PaperlayerView exact revision on arXiv

Source-linked AI summary

Affective publics in Arabic YouTube

Lynnette Hui Xian Ng, Craig Douglas Albert, Abdullah Melhem, Ahmed Aleroud, Lance Y. Hunter

arXiv:2609.03269v1cs.CY

TL;DR

The paper asks what emotional register organizes Arabic YouTube’s political affective publics across five MENA countries and how that register varies around US-related content. It analyzes 67,725 comments with a unified sentiment-and-emotion pipeline, finding a shared regional register marked by pervasive negativity, structurally similar emotion profiles, and geopolitically conditioned variation.

  • Problem

    Existing computational research has largely centered on Twitter/X, binary sentiment, or single-country studies, leaving the cross-national emotional organization of Arabic digital publics insufficiently examined.

  • Method

    The study analyzes 67,725 Arabic YouTube comments from five country-oriented corpora using unified sentiment classification, 28-category emotion detection, and analysis of US-related content.

  • Results

    Sentiment is predominantly negative across all five corpora, emotion profiles are structurally similar, Iraq is distinguished by grief, and US-related sentiment reverses direction across countries.

  • Takeaways & Limitations

    Arabic YouTube’s affective public is shared across borders but historically layered and geographically conditioned, with implications for MENA public diplomacy.

  • Takeaways & Limitations

    Emotion results should be interpreted as relative cross-country comparisons because machine translation, model biases, absent Arabic-native benchmarks, self-selection, and keyword-based country assignment constrain absolute or population-level interpretation.

Abstract

from arXiv · show

What is the emotional register of Arabic YouTube's affective publics? To investigate this, we analyzed 67,725 YouTube comments collected around socio-political topics associated with Yemen, Saudi Arabia, Iraq, Jordan, and Syria using a unified sentiment-and-emotion pipeline. Our results profile a single regional affective public rather than five separate national ones. Sentiment is overwhelmingly negative across all five country-oriented corpora, and the country-level emotion profiles are structurally similar. This shared register still accommodates some regional variations: discourse is organized around country-level political actors and cross-border historical trauma figures, and grief singularizes Iraq from the other countries. The differences in emotional register also tracks lived political causes rather than fixed categories, which we observe from patterns of the valence of US-related content, that tracks the presence or absence of direct US military engagement. Our work shows that the emotional register of Arabic YouTube's affective publics is a shared one that is historically layered and geographically conditioned, which has implications for public diplomacy in the MENA region.

1 Introduction

The study examines the emotional register of Arabic YouTube political discourse across five MENA countries, addressing gaps in scale, platform coverage, and fine-grained emotion analysis. It uses a unified pipeline to test cross-national patterns and shifts associated with US-related content.

  • Prior computational work has largely focused on Twitter/X, binary sentiment, or single-country designs rather than fine-grained cross-national emotional patterns.
  • The study maps sentiment and emotion across 67,725 Arabic YouTube comments from Yemen, Saudi Arabia, Iraq, Jordan, and Syria.It applies a unified computational pipeline to five country-oriented corpora.
  • The research asks how sentiment differs across countries, which emotions underlie it, and how the register changes around US-related content.
  • The study contributes a cross-national emotional mapping, argues that emotion provides more diagnostic information than aggregate sentiment, and examines geopolitically differentiated shifts linked to US-related content.

2 Background and Related Work

MENA digital publics operate under political and media conditions that differ from Western contexts, while Arabic computational research remains concentrated on sentiment, Twitter/X, and narrow national designs. This study extends that work through fine-grained emotion analysis on YouTube across five countries.

  • MENA digital publics are shaped by authoritarian constraint, information warfare, propaganda, instability, and high platform penetration relative to formal democratic participation.
  • Prior MENA research finds predominant negativity, links negative sentiment to US military presence and Israel-Palestine conflict dynamics, and reports context-dependent anti-Americanism.
  • Arabic sentiment research has advanced toward transformer models but remains largely focused on binary classification within single countries, while emotion analysis is less developed.
  • The study addresses these gaps by combining 28-category emotion detection with binary sentiment, moving from Twitter/X to YouTube, and comparing five national corpora.

3 Materials and Methods

The study constructs five country-oriented Arabic YouTube comment corpora and analyzes them with sentiment, emotion, named-entity, and US-mention procedures. It normalizes emotion scores for comment length and applies statistical tests to compare countries and US-related comments.

