nlp_research

介绍

本项目支持的NLP任务包括 分类、匹配、序列标注、文本生成等.
- 对于分类任务，目前支持多分类、多标签分类，通过选择不同的loss即可。
- 对于匹配任务，目前已支持交互模型和表示模型。
- 对于NER任务，目前已支持rnn+crf,idcnn+crf以及bert+crf

数据

训练数据(目前data下均内置了样例数据):
（1）对于分类任务的数据使用csv格式，csv头部包括列名‘target’和‘text’;
（2）对于匹配任务的数据使用csv格式，csv头部包括列名‘target’,‘text’ 或者 ‘target’,‘text_a’,‘text_b’
（3）对于NER任务的数据，参考"data/ner/train_data",或者使用其它格式的数据的话，修改task/ner.py中的read_data方法即可。
预训练数据(目前在分类和匹配任务上已支持):
- 如果使用到bert作为预训练(直接下载google训练好的模型即可)，直接运行"sh scripts/prepare.sh"
- 如果使用elmo作为预训练，需要准备一份corpus.txt训练语料放在language_model/bilm_tf/data/目录下
      然后执行指令进行预训练： 
            cd language_model/bilm_tf
            sh start.sh

快速开始

[依赖]
     环境：python3+tensorflow 1.10(python2.7已支持)
     pip3 install --user -r requirements.txt
     
各类任务的参数定义在conf/model/内的以任务名命名的yml文件中"conf/model/***.yml"
目前已支持的常见任务如下：       
[分类]
     1.生成tfrecords数据，训练:
        python3 run.py classify.yml mode=train
       或者直接使用脚本:
        sh scripts/restart.sh classify.yml
     
     2.测试：
       单个测试：python3 run.py classify.yml model=test_one
[匹配]
     1.生成tfrecords数据，训练:
         python3 run.py match.yml mode=train
       或者直接使用脚本:
         sh scripts/restart.sh match.yml
     2.测试：
        单个测试：python3 run.py match.yml model=test_one
[序列标注]
    ...
    sh scripts/restart.sh ner.yml
[翻译]    
    ...
    sh scripts/restart.sh translation.yml

模块

1. encoder
    cnn
    fasttext
    text_cnn
    dcnn
    idcnn
    dpcnn
    vdcnn
    rnn        
    rcnn
    attention_rnn
    capsule
    esim
    han
    matchpyramid
    abcnn
    transformer

2. common 
    loss
    attention
    lr
    ...

3. utils
    data process

参考文献

1. [2014 dcnn]A Convolutional Neural Network for Modelling Sentences
2. [2014 textcnn] Convolutional Neural Networks for Sentence Classification
3. [2015 charCNN] Character-level Convolutional Networks for TextClassification
4. [2016 HAN] Hierarchical Attention Networks for Document Classification
5. [2016-fasttext]Bag of Tricks for Efficient Text Classification
6. [2017 vdcnn] Very Deep Convolutional Networks for Text Classification
7. [2017_ACL dpcnn] Deep Pyramid Convolutional Neural Networks for Text Categorization
8. [2018] Investigating Capsule Networks with Dynamic Routing for Text Classification
9. [2018_ACL] Disconnected Recurrent Neural Networks for Text Categorization
10.[2018] Investigating Capsule Networks with Dynamic Routing for Text Classification
11.[2018] Topic Memory Networks for Short Text Classification
12.Learning Deep Structured Semantic Models for Web Search using Clickthrough Data
13.[2016] A Deep Relevance Matching Model for Ad-hoc Retrieval
14.[2016] ABCNN Attention-Based Convolutional Neural Network for Modeling Sentence Pairs
15.[2016] Person Re-Identification by Multi-Channel Parts-Based CNN with Improved Triplet Loss Function
16.[2016] Text Matching as Image Recognition
17.[2017 ACL,smn] Sequential Matching Network A New Architecture for Multi-turn Response Selection in Retrieval-based Chatbots
18.[2017 bimpm] Bilateral Multi-Perspective Matching for Natural Language Sentences
19.[2017 esim] Enhanced LSTM for Natural Language Inference
20.[2017] IRGAN A Minimax Game for Unifying Generative and Discriminative Information Retrieval Models
21.[2018] MIX Multi-Channel Information Crossing for Text Matching
22.[2017-NIPS] Attention-is-all-you-need
23.[2018-AAAI] DiSAN Directional Self-Attention Network for RNNCNN-Free Language Understanding
24.[2018-ICLR]Bi-Directional Block Self-Attention for Fast and Memory-Efficient Sequence Modeling
25.[2018] An Introductory Survey on Attention Mechanisms in NLP Problems
26.[2018] Universal Transformers
27.[2018 naacl ELMo]  Deep contextualized word representations
28.[2018 iclr quick-thoughts]An efficient framework for learning sentence representations
29.[2017 subword] EnrichingWord Vectors with Subword Information
30.[2018] Universal Language Model Fine-tuning for Text Classification

联系

如果有任何问题，欢迎发邮件到[email protected]
如果觉得我的工作对您有所帮助，请不要吝啬右上角的小星星哦！
欢迎Watch/Fork/Star！也欢迎一起建设这个项目！