搜狗“机器阅读理解工具包”及相关论文

  • A+
所属分类:医信大学

开源地址:https://github.com/sogou/SMRCToolkit

The Sogou Machine Reading Comprehension (SMRC) toolkit was designed for the fast and efficient development of modern machine comprehension models, including both published models and original prototypes.

avatar

SMRC 主要针对抽取式机器阅读理解任务,提供了包含数据集读取、预处理、神经网络模型基件、训练和评估的完整框架,同时对已发表的数十种机器阅读理解模型进行了复现和整合,用以帮助使用者快速、高效地重现已有模型或开发自己的阅读理解模型。

该工具集合也是目前业内最全的TensorFlow版本的阅读理解工具集合。目前与SMRC类似的工具集有 Allen AI的AllenNLP和UCL机器阅读理解组的Jack The Reader等。但AllenNLP和Jack The Reader更多的是提供适用于多个自然语言处理任务的框架,其中仅包含少量阅读理解模型(例如AllenNLP中针对SQuAD数据集的模型只有BiDAF和QANet)。SMRC则更多地深挖阅读理解任务,在建立一套模型构建和训练框架的同时,提供了10+ 个模型的实现。

机器阅读理解

经典数据集,斯坦福 SQuAD :https://rajpurkar.github.io/SQuAD-explorer/

SQuAD 是斯坦福大学于2016年推出的数据集,一个阅读理解数据集,给定一篇文章,准备相应问题,需要算法给出问题的答案。此数据集所有文章选自维基百科,数据集的量为当今其他数据集(例如,WikiQA)的几十倍之多。一共有107,785问题,以及配套的 536 篇文章。数据集的贡献者为斯坦福Percy Liang等人,Percy Liang是自然语言处理界的一位全才,在Semantic Parsing, QA, Optimization等多个领域都有重要贡献。 SQuAD通过众包的方式,从wikipedia上的536篇文章提取超过10w个问题-答案对,且其中的答案是原文的一个片段而不是单一的实体对象
该数据集包含 10 万个(问题,原文,答案)三元组,原文来自于 536 篇维基百科文章,而问题和答案的构建主要是通过众包的方式,让标注人员提出最多 5 个基于文章内容的问题并提供正确答案,且答案出现在原文中。SQuAD 和之前的完形填空类阅读理解数据集如 CNN/DM[2],CBT[3]等最大的区别在于:SQuAD 中的答案不在是单个实体或单词,而可能是一段短语,这使得其答案更难预测。SQuAD 包含公开的训练集和开发集,以及一个隐藏的测试集,其采用了与 ImageNet 类似的封闭评测的方式,研究人员需提交算法到一个开放平台,并由 SQuAD 官方人员进行测试并公布结果。

机器阅读理解主要内容:
(1)词性的标注(动词,名词,形容词等。。)
(2)分词
(3)信息的提取
(4)词性的匹配


清华大学推荐的阅读理解论文列表(https://github.com/thunlp/RCPapers)

Must-read papers on Machine Reading Comprehension.Contributed by Yankai Lin, Deming Ye and Haozhe Ji.

