---
title: "Sentiment Classification of Gaza War — Mustafa; 2026"
description: "The paper asks whether automatic sentiment tools agree when they read the same war coverage."
authors: ["Amr Eleraqi; Hager H. Mustafa;", "Abdul Hadi N. Ahmed"]
year: 2026
venue: "arXiv:2604.08566 · SSRN (2026)"
source: "https://arxiv.org/abs/2604.08566"
doi: "10.2139/ssrn.6332158"
arxiv: "2604.08566"
licence: "CC BY-NC-SA 4.0 (arXiv)"
licence_url: "http://creativecommons.org/licenses/by-nc-sa/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/eleraqi-2026-sentiment-classification-gaza-war-headlines/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/eleraqi-2026-sentiment-classification-gaza-war-headlines/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/9/) / Sentiment Classification of Gaza War Headlines: A Comparative Analysis of Large Language Models and Arabic Fine-Tuned BERT Models

# Sentiment Classification of Gaza War Headlines: A Comparative Analysis of Large Language Models and Arabic Fine-Tuned BERT Models

**Authors:** Amr Eleraqi; Hager H. Mustafa;, Abdul Hadi N. Ahmed · **Year:** 2026 · **Venue:** arXiv:2604.08566 · SSRN (2026) · **Licence:** [CC BY-NC-SA 4.0 (arXiv)](http://creativecommons.org/licenses/by-nc-sa/4.0/)

**Links:** [Publisher's page](https://doi.org/10.2139/ssrn.6332158) · [PDF](https://arxiv.org/pdf/2604.08566) · [DOI 10.2139/ssrn.6332158](https://doi.org/10.2139/ssrn.6332158) · [arXiv 2604.08566](https://arxiv.org/abs/2604.08566) · [OpenAlex](https://openalex.org/W7137919351)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
The paper asks whether automatic sentiment tools agree when they read the same war coverage. The authors take 10,990 Arabic headlines about the 2023 Gaza War (Eleraqi 2026) and label them with three large language models (GPT-4.1, Qwen2.5-7B, LLaMA-3.1-8B) and six Arabic BERT models fine-tuned for sentiment. They deliberately use no human gold standard and treat each model's output as an interpretation rather than as right or wrong. They compare the label distributions with three measures. Entropy is computed in Shannon's sense, and distances between models are computed in the Jensen–Shannon sense. The third is a Variance Score, which measures how far a model departs from the pooled behaviour of all models. For the LLMs, they also break sentiment down by narrative frame (humanitarian, legal, security and others) and analyse GPT-4.1's written rationales. The divergence turns out to be large and systematic. BERT models, MARBERT above all, default to neutral, while LLMs push towards negative, and LLaMA-3.1-8B labels almost everything negative. Only GPT-4.1 clearly shifts its judgments with the frame. The authors' point is that choosing a model means choosing an interpretive lens, so sentiment scores should not be treated as neutral, interchangeable measures of media tone.

## Key points
- Corpus: 10,990 Arabic Gaza War headlines. The models compared are GPT-4.1, Qwen2.5-7B and LLaMA-3.1-8B against six fine-tuned Arabic BERT models, including MARBERT.
- MARBERT labels 95.80% of headlines Neutral, while LLaMA-3.1-8B labels 98.94% Negative. Qwen2.5-7B sits between GPT-4.1 and the BERT models and is more reluctant than GPT-4.1 to assign non-neutral labels.
- Disagreement is measured with three tools: Shannon Entropy for how spread out a model's labels are, Jensen–Shannon Distance between models' label distributions, and a Variance Score for deviation from the aggregate of all models.
- Frame-conditioned analysis shows that GPT-4.1 adapts its sentiment to the frame, while the other LLMs barely react to frame differences.
- Stated limitations:
  - there is no human gold standard, by design;
  - the frame analysis rests on GPT-4.1 rationales only;
  - only headlines are analysed, not article bodies.

