KnowledgeGraphEval 2026 shared task - First Call for Participation
Alaa Aljabari
KnowledgeGraphEval 2026
Dear All,
We are pleased to announce KnowledgeGraphEval 2026, a new shared task at ArabicNLP 2026 and the first shared task dedicated to Arabic knowledge graph construction from text.
The shared task focuses on building structured knowledge representations that support downstream applications such as information retrieval, semantic search, question answering, and modern NLP systems, including Retrieval-Augmented Generation (RAG) and Large Language Models (LLMs).
To simplify the construction of knowledge graphs, we decompose the problem into two core tasks that are well established within the NLP community: named entity recognition and semantic relation extraction.
KnowledgeGraphEval 2026 includes two subtasks:
• Cross-Domain Named Entity Recognition (AdaptNER): Adapting NER models across 10 different domains.
• Relation Extraction (RE): Identify and classify semantic relations between entity pairs.
The shared task emphasizes robust evaluation under both in-domain and out-of-domain settings across multiple domains.
We also introduce two new resources: Konooz, a large-scale multi-domain and multi-dialect Arabic NER dataset, and WojoodRelations, a large-scale Arabic relation extraction dataset.
Datasets, baselines, and evaluation scripts are available on the website.
🔗 Website: https://sina.birzeit.edu/KnowledgeGraphEval
🔗 Registration: https://forms.gle/xGuSr76W6jnxV7At9
We encourage researchers and students working on Arabic NLP, information extraction, LLMs, domain adaptation, and knowledge graphs to participate.
Best Regards,
The KnowledgeGraphEval 2026 Organizing Team