Model Architecture

  1. Memory networks. Jason Weston, Sumit Chopra, and Antoine Bordes. arXiv preprint arXiv:1410.3916 (2014). paper
  2. Teaching Machines to Read and Comprehend. Karl Moritz Hermann, Tomáš Kočiský, Edward Grefenstette, Lasse Espeholt, Will Kay, Mustafa Suleyman, and Phil Blunsom. NIPS 2015. paper
  3. Text Understanding with the Attention Sum Reader Network. Rudolf Kadlec, Martin Schmid, Ondrej Bajgar, and Jan Kleindienst. ACL 2016. paper
  4. A Thorough Examination of the Cnn/Daily Mail Reading Comprehension Task. Danqi Chen, Jason Bolton, and Christopher D. Manning. ACL 2016. paper
  5. Long Short-Term Memory-Networks for Machine Reading. Jianpeng Cheng, Li Dong, and Mirella Lapata. EMNLP 2016. paper
  6. Key-value Memory Networks for Directly Reading Documents. Alexander Miller, Adam Fisch, Jesse Dodge, Amir-Hossein Karimi, Antoine Bordes, and Jason Weston. EMNLP 2016. paper
  7. Modeling Human Reading with Neural Attention. Michael Hahn and Frank Keller. EMNLP 2016. paper
  8. Learning Recurrent Span Representations for Extractive Question Answering Kenton Lee, Shimi Salant, Tom Kwiatkowski, Ankur Parikh, Dipanjan Das, and Jonathan Berant. arXiv preprint arXiv:1611.01436 (2016). paper
  9. Multi-Perspective Context Matching for Machine Comprehension. Zhiguo Wang, Haitao Mi, Wael Hamza, and Radu Florian. arXiv preprint arXiv:1612.04211. paper
  10. Natural Language Comprehension with the Epireader. Adam Trischler, Zheng Ye, Xingdi Yuan, and Kaheer Suleman. EMNLP 2016. paper
  11. Iterative Alternating Neural Attention for Machine Reading. Alessandro Sordoni, Philip Bachman, Adam Trischler, and Yoshua Bengio. arXiv preprint arXiv:1606.02245 (2016). paper
  12. Bidirectional Attention Flow for Machine Comprehension. Minjoon Seo, Aniruddha Kembhavi, Ali Farhadi, and Hannaneh Hajishirzi. ICLR 2017. paper
  13. Machine Comprehension Using Match-lstm and Answer Pointer. Shuohang Wang and Jing Jiang. arXiv preprint arXiv:1608.07905 (2016). paper
  14. Gated Self-matching Networks for Reading Comprehension and Question Answering. Wenhui Wang, Nan Yang, Furu Wei, Baobao Chang, and Ming Zhou. ACL 2017. paper
  15. Attention-over-attention Neural Networks for Reading Comprehension. Yiming Cui, Zhipeng Chen, Si Wei, Shijin Wang, Ting Liu, and Guoping Hu. ACL 2017. paper
  16. Gated-attention Readers for Text Comprehension. Bhuwan Dhingra, Hanxiao Liu, Zhilin Yang, William W. Cohen, and Ruslan Salakhutdinov. ACL 2017. paper
  17. A Constituent-Centric Neural Architecture for Reading Comprehension. Pengtao Xie and Eric Xing. ACL 2017. paper
  18. Structural Embedding of Syntactic Trees for Machine Comprehension. Rui Liu, Junjie Hu, Wei Wei, Zi Yang, and Eric Nyberg. EMNLP 2017. paper
  19. Accurate Supervised and Semi-Supervised Machine Reading for Long Documents. Izzeddin Gur, Daniel Hewlett, Alexandre Lacoste, and Llion Jones. EMNLP 2017. paper
  20. MEMEN: Multi-layer Embedding with Memory Networks for Machine Comprehension. Boyuan Pan, Hao Li, Zhou Zhao, Bin Cao, Deng Cai, and Xiaofei He. arXiv preprint arXiv:1707.09098 (2017). paper
