts are platform-generated and reflect the cumulative behaviour of all users who encountered a given post, not a representative sample. The observation window limits longitudinal inference. Content-type classification relies on platform-assigned labels and may not fully capture hybrid post formats.

Тhe analysis, visualization, and summarization of data presented in this report were assisted by large language models – ChatGPT 5.4 Thinking and Claude 4.6 Sonnet Thinking using a methodology developed by the researcher.

Appendix 3. Google Trends top 50 queries worldwide on May 17, 2026