[关闭]
@songying 2018-12-21T16:49:25.000000Z 字数 3819 阅读 1527

Neural Reading Comprehension and Beyond

综述类论文


Abstract

本文针对的问题: how to build computer systems to read a passage of text and answer comprehension questions。一方面,我们认为阅读理解能够更好的评估机器是否能够理解人类语言;另一方面,如果我们能够创建高性能的阅读理解系统,这将是一些应用如question answering 和 dialogue systems 的关键技术。

在本文中,我们专注于neural reading comprehension。
本文包含两部分:

Chapter 1. Introduction

Motivation

  • the formulation of the problem
  • the building blocks of these systems
  • the key ingredients of these systems
  • understanding of where current neural reading comprehension systems can excel
  • where current neural reading comprehension systems still lag behind.
  • Open-domain question answering combines the challenges from both information retrieval and reading comprehension and aims to answer general questions from either the Web or a large encyclopedia
  • Conversational question answering combines the challenges from dialogue and reading comprehension, and tackles the problem of multi-turn question answering over a passage of text, like how users would engage with conversational agents.

Thesis Outline -- 本文大纲

  • 第二章: 我们首先描述了阅读理解的历史以及最近的发展;然后我们正式的定义了问题的格式以及分类;然后简单的讨论了以下 reading comprehension 与 general question answering 的不同点。最后,我们讨论了最近成功的neural reading comprehension。
  • 第三章:我们介绍了现有的 neural reading comprehension models. 我们先介绍了传统方法,然后讨论它们与端到端方法的区别。我们然后引入一个神经网络方法“The Standford Attentive Reader" 并描述它的基本组件。最后,我们从不同角度总结最近的改进。
  • 第四章: 我们讨论了该领域的未来工作以及开放问题。我们先检测了现有模型的准确度。然后从数据集以及模型的角度讨论未来的方向;最后,我们回顾了该领域中几个重要的研究问题,这些问题目前仍待解决。
  • 第五章: 我们将 reading comprehension用于解决 open domain question answering。
  • 第六章: we study the problem of conversational question answering, where a machine has to understand a text passage and answer a series of questions that appear in a conversation

Contributions -- 本文贡献


Chapter 2. An Overview of reading comprehension

本章中,我们对reading comprehension 进行了简单综述。 我们从 reading comprehension 历史开始,然后讲述最近神经网络方法的兴起。
然后,我们定义了reading comprehension task 并根据答案类型描述了四种不同的分类并讨论了他们的评价标准。
其次,我们讨论了reading comprehension 与 question answering 的区别。

1. History

跳过

2. Task Definition

1. Problem Formulation -- 问题格式

2. Evaluation -- 评价标准

3. Reading Comprehension vs. Question Answering

4. Datasets and models

Chapter 3: Neural Reading Comprehension Models

1. Previous Approaches: Feature-based Models

2. A Neural Approach: The Stanford Attentive Reader

3. Experiments

4. Further Advances

Chapter 4 The Future of Reading Comprehension

1. Is SQuAD Solved Yet?

2. Future Work: Datasets

3. Future Work: Models

4. Research Questions


Chapter 5 Open Domain Question Answering

1. A Brief History of Open-domain QA

2. Our System: D R QA

3. Evaluation

4. Future Work

Chapter 6 Conversational Question Answering

添加新批注
在作者公开此批注前,只有你和作者可见。
回复批注