Multi-step Retriever-Reader Interaction for Scalable Open-domain Question Answering
Rajarshi Das1, Shehzaad Dhuliawala2, Manzil Zaheer3、Andrew McCallum1 1 University of Massachusetts, Amherst, 2 Microsoft Research, Montréal 3 Google AI, Mountain View
Abstract
This paper introduces a new framework for open-domain question answering in which the retriever and the reader iteratively interact with each other. The framework is agnostic to the architecture of the machine reading model, only requiring access to the token-level hidden representations of the reader. The retriever uses fast nearest neighbor search to scale to corpora containing millions of paragraphs. A gated recurrent unit updates the query at each step conditioned on the state of the reader and the reformulated query is used to re-rank the paragraphs by the retriever. We conduct analysis and show that iterative interaction helps in retrieving informative paragraphs from the corpus. Finally, we show that our multi-step-reasoning framework brings consistent improvement when applied to two widely used reader architectures (Dr.qa and bidaf) on various large open-domain datasets — Triviaqa-unfiltered, Quasar-t, Searchqa, and Squad-open 111Code and pretrained models are available at https://github.com/rajarshd/Multi-Step-Reasoning.
中文速览
开放域问答系统通常依赖一个"检索器"先找到相关段落，再交给"阅读器"提取答案，但检索器一旦找错段落，整个系统就会失败。这篇论文提出了一种让检索器与阅读器反复"对话"的框架：检索器先用预先缓存好的段落向量做快速近邻搜索（Maximum Inner Product Search）找出候选段落，阅读器读完后通过一个门控循环单元（GRU）更新查询向量，再把新的查询交还给检索器重新检索，如此循环多轮，逐步找到更相关的段落并跨段落整合证据。由于段落向量与查询无关、可离线存储，系统能扩展到数百万段落的大规模语料库，同时兼容DrQA、BiDAF等不同阅读器架构，并用强化学习联合训练两个模块。实验在TriviaQA、Quasar-T、SearchQA、SQuAD-open四个大型数据集上均取得持续提升，证明这种"迭代检索-阅读"机制是提升开放域问答性能的有效且高效的通用方案。
原文 arXiv:1905.05733;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1905.05733v1