Abdullah Ibne Hanif Arean

I am excited to announce that our new Bangla question-answering dataset, SOMAJGYAAN, is now publicly available—and has also been accepted for publication as a research paper at IJCNLP-AACL 2025! Credit goes to Fariha Anjum Shifa Apu & Md. Fahim Bhai for the hardwork and putting everything together. Why does this matter? 1. SOMAJGYAAN is Bangladesh’s one of the largest and most diverse MCQ dataset, which is manually annotated(4,234 questions) covering history, geography, social studies, economics, law, and politics—all sourced directly from national curriculum textbooks and guides. 2.The dataset is meticulously annotated to test not only factual recall but also reasoning, contextual understanding, and knowledge of uniquely Bangladeshi topics. 3. This resource empowers researchers and developers working on Bangla LLMs, AI for low-resource languages, educational technology, and social knowledge tasks to perform fair and rigorous benchmarking. By making this dataset open-access, we invite participation from the global NLP community to advance fairness, inclusiveness, and innovation in AI for low-resource languages. The dataset supports model training, transfer learning, LoRA fine-tuning, and evaluation of both open- and closed-source models on culture-rich and socially relevant question answering. We hope SOMAJGYAAN will inspire new directions in Bangla NLP and make AI research more representative and rigorous for our community. Explore and download now on Kaggle & Hugging Face! Kaggle Link: https://lnkd.in/guuSsHyC Hugging Face Link: https://lnkd.in/gXBcuUiS #BanglaNLP #LLM #Kaggle #OpenData #AACL2025 #IJCNLP #AIForGood #SocialKnowledge #MachineLearning #PEFT #ZeroShot #FairAI