This dataset contains the artifacts from an evidence-grounded multimodal knowledge graph construction pipeline over three neural-network lectures.
Raw videos, audio, and 1 FPS frames.
Faster-Whisper transcripts.
High-recall semantic anchors with transcript windows.
EasyOCR outputs for anchor frames.
Qwen2.5-VL raw extractions.
Validated concept and relationship mentions.
Canonical concepts and… See the full description on the dataset page:
https://huggingface.co/datasets/sahilfarib/evidence-grounded-multimodal-kg-multi-lecture-reasoning.
