RACE数据集上相关的研究

根据引用RACE的文献，看看RACE数据集都被用来做了什么事情
主要分为以下几个方面：

1 一些预训练模型的优化与研究。

MPNet: Masked and Permuted Pre-training for Language Understanding
一种新的预训练的模型，在多个数据集上表现良好包括RACE阅读理解数据集上
Improving Language Understanding by Generative Pre-Training
OpenAI GPT
Cracking the Contextual Commonsense Code: Understanding Commonsense Reasoning Aptitude of Deep Contextual Representations
预训练模型的能力理解。针对bert,实验证明bert对一些常识信息有很好的编码能力。哪些因素影响其表现
What does BERT Learn from Multiple-Choice Reading Comprehension Datasets?
通过对数据集加噪变化，微调BERT主要学习关键字如何导致正确的预测，而不是学习语义理解和推理；以及 BERT不需要正确的句法信息来完成任务

2 一些改善阅读理解问题的模型

DCMN+: Dual Co-Matching Network for Multi-choice Reading Comprehension
通过结合bert提取的Question, Article, Answer之间的语义表示，联合匹配获得更好的结果
A Study of Multi-Choice Questions Solution Strategy in Machine Reading Comprehension
多种注意机制获得给定文章和问题的交互语义
Unified Generative Adversarial Networks for Multiple-Choice Oriented Machine Comprehension
生成网络根据文章和问题生成一个与之相关的答案，判断网络，预测答案，文章，问题之间的相关性
A Multiple Granularity Co-Reasoning Model for Multi-choice Reading Comprehension
Residual Connection-Based Multi-step Reasoning via Commonsense Knowledge for Multiple Choice Machine Reading Comprehension
A BERT based model for Multiple-Choice Reading Comprehension
Option Attentive Capsule Network for Multi-choice Reading Comprehension
Machine comprehension using semantic graphs
Multi-task Learning with Multi-head Attention for Multi-choice Reading Comprehension
Inferential Machine Comprehension: Answering Questions by Recursively Deducing the Evidence Chain from Text
A Co-Matching Model for Multi-choice Reading Comprehension
Hierarchical Attention Flow for Multiple-Choice Reading Comprehension
Multi-range Reasoning for Machine Comprehension
Densely Connected Attention Propagation for Reading Comprehension
R-Trans: RNN Transformer Network for Chinese Machine Reading Comprehension中文阅读理解模型
Convolutional Spatial Attention Model for Reading Comprehension with Multiple-Choice Questions
FreeLB: Enhanced Adversarial Training for Natural Language Understanding
SG-Net: Syntax-Guided Machine Reading Comprehension
Exploiting Explicit Paths for Multi-hop Reading Comprehension
Dual Multi-head Co-attention for Multi-choice Reading Comprehension
Option Attentive Capsule Network for Multi-choice Reading Comprehension
MultiQA: An Empirical Investigation of Generalization and Transfer in Reading Comprehension多个数据集是否可以通用

3 多选项阅读理解的方法
主要包括以下几个方面

• 阅读理解与段落定位

Selecting Paragraphs to Answer Questions for Multi-passage Machine Reading Comprehension
将阅读理解问题定位到段落
Efficient and Robust Question Answering from Minimal Context over Documents
筛选出部分句子，再放入模型中生成答案，减少训练时间
From Answer Extraction to Answer Generation for Machine Reading Comprehension
先抽取答案所在的段落，再生成答案
Evidence Sentence Extraction for Machine Reading Comprehension
找出支持正确答案的句子
Label Distribution Augmented Maximum Likelihood Estimation for Reading Comprehension
选出阅读理解答案所在的文章内容片段起始和结束点。更改loss使答案更好

