Mohamed Bayan Kmainasi — محمد بيان قميناسي

Research Assistant · QCRI

Mohamed Bayan Kmainasi

محمد بيان قميناسي

I'm a Research Assistant at the Qatar Computing Research Institute (QCRI), where I work on natural language processing and multimodal models. Most of my research looks at how models can read and reason about memes, and how to detect harmful content such as hate speech and propaganda, with a particular focus on Arabic.

Lately I've been using reinforcement learning (GRPO) and chain-of-thought methods to help language and vision-language models reason better and explain their decisions. I finished my MSc in Computing at Qatar University in early 2026, after a BSc in Computer Engineering where I graduated first in my class. My work has appeared at NAACL, EMNLP, The Web Conference, and WISE.

Download CV Get in touch
4.0MSc GPA · Qatar University
3.99BSc GPA · Ranked 1st
11Papers & preprints
NAACL · EMNLP · WWW · WISEWhere my work appears

What I work on

🖼️

Multimodal NLP

Vision-language models that read and reason about memes, including hateful and propagandistic content across several languages.

🛡️

Content Moderation

Detecting harmful and propagandistic content in a way that explains the reasoning, not just the label.

🌍

Arabic Language Technology

Datasets, benchmarks, and multilingual models that make NLP work better for Arabic and other low-resource languages.

🎯

RL for LLMs

Using GRPO and chain-of-thought supervision to post-train models for stronger reasoning and explainability.

News

  • Jun 2026 New preprint on reinforcement learning with chain-of-thought supervision for explainable meme detection (arXiv).
  • 2026 Received the 3rd Best Poster Award at the MenaML Winter School (KAUST) for my thesis research.
  • 2026 MemeLens, a multilingual multimodal meme benchmark, released as a preprint (under review for ACL 2026).
  • 2026 Presented work on reasoning models for hateful-meme detection at The Web Conference (WWW 2026).
  • Jan 2026 Completed my MSc in Computing at Qatar University with a 4.0 GPA.
  • 2025 Two papers accepted at EMNLP 2025: MemeIntel (main) and PropXplain (Findings).

Publications

Selected papers below. The full, up-to-date list lives on my Google Scholar profile.

2026

  1. MemeLens: Multilingual Multitask VLMs for Memes

    A. E. Shahroor, M. B. Kmainasi, A. Hasnat, D. Dimitrov, G. Da San Martino, P. Nakov, F. Alam

    Preprint · ACL 2026 (under review) A multilingual, multimodal benchmark and models for understanding memes.

    arXiv Model Dataset 
    @article{shahroor2026memelens,
  title   = {MemeLens: Multilingual Multitask VLMs for Memes},
  author  = {Shahroor, Ali Ezzat and Kmainasi, Mohamed Bayan and Hasnat, Abul and Dimitrov, Dimitar and Da San Martino, Giovanni and Nakov, Preslav and Alam, Firoj},
  journal = {arXiv preprint arXiv:2601.12539},
  year    = {2026}
}

  2. Can Thinking Models Think to Detect Hateful Memes?

    M. B. Kmainasi, M. Kutlu, A. E. Shahroor, A. Hasnat, F. Alam

    WWW 2026 Companion Do reasoning ("thinking") models actually help for hateful-meme detection?

    ACM arXiv

  3. Adapting Reinforcement Learning with Chain-of-Thought Supervision for Explainable Detection of Hateful and Propagandistic Memes

    M. B. Kmainasi, M. Kutlu, A. E. Shahroor, A. Hasnat, F. Alam

    Preprint GRPO with chain-of-thought supervision for explainable multimodal detection.

    arXiv Code

  4. CritiSense: Critical Digital Literacy and Resilience Against Misinformation

    F. Alam, F. Ahmad, A. E. Shahroor, M. B. Kmainasi, E. Sartori, G. Da San Martino, et al.

    Preprint · ACL 2026 demo (under review) A tool for critical digital literacy and resilience against misinformation.

    arXiv

2025

  1. LlamaLens: Specialized Multilingual LLM for Analyzing News and Social Media Content

    M. B. Kmainasi, A. E. Shahroor, M. Hasanain, S. R. Laskar, N. Hassan, F. Alam

    NAACL 2025 Findings An instruction-tuned multilingual LLM for analyzing news and social-media content.

    ACL Anthology arXiv Model Dataset 
    @inproceedings{kmainasi2025llamalens,
  title     = {LlamaLens: Specialized Multilingual LLM for Analyzing News and Social Media Content},
  author    = {Kmainasi, Mohamed Bayan and Shahroor, Ali Ezzat and Hasanain, Maram and Laskar, Sahinur Rahman and Hassan, Naeemul and Alam, Firoj},
  booktitle = {Findings of the Association for Computational Linguistics: NAACL 2025},
  pages     = {5642--5664},
  year      = {2025},
  url       = {https://aclanthology.org/2025.findings-naacl.313/}
}

  2. MemeIntel: Explainable Detection of Propagandistic and Hateful Memes

    M. B. Kmainasi, A. Hasnat, M. A. Hasan, A. E. Shahroor, F. Alam

    EMNLP 2025 Explainable multimodal detection, with a dedicated instruction dataset.

