References & Citations
Computer Science > Computation and Language
Title: Grape: Knowledge Graph Enhanced Passage Reader for Open-domain Question Answering
(Submitted on 6 Oct 2022 (v1), last revised 10 Oct 2022 (this version, v2))
Abstract: A common thread of open-domain question answering (QA) models employs a retriever-reader pipeline that first retrieves a handful of relevant passages from Wikipedia and then peruses the passages to produce an answer. However, even state-of-the-art readers fail to capture the complex relationships between entities appearing in questions and retrieved passages, leading to answers that contradict the facts. In light of this, we propose a novel knowledge Graph enhanced passage reader, namely Grape, to improve the reader performance for open-domain QA. Specifically, for each pair of question and retrieved passage, we first construct a localized bipartite graph, attributed to entity embeddings extracted from the intermediate layer of the reader model. Then, a graph neural network learns relational knowledge while fusing graph and contextual representations into the hidden states of the reader model. Experiments on three open-domain QA benchmarks show Grape can improve the state-of-the-art performance by up to 2.2 exact match score with a negligible overhead increase, with the same retriever and retrieved passages. Our code is publicly available at this https URL
Submission history
From: Mingxuan Ju [view email][v1] Thu, 6 Oct 2022 14:06:04 GMT (615kb,D)
[v2] Mon, 10 Oct 2022 01:12:29 GMT (1086kb,D)
Link back to: arXiv, form interface, contact.