## Interesting ideas
- Instead of scoring each model against one gold label set, the paper compares models with each other at the level of whole label distributions (entropy, Jensen–Shannon distance, distance from the consensus), which exposes systematic leanings even without annotation. — For Mova (learn): when several witness parsers produce silver data, track each witness's label distribution and its JS distance from the pooled consensus to catch a parser that has drifted or collapsed before its votes contaminate agreement.
- A model collapsing almost entirely into one class (LLaMA-3.1-8B to negative, MARBERT to neutral) is treated as a finding in its own right, detected through very low entropy of its outputs. — For Mova (learn): add a cheap entropy check on LLM teacher-tier outputs per label type, so a teacher that answers nearly everything the same way is demoted or flagged as a negative control.
- Sentiment is analysed conditionally on narrative frame, so the same polarity label can be read differently depending on whether a text is framed as humanitarian, legal or security. — For Mova (frames): speech-act or evaluative judgments in prag could be stored together with a frame or perspective tag, so that the evaluation is explained by the frame and not given as a bare score.
- The paper argues that a neutral label on factual-sounding text about violence can hide evaluation that lies in world knowledge rather than in wording. — For Mova (world): evaluative stance could be derived from world states (harm, casualties, loss of holdings) recorded in the text's world, giving an explainable non-lexical source of negativity.
- LLM rationales are linked back to frames, turning a black-box label into something inspectable. — For Mova (learn): require LLM teachers to return a short rationale tied to a named frame or role, so their silver labels can be audited and turned into repair rules instead of accepted blindly.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

Fields: [Linguistics & languages](https://md.research.elamur.ai/free/en/shelves/linguistics/)

Topics: [Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) · [Sentiment & emotion](https://md.research.elamur.ai/free/en/topics/sentiment-emotion/) · [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/)