  • 3.1 Data collection: The dataset contains 67,725 comments collected with country-specific Arabic socio-political keywords, representing discourse contexts rather than verified commenter nationalities.Search themes include Gaza, domestic conflicts, armed conflict, and US involvement.
  • 3.1 Data collection: Table 1 organizes search keywords into thematic domains by country and points to the full Arabic keyword list in Supporting Information.
  • 3.2 Country contexts: The five country contexts differ in history, platform penetration, and relationships with the United States, providing background for interpreting cross-context results.
  • 3.3 Computational pipeline: Comments pass through sentiment and emotion classification, with sentiment scored by CAMeL-Lab BERT and emotion classified by translated Arabic trigrams using EmoRoBERTa’s 28 categories.
  • 3.3 Computational pipeline: Emotion scores are length-normalized because summed trigram probabilities mechanically increase with comment length, and comments shorter than three tokens are excluded.3,411 of 67,725 comments, or 5.0%, are excluded from the length-normalized analysis.
  • 3.3 Computational pipeline: Named-entity recognition and text matching add a binary indicator for United States mentions in comments or associated video titles.
  • 3.3 Computational pipeline: Country-level comparisons use ANOVA with corrected pairwise Welch tests for sentiment, Mann–Whitney U tests for US-mentioned comments, and cosine similarity for emotion profiles.

4 Results

Across five country-oriented Arabic YouTube corpora, sentiment is overwhelmingly negative, while detailed emotion profiles reveal a shared regional affective grammar with specific national differences. Iraq is distinguished by grief, and US-related content reverses negativity direction across countries.

  • 4.1 Sentiment is predominantly negative and significantly different across countries: Mean negativity ranges from 82.0% in Syria to 84.7% in Iraq, indicating pronounced negativity across all five countries despite modest practical differences.Pairwise effect sizes are small to negligible, with |d| = 0.047–0.132.
  • 4.2 Communal solidarity and singular grief: Caring and admiration dominate all five countries, revealing a communal solidarity register that aggregate sentiment alone would not show.The highest mean caring scores occur in Saudi Arabia (M = 31.6) and Yemen (M = 24.1), while admiration ranks second in every country.
  • 4.2 Communal solidarity and singular grief: Iraq is singularized by grief, with a mean score of 10.03—5.4 times Yemen’s 1.87—and also has the highest anger.Iraq’s profile contrasts with the shared regional structure while retaining high admiration.
  • 4.2 Communal solidarity and singular grief: Country emotion profiles remain highly similar, with every pairwise cosine similarity above 0.927 and Yemen–Syria the closest pair at 0.982.Saudi Arabia–Iraq is the least similar pair at 0.927.
  • 4.3 US-related content produces a directional reversal by country: USA-mentioned comments are less negative in Yemen, Saudi Arabia, and Jordan but more negative in Syria and Iraq, with all comparisons significant at p < 0.001.The reversal is not explained by USA-mention prevalence: Syria and Yemen have similar rates but opposite directions.

5 Discussion

Arabic YouTube discourse forms a shared regional affective public: intense negativity coexists with caring and admiration, while emotional patterns remain historically layered and geographically conditioned. Cross-national convergence accommodates country-specific political actors, Iraq’s elevated grief, and US-related sentiment shifts linked to local geopolitical experience.

  • 5.1 Coexistence of negativity and communal solidarity: Caring and admiration dominate across all five countries alongside mean negativity rates of 82–85%, challenging a simple grievance model of online political negativity.The pattern is consistent across independently collected national corpora and suggests that negative political evaluations coexist with communal solidarity orientations.
  • 5.5 Geopolitical context and the US-sentiment reversal: US-mentioned comments are significantly less negative in Yemen, Saudi Arabia, and Jordan but significantly more negative in Syria and Iraq, corresponding to differences in active US military operations.This directional reversal contrasts with treating hostility to the US as a fixed reaction independent of its actions.
  • 5.3 Cross-national emotional convergence and regional affective structure: Country-level emotion profiles show high cosine similarity of 0.927–0.982, indicating regional emotional structure rather than five wholly separate national publics.The convergence appears across independently collected corpora built around distinct country-specific political contexts.
  • 5.2 Affective publics as an interpretive framework: Negativity and solidarity emotions cluster around recurring political causes, including Gaza, Yemen’s civil-war inheritance, and Iraq’s US-led invasion and aftermath.These causes organize affective attention rather than emotions distributing uniformly across each corpus.
  • 5.3 Cross-national emotional convergence and regional affective structure: The shared emotional template accommodates regional variation, including elevated grief in Iraq and elevated anger in Jordan, without implying emotional uniformity.Iraq’s grief is anchored in historical trauma associated with Saddam Hussein, Tariq Aziz, and Ali Hassan al-Majid.
  • 5.4 Named actors as affective anchors: Named-actor patterns show that affect is organized through layered contemporary and historical political figures, including Yemen’s Houthi and Saleh figures, Jordan’s Palestinian-cause figures, and Saudi Arabia’s Putin reference.These profiles connect present discourse with civil-war inheritance, regional intervention, multigenerational historical frames, and comparative geopolitical judgments.

