← all repositories

THUDM/CogQA

A multi-hop question answering system that builds cognitive graphs by coordinating BERT-based extraction and graph neural network reasoning.

458 stars Python Language ModelsRAG · Search
CogQA
Velocity · 7d
+0.2
★ / day
Trend
steady
star history

CogQA implements a dual-process theory framework for multi-hop reading comprehension at scale. It coordinates an implicit extraction module (System 1) using BERT with an explicit reasoning module (System 2) using graph neural networks to iteratively build cognitive graphs and answer questions. The system retrieves and reasons over Wikipedia documents to provide explainable reasoning paths alongside accurate answers.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.