## Cite

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@misc{mustafa2026sentiment,
  title = {{Sentiment Classification of Gaza War Headlines: A Comparative Analysis of Large Language Models and Arabic Fine-Tuned BERT Models}},
  author = {Mustafa;, Amr Eleraqi; Hager H. and Ahmed, Abdul Hadi N.},
  year = {2026},
  doi = {10.2139/ssrn.6332158},
  eprint = {2604.08566},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.2139/ssrn.6332158}
}
```

- [← previous Sentiment Analysis of German Sign Language Fairy Tales](https://md.research.elamur.ai/free/en/papers/nunnari-2026-sentiment-analysis-german-sign-language/)
- [next → Sequence Labeling for Constituent Parsing: A Comparative Study and Encoding Innovations](https://md.research.elamur.ai/free/en/papers/roca-2026-sequence-labeling-constituent-parsing-comparative/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/9/)

## References

27 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Abdul-Mageed, Muhammad, AbdelRahim Elmadany, and El Moatez Billah Nagoudi. 2021. “ARBERT & MARBERT: Deep Bidirectional Transformers for Arabic.” Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics (ACL 2021), Long Papers, 7088–7105. Association for Computational Linguistics. https://aclanthology.org/2021.acl-long.551.pdf [↗](https://aclanthology.org/2021.acl-long.551/) [our card](https://md.research.elamur.ai/free/en/papers/abdul-mageed-2021-arbert-marbert-deep-bidirectional-transformers/)
2. Abuasaker, Walaa, Mónica Sánchez, Jennifer Nguyen, Nil Agell, Núria Agell, and Francisco J. Ruiz. 2025. “A Comparative Analysis of European Media Coverage of the Israel–Gaza War Using Hesitant Fuzzy Linguistic Term Sets.” Machine Learning and Knowledge Extraction 7 (1): 8. https://doi.org/10.3390/make7010008. [↗](https://doi.org/10.3390/make7010008)
3. Almutrash, Salman, and Shadi Abudalfa. 2024. “Comparative Study on the Efficiency of Using PaLM and CAMeLBERT for Arabic Entity Sentiment Classification.” In SaudiCIS 2024 Proceedings (1st Saudi Conference on Information Systems, Dhahran, Saudi Arabia, November 19–21, 2024). AIS eLibrary. [https://aisel.aisnet.org/saudicis2024/66](https://aisel.aisnet.org/saudicis2024/66) [↗](https://aisel.aisnet.org/saudicis2024/66)
4. Antoun, Wissam, Fady Baly, and Hazem Hajj. 2020. “AraBERT: Transformer-based Model for Arabic Language Understanding.” In Proceedings of the 4th Workshop on Open-Source Arabic Corpora and Processing Tools, with a Shared Task on Offensive Language Detection (OSACT), 9–15. Marseille, France: European Language Resource Association. https://aclanthology.org/2020.osact-1.2/. [↗](https://aclanthology.org/2020.osact-1.2/) [our card](https://md.research.elamur.ai/free/en/papers/antoun-2020-arabert-transformer-based-model-arabic/)
5. Bommasani, Rishi, et al. 2021. “On the Opportunities and Risks of Foundation Models.” arXiv (August 2021). https://doi.org/10.48550/arXiv.2108.07258 [↗](https://doi.org/10.48550/arxiv.2108.07258) [our card](https://md.research.elamur.ai/free/en/papers/bommasani-2021-opportunities-risks-foundation-models/)
6. Boudad, Naima, Rdouan Faizi, Rachid Oulad Haj Thami, and Raddouane Chiheb. 2018. “Sentiment Analysis in Arabic: A Review of the Literature.” Ain Shams Engineering Journal 9 (4): 2479–2490. https://doi.org/10.1016/j.asej.2017.04.007. [↗](https://doi.org/10.1016/j.asej.2017.04.007)
7. Ceron, Andrea, Luigi Curini, and Stefano M. Iacus. 2015. “Using Sentiment Analysis to Monitor Electoral Campaigns: Method Matters—Evidence From the United States and Italy.” Social Science Computer Review 33 (1): 3–20. https://doi.org/10.1177/0894439314521983. [↗](https://doi.org/10.1177/0894439314521983)
8. Devlin, Jacob, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding.” In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), 4171–4186. Minneapolis, Minnesota: Association for Computational Linguistics. https://doi.org/10.18653/v1/N19-1423. [↗](https://doi.org/10.18653/v1/n19-1423) [our card](https://md.research.elamur.ai/free/en/papers/devlin-2019-bert-pre-training-deep-bidirectional/)
9. Eleraqi, Amr. 2026. “Arabic News Corpus on the Gaza War and Geopolitical Narratives (2023–2025).” Harvard Dataverse, V1.0 (January 4, 2026). https://doi.org/10.7910/DVN/FFENX3. [↗](https://doi.org/10.7910/dvn/ffenx3)
10. Entman, Robert M. 1993. “Framing: Toward Clarification of a Fractured Paradigm.” Journal of Communication 43 (4): 51–58. https://doi.org/10.1111/j.1460-2466.1993.tb01304.x [↗](https://doi.org/10.1111/j.1460-2466.1993.tb01304.x)
11. Grimmer, Justin, and Brandon M. Stewart. 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. [↗](https://doi.org/10.1093/pan/mps028)
12. Gururangan, Suchin, Ana Marasović, Swabha Swayamdipta, Kyle Lo, Iz Beltagy, Doug Downey, and Noah A. Smith. 2020. “Don’t Stop Pretraining: Adapt Language Models to Domains and Tasks.” In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 8342–8360. Online: Association for Computational Linguistics. https://doi.org/10.18653/v1/2020.acl-main.740. [↗](https://doi.org/10.18653/v1/2020.acl-main.740)
