Exploring Graph-structured Passage Representation for Multi-hop Reading Comprehension with Graph Neural Networks

TL;DR

This paper proposes a novel graph neural network approach for multi-hop reading comprehension, leveraging complex graph structures to improve evidence integration and answer accuracy.

Contribution

It introduces a new method that uses richer graph structures and GCN/GRN models for better evidence connection in multi-hop QA tasks.

Findings

Richer global evidence improves answer accuracy.

The proposed method outperforms all previous results on standard datasets.

Graph neural networks effectively enhance multi-hop reasoning.

Abstract

Multi-hop reading comprehension focuses on one type of factoid question, where a system needs to properly integrate multiple pieces of evidence to correctly answer a question. Previous work approximates global evidence with local coreference information, encoding coreference chains with DAG-styled GRU layers within a gated-attention reader. However, coreference is limited in providing information for rich inference. We introduce a new method for better connecting global evidence, which forms more complex graphs compared to DAGs. To perform evidence integration on our graphs, we investigate two recent graph neural networks, namely graph convolutional network (GCN) and graph recurrent network (GRN). Experiments on two standard datasets show that richer global information leads to better answers. Our method performs better than all published results on these datasets.