• 阅读理解需要的文章和内容知识

How Much Reading Does Reading Comprehension Require? A Critical Investigation of Popular Benchmarks
要回答问题需要结合多少文章和问题的信息，只提供问题或者段落，对与回答问题的准确度会造成多大的影响。
Assessing the Benchmarking Capacity of Machine Reading Comprehension Datasets
仅含内容词(非停用词)、混排句子词和混排句子顺序的问题的相对表现平均为原始表现的80%以上，表明这些问题可能不足以评估语法和复杂推理。
在模型能够正确回答的情况下，改变句子顺序，等对于结果的影响变化比对人的判断的影响小
MRQA 2019 Shared Task: Evaluating Generalization in Reading Comprehension
18个不同的数据集组合成相同的格式，6个用来训练，6个用来验证，6个用来测试。考量不同的数据集下的泛化效果

• 阅读理解如何作出正确答案

GenNet : Reading Comprehension with Multiple Choice Questions using Generation and Selection model
先生成一个答案，再从选项中选择最接近的一个
Explaining Question Answering Models through Text Generation
先生成一个文本，根据这个文本去选择正确答案
Joint Training of Candidate Extraction and Answer Selection for Reading Comprehension
先根据文章段落，选出topk可能答案集合。再对其整合得到最终的正确答案
ElimiNet: A Model for Eliminating Options for Reading Comprehension with Multiple Choice Questions
消去最不适合的选项
Option Comparison Network for Multiple-choice Reading Comprehension
比较选项之间的相关性，进而选择正确答案
Multi-Perspective Context Aggregation for Semi-supervised Cloze-style Reading Comprehension
四个选项和question聚合模块获取信息，最终输出一个正确答案
Retrospective Reader for Machine Reading Comprehension
提出了一种改进的模型，识别出没有办法回答的问题

• 生成问题或者干扰项

Generating Distractors for Reading Comprehension Questions from Real Examinations
尽量生成长的干扰项，动态注意和静态注意机制
Difficulty Controllable Generation of Reading Comprehension Questions
生成不同难度的阅读理解问题
使用两个模型回答问题，都能回答的标为简单，都回答不了标为难，其他不用做数据集
给定一个句子，和它的片段。去生成问题。这个片段当作问题的答案
与人打分的相关性作为评估标准
Neural Models for Key Phrase Detection and Question Generation
根据文章提问题
Question Generation for Reading Comprehension of Language Learning Test : -A Method using Seq2Seq Approach with Transformer Model

阅读理解数据集

ReClor: A Reading Comprehension Dataset Requiring Logical Reasoning
从开放网站中获得的研究生入学考试的理解题
ReCO: A Large Scale Chinese Reading Comprehension Dataset on Opinion
对每一个问题，有yes/no/uncertain三个答案。众包完成数据收集
Molweni: A Challenge Multiparty Dialogues-based Machine Reading Comprehension Dataset with Discourse Structure
多人多轮对话，问题
JEC-QA: A Legal-Domain Question Answering Dataset
多选项问题。来自中国国家司法考试
Triviaqa: A large scale distantly supervised challenge dataset for reading comprehension
问题，答案，文章
Dureader: a chinese machine reading comprehension dataset from real-world applications
来自百度，数据量多。问题，描述，正确答案
Ms marco: A human generated machine reading comprehension dataset
来自bing搜索的日志。问题，passage,回答，最好的回答，documents,问题类型
CoQA: A Conversational Question Answering Challenge
文章，问题s。答案
Natural Questions: A Benchmark for Question Answering Research
文章，问题。答案
Looking Beyond the Surface: A Challenge Set for Reading Comprehension over Multiple Sentences
文章，问题，多正确答案
Drcd: a chinese machine reading comprehension dataset
文章，问题s。答案
Docred: A large-scale document-level relation extraction dataset
文章级别的关系提取
A span-extraction dataset for Chinese Machine Reading Comprehension
文章，问题，答案区间
DuoRC: Towards Complex Language Understanding with Paraphrased Reading Comprehension
电影情节，问题，答案
LearningQ: A Large-Scale Dataset for Educational Question Generation
教育类的问题
Cosmos QA: Machine Reading Comprehension with Contextual Commonsense Reasoning
文章，问题，答案