Cognitive Graph for Multi-Hop Reading Comprehension at Scale

Abstract

We propose a new CogQA framework for multi-hop question answering in web-scale documents. Inspired by the dual process theory in cognitive science, the framework gradually builds a \textit{cognitive graph} in an iterative process by coordinating an implicit extraction module (System 1) and an explicit reasoning module (System 2). While giving accurate answers, our framework further provides explainable reasoning paths. Specifically, our implementation based on BERT and graph neural network efficiently handles millions of documents for multi-hop reasoning questions in the HotpotQA fullwiki dataset, achieving a winning joint $F_1$ score of 34.9 on the leaderboard, compared to 23.6 of the best competitor.

ACL 2019

References in corpus (5)

• Teaching Machines to Read and Comprehend
• Bidirectional Attention Flow for Machine Comprehension
• The Goldilocks Principle: Reading Children's Books with Explicit Memory Representations
• ReasoNet: Learning to Stop Reading in Machine Comprehension
• HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering

Cited by in corpus (8)

• Learning to Retrieve Reasoning Paths over Wikipedia Graph for Question Answering
• A Survey on Explainability in Machine Reading Comprehension
• Accelerating Training of Transformer-Based Language Models with Progressive Layer Dropping
• Cognitive Knowledge Graph Reasoning for One-shot Relational Learning
• Pre-Trained Models: Past, Present and Future
• Hopper: Multi-hop Transformer for Spatiotemporal Reasoning
• Simple yet Effective Bridge Reasoning for Open-Domain Multi-Hop Question Answering
• Fine-tuning Multi-hop Question Answering with Hierarchical Graph Network