  21. Dynamic Coattention Networks For Question Answering. Caiming Xiong, Victor Zhong, and Richard Socher. ICLR 2017 paper
  22. R-NET: Machine Reading Comprehension with Self-matching Networks. Natural Language Computing Group, Microsoft Research Asia. paper
  23. Reasonet: Learning to Stop Reading in Machine Comprehension. Yelong Shen, Po-Sen Huang, Jianfeng Gao, and Weizhu Chen. KDD 2017. paper
  24. FusionNet: Fusing via Fully-Aware Attention with Application to Machine Comprehension. Hsin-Yuan Huang, Chenguang Zhu, Yelong Shen, and Weizhu Chen. ICLR 2018. paper
  25. Making Neural QA as Simple as Possible but not Simpler. Dirk Weissenborn, Georg Wiese, and Laura Seiffe. CoNLL 2017. paper
  26. Efficient and Robust Question Answering from Minimal Context over Documents. Sewon Min, Victor Zhong, Richard Socher, and Caiming Xiong. ACL 2018. paper
  27. Simple and Effective Multi-Paragraph Reading Comprehension. Christopher Clark and Matt Gardner. ACL 2018. paper
  28. Neural Speed Reading via Skim-RNN. Minjoon Seo, Sewon Min, Ali Farhadi, and Hannaneh Hajishirzi. ICLR2018. paper
  29. Hierarchical Attention Flow forMultiple-Choice Reading Comprehension. Haichao Zhu,� Furu Wei, Bing Qin, and Ting Liu. AAAI 2018. paper
  30. Towards Reading Comprehension for Long Documents. Yuanxing Zhang, Yangbin Zhang, Kaigui Bian, and Xiaoming Li. IJCAI 2018. paper
  31. Joint Training of Candidate Extraction and Answer Selection for Reading Comprehension. Zhen Wang, Jiachen Liu, Xinyan Xiao, Yajuan Lyu, and Tian Wu. ACL 2018. paper
  32. Multi-Passage Machine Reading Comprehension with Cross-Passage Answer Verification. Yizhong Wang, Kai Liu, Jing Liu, Wei He, Yajuan Lyu, Hua Wu, Sujian Li, and Haifeng Wang. ACL 2018. paper
  33. Reinforced Mnemonic Reader for Machine Reading Comprehension. Minghao Hu, Yuxing Peng, Zhen Huang, Xipeng Qiu, Furu Wei, and Ming Zhou. IJCAI 2018. paper
  34. Stochastic Answer Networks for Machine Reading Comprehension. Xiaodong Liu, Yelong Shen, Kevin Duh, and Jianfeng Gao. ACL 2018. paper
  35. Multi-Granularity Hierarchical Attention Fusion Networks for Reading Comprehension and Question Answering. Wei Wang, Ming Yan, and Chen Wu. ACL 2018. paper
  36. A Multi-Stage Memory Augmented Neural Networkfor Machine Reading Comprehension. Seunghak Yu, Sathish Indurthi, Seohyun Back, and Haejun Lee. ACL 2018 workshop. paper
  37. S-NET: From Answer Extraction to Answer Generation for Machine Reading Comprehension. Chuanqi Tan, Furu Wei, Nan Yang, Bowen Du, Weifeng Lv, and Ming Zhou. AAAI2018. paper
  38. Ask the Right Questions: Active Question Reformulation with Reinforcement Learning. Christian Buck, Jannis Bulian, Massimiliano Ciaramita, Wojciech Gajewski, Andrea Gesmundo, Neil Houlsby, and Wei Wang. ICLR2018. paper
  39. QANet: Combining Local Convolution with Global Self-Attention for Reading Comprehension. Adams Wei Yu, David Dohan, Minh-Thang Luong, Rui Zhao, Kai Chen, Mohammad Norouzi, and Quoc V. Le. ICLR2018. paper
  40. Read + Verify: Machine Reading Comprehension with Unanswerable Questions. Minghao Hu, Furu Wei, Yuxing Peng, Zhen Huang, Nan Yang, and Ming Zhou. AAAI2019. paper

