Graph Reasoning for Question Answering with Triplet Retrieval
Shiyang Li1, Yifan Gao2, Haoming Jiang2, Qingyu Yin2, Zheng Li2, Xifeng Yan1 Chao Zhang3, Bing Yin2 1University of California, Santa Barbara 2Amazon Inc. 3Georgia Institute of Technology {yifangao,jhaoming, qingyy, amzzhe, Work was done during internship at Amazon.
Abstract
Answering complex questions often requires reasoning over knowledge graphs (KGs). State-of-the-art methods often utilize entities in questions to retrieve local subgraphs, which are then fed into KG encoder, e.g. graph neural networks (GNNs), to model their local structures and integrated into language models for question answering. However, this paradigm constrains retrieved knowledge in local subgraphs and discards more diverse triplets buried in KGs that are disconnected but useful for question answering. In this paper, we propose a simple yet effective method to first retrieve the most relevant triplets from KGs and then rerank them, which are then concatenated with questions to be fed into language models. Extensive results on both CommonsenseQA and OpenbookQA datasets show that our method can outperform state-of-the-art up to 4.6% absolute accuracy.
中文速览
用知识图谱（Knowledge Graph, KG）辅助回答复杂问题时，现有主流方法依赖实体链接来抽取局部子图，再用图神经网络编码后融入语言模型，但这种方式会遗漏大量与问题相关却游离于子图之外的三元组知识，而且需要从头训练额外的图编码器。为此，研究者提出了一种更简洁的方案：把知识图谱中的每条三元组转化成自然语言短句，构成一个可检索的语料库，然后用稀疏检索（BM25）和稠密检索（DPR）混合召回最相关的条目，再经过预训练交叉编码器重排序，最后将筛选出的知识片段直接拼接到问题后面，送入RoBERTa等语言模型作答。在CommonsenseQA和OpenbookQA两个常识与科学推理基准上，这一方法相比当前最优系统最高提升4.6个百分点的绝对准确率，且全程无需引入任何额外可训练参数。这项工作表明，用"检索+重排序"替代复杂的图编码器，既能更充分地利用知识图谱中的全局信息，又能以更低的模型复杂度实现更强的性能，为知识图谱与语言模型的融合研究提供了新的思路。
原文 arXiv:2305.18742;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2305.18742v1