6 Conclusion

The study finds predominantly negative sentiment alongside communal solidarity emotions across five country-oriented Arabic YouTube corpora. Emotion profiles are regionally shared but retain Iraq-specific grief and geopolitically conditioned differences in US-related sentiment.

  • Sentiment is predominantly negative across all five countries, while caring and admiration are the most frequent emotions.Negative political evaluation co-occurs with communal affective solidarity rather than anger or disgust.
  • Country-level emotion profiles are structurally similar, supporting a shared regional affective structure in Arabic YouTube political discourse.The conclusion interprets the cross-country similarities through the affective publics framework.
  • Iraq differs most from the regional pattern in grief, consistent with its accumulation of mass-violence events and atrocity-linked historical figures.
  • US-related comments are less negative in Yemen, Saudi Arabia, and Jordan but more negative in Syria and Iraq.This reversal corresponds with differences between direct and indirect US military presence.

Declarations

The declarations identify author contributions, report no competing interests, disclose U.S. Office of Naval Research funding, and state that anonymized data are available upon request under YouTube’s Terms of Service.

  • Contributors are credited across conceptualization, methodology, analysis, software, writing, funding acquisition, project administration, validation, and data curation.
  • The authors declare no competing interests and report funding from the U.S. Office of Naval Research.The award number is N000142212549.
  • The anonymized comment dataset and video data can be made available upon request in accordance with YouTube’s Terms of Service.

S1 Search Keywords

The supplementary search-keyword list documents the Arabic terms and English translations used to collect YouTube data across countries and thematic domains.

  • Search terms are organized by country and thematic domain, with Arabic keywords paired with English translations.
  • Researchers conducted the searches during summer 2024 using the youtubesearchpython library.
  • Native-speaker researchers with MENA regional-politics expertise developed the keywords and validated them against pilot results.

S2 Sentiment Scores by Country

The supplementary tables report country-level sentiment comparisons and pairwise similarities between 28-dimensional emotion profiles, documenting both significant sentiment differences and a shared emotional grammar.

  • Negativity differs significantly across countries, with ANOVA F(4, 67,720) = 29.36, p < 0.001.The table reports mean and standard deviation sentiment scores on a 0–100 scale.
  • Positivity also differs significantly across countries, with ANOVA F(4, 67,720) = 40.52, p < 0.001.The table reports mean and standard deviation sentiment scores on a 0–100 scale.
  • All pairwise emotion-profile cosine similarities exceed 0.927, indicating a shared regional affective grammar.Yemen–Syria is the most similar pair, while Saudi Arabia–Iraq is the least similar.

S3 Named Persons by Country

Political leaders and resistance figures are the most frequently mentioned named persons across country-oriented corpora, with cross-country overlap indicating regionally integrated discourse.

  • Political leaders and resistance figures dominate named-person mentions, while overlapping actors across countries reflect the regional integration of Arabic YouTube political discourse.

S4 Comment Length by Country

Comment lengths differ statistically across countries but only modestly, with a common median of 10 tokens; short comments are excluded from normalized emotion analysis, and normalization preserves the headline emotion patterns.

  • The median comment is 10 tokens in every country, while country means differ significantly but only modestly, ranging from 13.2 in Syria to 14.2 in another country.The supplied passage truncates the country label for the 14.2 mean.
  • 3,411 comments, or 5.0%, contain fewer than three tokens and are excluded because they cannot form a trigram.
  • Length normalization leaves the three headline patterns unchanged: caring and admiration lead in all countries, Iraq’s grief elevation persists, and cross-country profiles remain highly similar.Iraq’s grief ratio increases from 4.2× without normalization to 5.4× with normalization; the supplied passage truncates the normalized similarity range.
Loading 2609.03269v1…