Multi-Paragraph Reasoning with Knowledge-enhanced Graph Neural Network
This work addresses a bottleneck in open-domain question answering systems by improving reasoning over multiple paragraphs, which is incremental as it builds on existing methods with knowledge graph enhancements.
The paper tackles the problem of multi-paragraph reasoning in open-domain question answering by proposing a knowledge-enhanced graph neural network (KGNN) that uses relational facts from knowledge graphs to build entity graphs, resulting in outperforming baseline methods on the HotpotQA dataset in both distractor and full wiki settings.
Multi-paragraph reasoning is indispensable for open-domain question answering (OpenQA), which receives less attention in the current OpenQA systems. In this work, we propose a knowledge-enhanced graph neural network (KGNN), which performs reasoning over multiple paragraphs with entities. To explicitly capture the entities' relatedness, KGNN utilizes relational facts in knowledge graph to build the entity graph. The experimental results show that KGNN outperforms in both distractor and full wiki settings than baselines methods on HotpotQA dataset. And our further analysis illustrates KGNN is effective and robust with more retrieved paragraphs.