Utilizing External Knowledge

  1. Leveraging Knowledge Bases in LSTMs for Improving Machine Reading. Bishan Yang and Tom Mitchell. ACL 2017. paper
  2. Learned in Translation: Contextualized Word Vectors. Bryan McCann, James Bradbury, Caiming Xiong, and Richard Socher. arXiv preprint arXiv:1708.00107 (2017). paper
  3. Knowledgeable Reader: Enhancing Cloze-Style Reading Comprehension with External Commonsense Knowledge.Todor Mihaylov and Anette Frank. ACL 2018. paper
  4. A Comparative Study of Word Embeddings for Reading Comprehension. Bhuwan Dhingra, Hanxiao Liu, Ruslan Salakhutdinov, and William W. Cohen. arXiv preprint arXiv:1703.00993 (2017). paper
  5. Deep contextualized word representations. Matthew E. Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. NAACL 2018. paper
  6. Improving Language Understanding by Generative Pre-Training. Alec Radford, Karthik Narasimhan, Tim Salimans, and Ilya Sutskever. OpenAI. paper
  7. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. arXiv preprint arXiv:1810.04805 (2018). paper

Exploration

  1. Adversarial Examples for Evaluating Reading Comprehension Systems. Robin Jia, and Percy Liang. EMNLP 2017. paper
  2. Did the Model Understand the Question? Pramod Kaushik Mudrakarta, Ankur Taly, Mukund Sundararajan, and Kedar Dhamdhere. ACL 2018. paper

Open Domain Question Answering

  1. Reading Wikipedia to Answer Open-Domain Questions. Danqi Chen, Adam Fisch, Jason Weston, and Antoine Bordes. ACL 2017. paper
  2. R^3: Reinforced Reader-Ranker for Open-Domain Question Answering. Shuohang Wang, Mo Yu, Xiaoxiao Guo, Zhiguo Wang, Tim Klinger, Wei Zhang, Shiyu Chang, Gerald Tesauro, Bowen Zhou, and Jing Jiang. AAAI 2018. paper
  3. Evidence Aggregation for Answer Re-Ranking in Open-Domain Question Answering. Shuohang Wang, Mo Yu, Jing Jiang, Wei Zhang, Xiaoxiao Guo, Shiyu Chang, Zhiguo Wang, Tim Klinger, Gerald Tesauro, and Murray Campbell. ICLR 2018. paper
  4. Denoising Distantly Supervised Open-Domain Question Answering. Yankai Lin, Haozhe Ji, Zhiyuan Liu, and Maosong Sun. ACL 2018. paper