13. Habash, Nizar Y. 2010. Introduction to Arabic Natural Language Processing. Synthesis Lectures on Human Language Technologies, no. 10. San Rafael, CA: Morgan & Claypool Publishers. https://doi.org/10.2200/S00277ED1V01Y201008HLT010. [↗](https://doi.org/10.2200/s00277ed1v01y201008hlt010)
14. Hannani, Mohamed, Abdelhadi Soudi, and Kristof Van Laerhoven. 2024. “Assessing the Performance of ChatGPT-4, Fine-tuned BERT and Traditional ML Models on Moroccan Arabic Sentiment Analysis.” In Proceedings of the 4th International Conference on Natural Language Processing for Digital Humanities (NLP4DH 2024). https://aclanthology.org/2024.nlp4dh-1.47.pdf. [↗](https://aclanthology.org/2024.nlp4dh-1.47/)
15. Haselmayer, Martin, and Marcelo Jenny. 2017. “Sentiment Analysis of Political Communication: Combining a Dictionary Approach with Crowdcoding.” Quality & Quantity 51 (6): 2623–2646. https://doi.org/10.1007/s11135-016-0412-4. [↗](https://doi.org/10.1007/s11135-016-0412-4)
16. Huang, Lei, et al. 2023. “A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions.” arXiv (November 2023). https://doi.org/10.48550/arXiv.2311.05232. [↗](https://doi.org/10.48550/arxiv.2311.05232) [our card](https://md.research.elamur.ai/free/en/papers/huang-2023-survey-hallucination-large-language-models/)
17. Inoue, Go, Bashar Alhafni, Nurpeiis Baimukan, Houda Bouamor, and Nizar Habash. 2021. “The Interplay of Variant, Size, and Task Type in Arabic Pre-trained Language Models.” In Proceedings of the Sixth Arabic Natural Language Processing Workshop, 92–104. Kyiv, Ukraine (Virtual): Association for Computational Linguistics. https://aclanthology.org/2021.wanlp-1.10/. [↗](https://aclanthology.org/2021.wanlp-1.10/) [our card](https://md.research.elamur.ai/free/en/papers/inoue-2021-interplay-variant-size-task-type/)
18. Ke, Zixuan, Yijia Shao, Haowei Lin, Hu Xu, Lei Shu, and Bing Liu. 2022. “Adapting a Language Model While Preserving its General Knowledge.” In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, 10177–10188. Abu Dhabi, United Arab Emirates: Association for Computational Linguistics. https://doi.org/10.18653/v1/2022.emnlp-main.693. [↗](https://doi.org/10.18653/v1/2022.emnlp-main.693)
19. Kim, Yoon. 2014. “Convolutional Neural Networks for Sentence Classification.” In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), 1746–1751. Doha, Qatar: Association for Computational Linguistics. https://doi.org/10.3115/v1/D14-1181. [↗](https://doi.org/10.3115/v1/d14-1181) [our card](https://md.research.elamur.ai/free/en/papers/kim-2014-convolutional-neural-networks-sentence-classification/)
20. Krippendorff, Klaus. 2019. Content Analysis: An Introduction to Its Methodology. 4th ed. Thousand Oaks, CA: SAGE Publications, Inc. https://doi.org/10.4135/9781071878781. [↗](https://doi.org/10.4135/9781071878781)
21. Kullback, Solomon, and Richard A. Leibler. 1951. “On Information and Sufficiency.” The Annals of Mathematical Statistics 22 (1): 79–86. https://doi.org/10.1214/aoms/1177729694. [↗](https://doi.org/10.1214/aoms/1177729694)
22. Liu, Pengfei, Weizhe Yuan, Jinlan Fu, Zhengbao Jiang, Hiroaki Hayashi, and Graham Neubig. 2021. “Pre-train, Prompt, and Predict: A Systematic Survey of Prompting Methods in Natural Language Processing.” arXiv (July 2021). https://doi.org/10.48550/arXiv.2107.13586. [↗](https://doi.org/10.48550/arxiv.2107.13586) [our card](https://md.research.elamur.ai/free/en/papers/liu-2021-pre-train-prompt-predict-systematic/)
23. McCombs, Maxwell E., and Donald L. Shaw. 1972. “The Agenda-Setting Function of Mass Media.” Public Opinion Quarterly 36 (2): 176–187. https://doi.org/10.1086/267990. [↗](https://doi.org/10.1086/267990)
24. Mikolov, Tomas, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013. “Efficient Estimation of Word Representations in Vector Space.” arXiv:1301.3781. https://doi.org/10.48550/arXiv.1301.3781. [↗](https://doi.org/10.48550/arxiv.1301.3781) [our card](https://md.research.elamur.ai/free/en/papers/mikolov-2013-efficient-estimation-word-representations-vector/)
25. Mulki, Hala, Hatem Haddad, and Ismail Babaoğlu. 2017. “Modern Trends in Arabic Sentiment Analysis: A Survey.” Traitement Automatique des Langues 58 (3): 15–39. https://aclanthology.org/2017.tal-3.3/ [↗](https://aclanthology.org/2017.tal-3.3/)
26. OpenAI. 2023. “GPT-4 Technical Report.” arXiv 2303.08774. https://doi.org/10.48550/arXiv.2303.08774. [↗](https://doi.org/10.48550/arxiv.2303.08774) [our card](https://md.research.elamur.ai/free/en/papers/openai-2023-gpt-4-technical-report/)
27. Shannon, Claude E. 1948. “A Mathematical Theory of Communication.” Bell System Technical Journal 27 (3): 379–423; 27 (4): 623–656. https://doi.org/10.1002/j.1538-7305.1948.tb01338.x. [↗](https://doi.org/10.1002/j.1538-7305.1948.tb01338.x)