intfloat/multilingual-e5-large fine-tuned with MultipleNegativesRankingLoss on Arabic text paired with hand-augmented English genre definitions, for hierarchical Arabic genre classification (broad_genre + specific_genre).

Authors: Hassan Barmandah (NAMAA Community; Umm Al-Qura University), Israa Elhosiny (NAMAA Community), Yousra El-Ghawi (NAMAA Community), Omer Nacar (NAMAA Community)

This model's development-set score (0.9316 hierarchical F1) is not representative of real-world performance. Per the project's system-description paper, this entire lineage of fine-tuned/ensembled sentence encoders — which scored well on the 110-item, 6-genre AraGenre dev set — collapsed to 0.22–0.44 hierarchical F1 on the actual 27,972-item hidden test set (74 specific genres under 6 broad genres). The cause: these models were calibrated to a hand-augmented definition style that does not match the test set's official genre definitions or its much larger taxonomy.

The system that actually won for this team — 0.7013 hierarchical F1, 3rd of 18 teams on the official CodaBench leaderboard — was a separate, zero-shot DeepSeek-LLM pipeline with no fine-tuning at all (stage2_llm_zeroshot_pipeline/ in the project repo). This model is not that system. It is released here for transparency and reproducibility of the project's full experimental record, not as a recommended production classifier.

intfloat/multilingual-e5-large fine-tuned with MultipleNegativesRankingLoss (single phase) on (text, augmented definition) pairs from the 7 AraGenre TRAIN genres. This was the first recipe in its lineage to cross 0.90 hierarchical F1 on dev.

AraGenre TRAIN genres only (7), with hand-augmented definitions. No dev labels were used in training.

See the project repository for the full script and data-loading requirements.

If you use this work, please cite our system-description paper:

Please also cite the AraGenre 2026 shared task overview paper:

Apache 2.0