Datasets

  1. (SQuAD 1.0) SQuAD: 100,000+ Questions for Machine Comprehension of Text. Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and Percy Liang. EMNLP 2016. paper
  2. (SQuAD 2.0) Know What You Don't Know: Unanswerable Questions for SQuAD. Pranav Rajpurkar, Robin Jia, and Percy Liang. ACL 2018. paper
  3. (MS MARCO) MS MARCO: A Human Generated MAchine Reading COmprehension Dataset. Tri Nguyen, Mir Rosenberg, Xia Song, Jianfeng Gao, Saurabh Tiwary, Rangan Majumder, and Li Deng. arXiv preprint arXiv:1611.09268 (2016). paper
  4. (Quasar) Quasar: Datasets for Question Answering by Search and Reading. Bhuwan Dhingra, Kathryn Mazaitis, and William W. Cohen. arXiv preprint arXiv:1707.03904 (2017). paper
  5. (TriviaQA) TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehension. Mandar Joshi, Eunsol Choi, Daniel S. Weld, Luke Zettlemoyer. arXiv preprint arXiv:1705.03551 (2017). paper
  6. (SearchQA) SearchQA: A New Q&A Dataset Augmented with Context from a Search Engine. Matthew Dunn, Levent Sagun, Mike Higgins, V. Ugur Guney, Volkan Cirik, and Kyunghyun Cho. arXiv preprint arXiv:1704.05179 (2017). paper
  7. (QuAC) QuAC : Question Answering in Context. Eunsol Choi, He He, Mohit Iyyer, Mark Yatskar, Wen-tau Yih, Yejin Choi, Percy Liang, and Luke Zettlemoyer. arXiv preprint arXiv:1808.07036 (2018). paper
  8. (CoQA) CoQA: A Conversational Question Answering Challenge. Siva Reddy, Danqi Chen, and Christopher D. Manning. arXiv preprint arXiv:1808.07042 (2018). paper
  9. (MCTest) MCTest: A Challenge Dataset for the Open-Domain Machine Comprehension of Text. Matthew Richardson, Christopher J.C. Burges, and Erin Renshaw. EMNLP 2013. paper.
  10. (CNN/Daily Mail) Teaching Machines to Read and Comprehend. Hermann, Karl Moritz, Tomas Kocisky, Edward Grefenstette, Lasse Espeholt, Will Kay, Mustafa Suleyman, and Phil Blunsom. NIPS 2015. paper
  11. (CBT) The Goldilocks Principle: Reading Children's Books with Explicit Memory Representations. Felix Hill, Antoine Bordes, Sumit Chopra, and Jason Weston. arXiv preprint arXiv:1511.02301 (2015). paper
  12. (bAbi) Towards AI-Complete Question Answering: A Set of Prerequisite Toy Tasks. Jason Weston, Antoine Bordes, Sumit Chopra, Alexander M. Rush, Bart van Merriënboer, Armand Joulin, and Tomas Mikolov. arXiv preprint arXiv:1502.05698 (2015). paper
  13. (LAMBADA) The LAMBADA Dataset:Word Prediction Requiring a Broad Discourse Context. Denis Paperno, Germ ́an Kruszewski, Angeliki Lazaridou, Quan Ngoc Pham, Raffaella Bernardi, Sandro Pezzelle, Marco Baroni, Gemma Boleda, and Raquel Fern ́andez. ACL 2016. paper
  14. (SCT) LSDSem 2017 Shared Task: The Story Cloze Test. Nasrin Mostafazadeh, Michael Roth, Annie Louis,Nathanael Chambers, and James F. Allen. ACL 2017 workshop. paper
  15. (Who did What) Who did What: A Large-Scale Person-Centered Cloze Dataset Takeshi Onishi, Hai Wang, Mohit Bansal, Kevin Gimpel, and David McAllester. EMNLP 2016. paper
  16. (NewsQA) NewsQA: A Machine Comprehension Dataset. Adam Trischler, Tong Wang, Xingdi Yuan, Justin Harris, Alessandro Sordoni, Philip Bachman, and Kaheer Suleman. arXiv preprint arXiv:1611.09830 (2016). paper
  17. (RACE) RACE: Large-scale ReAding Comprehension Dataset From Examinations. Guokun Lai, Qizhe Xie, Hanxiao Liu, Yiming Yang, and Eduard Hovy. EMNLP 2017. paper
  18. (ARC) Think you have Solved Question Answering?Try ARC, the AI2 Reasoning Challenge. Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot,Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord. arXiv preprint arXiv:1803.05457 (2018). paper
  19. (MCScript) MCScript: A Novel Dataset for Assessing Machine Comprehension Using Script Knowledge. Simon Ostermann, Ashutosh Modi, Michael Roth, Stefan Thater, and Manfred Pinkal. arXiv preprint arXiv:1803.05223. paper
  20. (NarrativeQA) The NarrativeQA Reading Comprehension Challenge. Tomáš Kočiský, Jonathan Schwarz, Phil Blunsom, Chris Dyer, Karl Moritz Hermann, Gábor Melis, and Edward Grefenstette. TACL 2018. paper
  21. (DuoRC) DuoRC: Towards Complex Language Understanding with Paraphrased Reading Comprehension. Amrita Saha, Rahul Aralikatte, Mitesh M. Khapra, and Karthik Sankaranarayanan. ACL 2018. paper
  22. (CLOTH) Large-scale Cloze Test Dataset Created by Teachers. Qizhe Xie, Guokun Lai, Zihang Dai, and Eduard Hovy. EMNLP 2018. paper
  23. (DuReader) DuReader: a Chinese Machine Reading Comprehension Dataset from Real-world Applications. Wei He, Kai Liu, Yajuan Lyu, Shiqi Zhao, Xinyan Xiao, Yuan Liu, Yizhong Wang, Hua Wu, Qiaoqiao She, Xuan Liu, Tian Wu, and Haifeng Wang. ACL 2018 Workshop. paper
  24. (CliCR) CliCR: a Dataset of Clinical Case Reports for Machine Reading Comprehension. Simon Suster and Walter Daelemans. NAACL 2018. paper