    ACL Anthology arXiv Code Dataset 
    @inproceedings{kmainasi2025memeintel,
  title     = {MemeIntel: Explainable Detection of Propagandistic and Hateful Memes},
  author    = {Kmainasi, Mohamed Bayan and Hasnat, Abul and Hasan, Md. Arid and Shahroor, Ali Ezzat and Alam, Firoj},
  booktitle = {Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing},
  year      = {2025},
  url       = {https://aclanthology.org/2025.emnlp-main.1539/}
}

  3. PropXplain: Can LLMs Enable Explainable Propaganda Detection?

    M. Hasanain, M. A. Hasan, M. B. Kmainasi, E. Sartori, A. E. Shahroor, et al.

    EMNLP 2025 Findings Using LLMs to make propaganda detection explainable.

    ACL Anthology arXiv Dataset 
    @inproceedings{hasanain2025propxplain,
  title     = {PropXplain: Can LLMs Enable Explainable Propaganda Detection?},
  author    = {Hasanain, Maram and Hasan, Md. Arid and Kmainasi, Mohamed Bayan and Sartori, Elisa and Shahroor, Ali Ezzat and Da San Martino, Giovanni and Nakov, Preslav and Alam, Firoj},
  booktitle = {Findings of the Association for Computational Linguistics: EMNLP 2025},
  pages     = {23855--23863},
  year      = {2025},
  url       = {https://aclanthology.org/2025.findings-emnlp.1296/}
}

  4. EverydayMMQA: A Multilingual and Multimodal Framework for Culturally Grounded Spoken Visual QA

    F. Alam, A. E. Shahroor, M. A. Hasan, Z. S. Ali, H. H. Bhatti, M. B. Kmainasi, et al.

    Preprint · ICML 2026 (under review) Culturally grounded spoken visual question answering.

    arXiv

  5. Can Large Language Models Predict the Outcome of Judicial Decisions?

    M. B. Kmainasi, A. E. Shahroor, A. Al-Ghraibah

    Preprint LLMs for Arabic legal judgment prediction.

    arXiv

2024

  1. Native vs Non-Native Language Prompting: A Comparative Analysis

    M. B. Kmainasi, R. Khan, A. E. Shahroor, B. Bendou, M. Hasanain, F. Alam

    WISE 2024 My most-cited paper: does prompting in your native language help?

    Springer arXiv 
    @inproceedings{kmainasi2024native,
  title     = {Native vs Non-Native Language Prompting: A Comparative Analysis},
  author    = {Kmainasi, Mohamed Bayan and Khan, Rakif and Shahroor, Ali Ezzat and Bendou, Boushra and Hasanain, Maram and Alam, Firoj},
  booktitle = {Web Information Systems Engineering -- WISE 2024},
  pages     = {406--420},
  year      = {2024},
  url       = {https://doi.org/10.1007/978-981-96-0576-7_30}
}

  2. Automated Detection of Lung Cancer Levels Based on Patient's Lifestyle Information

    A. Al-Ghraibah, A. Al-Abbas, M. B. Kmainasi, I. E. Mustafa, K. Brmo

    IEEE · JIBEC 2024 Machine learning for lung-cancer level detection from lifestyle data.

    IEEE Xplore

Experience

  1. Research AssistantJan 2025 – Present
    Qatar Computing Research Institute (QCRI)

    Work on the Fanar speech project training large multilingual speech LLMs, with papers at EMNLP 2025 and WWW 2026 and more under review. Also contributed to EduLLM on QA and testing.

  2. Research InternMay 2024 – Jan 2025
    Qatar Computing Research Institute (QCRI)

    First-author papers at WISE 2024 and NAACL 2025. Placed 2nd in the QCRI Programming Contest 2024.

  3. Software Engineer InternJul 2022 – Sep 2022
    Qatar Fuel Additives Company (QAFAC)

    Built an internal safety app for incident reporting and team communication.

Education

  1. MSc in Computing2024 – 2026
    Qatar University · GPA 4.0/4.0
  2. BSc in Computer Engineering2019 – 2023
    Al-Ahliyya Amman University · GPA 3.99/4.0, ranked 1st

Awards

  • 3rd Best Poster — MenaML Winter School, KAUST (2026)
  • 2nd Place Research Poster — QCRI (2024)
  • 2nd Place — QCRI Programming Contest (2024)
  • Top Graduate (1st) — Computer Engineering, Al-Ahliyya Amman University (2023)

Projects & Open Resources

Datasets, models, and systems I've released with my research. Most are public on GitHub and Hugging Face.

LlamaLens

Specialized multilingual LLM with instruction datasets in Arabic, English, and Hindi for analyzing news and social media.

Multilingual LLMNAACL 2025Dataset
PaperModelDatasets

MemeLens

Multilingual, multimodal benchmark and vision-language models for understanding memes.

VLMBenchmarkMultimodal
PaperModelDataset

MemeIntel

Explainable detection of propagandistic and hateful memes, released with the MemeXplain dataset.

MultimodalEMNLP 2025Dataset
PaperCodeDataset

PropXplain

Explainable propaganda detection with LLMs, released with the PropXplain dataset.

PropagandaEMNLP 2025Dataset
PaperDataset

MemeReason

Code for the reinforcement-learning (GRPO) and chain-of-thought approach to explainable meme detection.

RLChain-of-ThoughtCode
PaperCode

LLMeBench

Contributor to QCRI's framework for benchmarking LLMs across many tasks and languages.

BenchmarkingOpen Source
GitHub

Contact

The fastest way to reach me is by email. I'm open to research collaborations, especially around multimodal NLP and Arabic language technology.

Email mkmainasi@hbku.edu.qa Google Scholar h5tFJ7MAAAAJ GitHub @MohamedBayan Hugging Face QCRI

📍 Doha, Qatar · Qatar Computing Research Institute (QCRI)