SciMKG: A Multimodal Knowledge Graph for Science Education with Text, Image, Video and Audio

SciMKG is a large-scale multimodal educational knowledge graph (MEKG) designed for Chinese K12 education (covering biology, physics, and chemistry). It leverages large language models (LLMs) to automatically extract and align concepts from diverse educational materials such as text, images, videos, and audio, enabling structured and intelligent educational content understanding.

🔍 Key Features

Automated Multimodal Construction: Introduce an Extraction–Verification–Integration–Augmentation pipeline to incrementally extract and refine disciplinary concepts.

Cross-Modal Alignment: Ensure semantic consistency between modalities via shared structural and semantic representations.

Large-Scale Coverage: 
1,356 knowledge points
34,630 multimodal concepts
403,400 relational triples