  • 机器阅读理解相关论文汇总
    • 综述
      • 1 Emergent Logical Structure in Vector Representations of Neural Readers
    • 论文
      • 1 Learning Recurrent Span Representations for Extractive Question Answering
      • 2 Answering Complicated Question Intents Expressed in Decomposed Question Sequences
      • 3 The Amazing Mysteries of the Gutter: Drawing Inferences Between Panels in Comic Book Narratives
      • 4 Broad Context Language Modeling as Reading Comprehension
      • 5 Easy Questions First? A Case Study on Curriculum Learning for Question Answering
      • 6 Machine Comprehension Using Match-LSTM and Answer Pointer
      • 7 Dataset and Neural Recurrent Sequence Labeling Model for Open-Domain Factoid Question Answering
    • 模型
      • 1 Memory Networks
        • 1.1 Memory Network
        • 1.2 End-To-End Memory Networks
        • 1.3 Ask Me Anything: Dynamic Memory Networks for Natural Language Processing
        • 1.4 Key-Value Memory Networks for Directly Reading Documents
        • 1.5 The Goldilocks Principle: Reading Children's Books with Explicit Memory Representations
        • 1.6 Can Active Memory Replace Attention?
      • 2 DeepMind Attentive Reader
        • 2.1 Teaching Machines to Read and Comprehend
      • 3 Danqi’s Stanford Reader
        • 3.1 [A thorough examination of the cnn/daily
      • 4 Attention Sum Reader
      • 5 Gated Attention Sum Reader
      • 6 Attention Over Attention Reader
        • 6.1 Attention-over-Attention Neural Networks for Reading Comprehension
    • 数据集
      • 1 SQuAD
      • 2 LAMBADA
      • 3 Who-did-What(WDW)
      • 4 CNN & DailyMail
      • 5 Children's Book Test
      • 6 BookTest

综述

Emergent Logical Structure in Vector Representations of Neural Readers

针对最近提出的各种各样的attention based reader models,本文作者做了一个比较全面的总结和分析,并且通过数学分析和实验展示了模型之间的相关性。

论文

Learning Recurrent Span Representations for Extractive Question Answering

【机器阅读】不同的阅读理解数据集产生答案的方式不同,有的是给定N个候选答案,有的是规定从原文中的entity中进行选择,有的是从原文中的任意token进行选择等等。本文所用的数据集是SQuAD,候选答案是原文中的任意字符串,难度较大,答案可能是一个词或者几个词都有可能。本文在前人研究的基础上提出了一种显式表示answer span的模型,取得了不错的效果。

Answering Complicated Question Intents Expressed in Decomposed Question Sequences

【复杂问答】基于语义分析的问答系统最近流行于解决长、难问题,本文研究的内容是如何处理多个相互关联的简单问题?(即将复杂问题分解成多个相关简答问题)并给出了一个任务数据集。这个问题的一大难点在于相互关联的问题需要共指消解的工作。本文将单轮问答对话分解成多轮问题过程,上下文的处理非常重要。建议研究聊天机器人的童鞋来精读此文。

The Amazing Mysteries of the Gutter: Drawing Inferences Between Panels in Comic Book Narratives

【问答系统】基于上下文的问答已经有很多数据集了,基于图像的问答也有一些数据集了。漫画是一类大家小时候都喜欢的读物,包含了丰富的图像和文本数据(对话)。本文给出了一个大型数据集,包括了丰富的图像和文本,规模在120万(120GB)左右。数据给出了几个任务,基于图像的问答任务,基于对话文本的问答任务和文本排序任务。对问答感兴趣,想找一些新数据来刷一刷榜的童鞋可以看过来。

Broad Context Language Modeling as Reading Comprehension

不久前发布的LAMBADA dataset中,作者尝试的各种baseline models都给出了比较差的结果。在观察了LAMBADA dataset之后,我们认为可以利用Reading comprehension models来提升准确率,而不必使用传统的language model。这篇文章中,作者利用了简单的方法和模型将LAMBADA dataset的准确率从7.3%提高到45.4%,非常简单有效。

Easy Questions First? A Case Study on Curriculum Learning for Question Answering

Machine Comprehension Using Match-LSTM and Answer Pointer

本文提出的模型结合了match-LSTM(mLSTM)和Pointer Net(Ptr-Net)两种网络结构。

Dataset and Neural Recurrent Sequence Labeling Model for Open-Domain Factoid Question Answering

作者给出了一个新的中文的QA数据集, 并且提出了一个非常有意思的baseline model.

模型

1 Memory Networks

1.1 Memory Network

Memory Networks为解决长期记忆问题, 提出一类称为Memory Networks的模型框架, 基于该框架构造的模型可以拥有长期(大量)和易于读写的记忆。

1.2 End-To-End Memory Networks

本文提出了一个可以端到端训练的Memory Networks,并且在训练阶段比原始的Memory Networks需要更少的监督信息。

1.3 Ask Me Anything: Dynamic Memory Networks for Natural Language Processing

Question Answering: 给定一段Context,一个与此Context相关的Question,利用模型生成一个单词的Answer。

1.4 Key-Value Memory Networks for Directly Reading Documents

鉴于知识库有知识稀疏、形式受限等问题,本文提出了一种可以通过直接读取文档来解决QA问题的新方法Key-Value Memory Networks。

1.5 The Goldilocks Principle: Reading Children's Books with Explicit Memory Representations

本文对于语言模型(RNN/LSTM/Memory Network生成)到底能够多好或者在多大程度上表示The Children’s Book做了一项测试。测试结果表面Memor Network上的效果最好。

1.6 Can Active Memory Replace Attention?

Memory Networks和Attention是解决长距离依赖问题的两大方法,Attention模型在NLP的很多任务中都有更好的表现,本文对Memory Networks类模型的缺点进行了分析,并且提出了一种改进模型。改进版的memory模型有不错的表现,并且在长句子机器翻译任务中得到了验证。本文作者来自Google Brain。建议关注自然语言处理的童鞋,不管是关注什么任务,都应该精读一下本文。

2 DeepMind Attentive Reader

2.1 Teaching Machines to Read and Comprehend

3 Danqi’s Stanford Reader

3.1 [A thorough examination of the cnn/daily

mail reading comprehension task]()

4 Attention Sum Reader

5 Gated Attention Sum Reader

6 Attention Over Attention Reader

6.1 Attention-over-Attention Neural Networks for Reading Comprehension

本文优化了attention机制,同时apply question-to-document and document-to-question attention,提升了已有模型在Cloze-Style Question Answering Task上的准确率。

数据集

SQuAD

SQuAD: 100,000+ Questions for Machine Comprehension of Text

LAMBADA

The LAMBADA dataset: Word prediction requiring a broad discourse context

Who-did-What(WDW)

Who did What: A Large-Scale Person-Centered Cloze Dataset

数据来自LDC English Gigaword newswire copus。该数据集为了防止文章摘要被使用,每一个问题都从两个独立的文章中生成,一篇用来做Context,一篇用来挖空作为问题。该数据集为了不像CNN&DailyMail那样将实体匿名,所有的问题都是人名实体。而且使用了一些简单的baselines来筛选掉那些容易解决的问题。

CNN & DailyMail

Teaching Machines to Read and Comprehend

数据来自CNN和Daily Mail新闻,文章中高亮显示而且挖空的就是问题。为了防止使用外界知识,将命名实体都用ID替换,给出答案候选集。

5 Children's Book Test

The goldilocks principle: Reading childrens books with explicit memory representations

数据来自一个儿童读物,每个问题都是从中挑选出21条连续的句子,将前20条作为Context,将第21条挖空作为问题。

6 BookTest

Embracing data abundance: BookTest Dataset for Reading Comprehension

本文发布了一个新的机器阅读理解数据集BookTest,该数据集最大的亮点是规模大,是Facebook发布的Children’s Book Test的60倍之大。

  • HealthIT.CN小程序
  • 产学研医创新平台
  • weinxin
  • 健康数据
  • HealthIT公众号
  • weinxin
admin

发表评论

:?: :razz: :sad: :evil: :!: :smile: :oops: :grin: :eek: :shock: :???: :cool: :lol: :mad: :twisted: :roll: :wink: :idea: :arrow: :neutral: :cry: :mrgreen: