DSG-SEG

KO | EN

gitlite — search

#javascript #hacktoberfest #ai #go #golang #rust #docker #macos #windows #cli #kubernetes #awesome-list #ios #open-source

(★ 11)

[2021] Topic Segmentation via Discourse Structure Graph Network

DSG-SEG Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# [《基于篇章结构图网络的话题分割》(Topic Segmentation via Discourse Structure Graph Network)](http://jcip.cipsc.org.cn/CN/Y2021/V35/I12/17)
该仓库包含了《基于篇章结构图网络的话题分割》文章的代码和数据集，代码使用pytorch框架实现。
我们的工作主要基于[《Text Segmentation as a Supervised Learning Task》](https://arxiv.org/abs/1803.09337) ，它的源代码是[text-segmentation](https://github.com/koomri/text-segmentation)。
另外，我们的模型灵感主要来自于 [《Every Document Owns Its Structure: Inductive Text Classification via
Graph Neural Networks》](https://arxiv.org/abs/2004.13826)([TextING的代码](https://github.com/CRIPAC-DIG/TextING)) 以及 [《Graph Convolutional Networks for Text Classification》](https://arxiv.org/abs/1809.05679)([TextGCN的代码](https://github.com/yao8839836/text_gcn))。
谢谢他们的工作。

# 论文简介
![image](https://user-images.githubusercontent.com/59757561/113157597-9ee79600-926d-11eb-9480-4c0f3214823b.png)

我们提出了一个基于篇章结构图网络的话题分割模型——DSG-SEG，如上图所示.具体的，模型把每一个篇章单独构建成图，图中包含了单词、句子节点，以及单词节点之间、单词和句子节点之间的邻接关系；接着，模型把图作为输入，对GGNN网络进行迭代，句子节点之间因此通过共有的邻接单词节点产生了间接的信息交互；最终，模型得到了具有全局语义信息的句子向量表示，并将其送入Bi-LSTM网络进行分割点的预测。

我们对比了六种基准模型:freq-sr\latent-sr\tree-sr-left\TextSeg\Bert+Bi-LSTM\Bert+Bi-LSTMnft（除TextSeg外，其它基准模型皆为我们复现）我们的模型在诸多数据集上同时取得了最好的结果指标和时间性能。

# 环境配置
python3.7
```bash
conda create -n textseg python=3.7
conda activate textseg
```
python依赖库
```bash
pip install -r requirements.txt
```

# 资源获取  
假设所有资源保存在/data路径下，完成这部分，你将得到:  
[查看文件结构](https://github.com/beiweixiaoxu/DSG-SEG/blob/master/doc/file_tree.md)

glove:  
wget http://nlp.stanford.edu/data/glove.6B.zip  
unzip glove.6B.zip  
rm glove.6B.(50d|100d|200d).txt  
  
word2vec:  
wget https://s3.amazonaws.com/dl4j-distribution/GoogleNews-vectors-negative300.bin.gz  
gzip -d GoogleNews-vectors-negative300.bin.gz  
解压完毕得到:GoogleNews-vectors-negative300.bin  
  
WIKI_727K:  
wget https://www.dropbox.com/sh/k3jh0fjbyr0gw0a/AADzAd9SDTrBnvs1qLCJY5cza?dl=0&preview=wiki_727K.tar.bz2  
tar -xjf wiki_727K.tar.bz2

wiki_test_50:  
wget https://www.dropbox.com/sh/k3jh0fjbyr0gw0a/AADzAd9SDTrBnvs1qLCJY5cza?dl=0&preview=wiki_test_50.tar.bz2  
tar -xjf wiki_test_50.tar.bz2

WIKI-SECTION:  
wget https://github.com/sebastianarnold/WikiSection/raw/master/wikisection_dataset_ref.tar.gz  
mkdir wikisection & tar -zxvf wikisection_dataset_ref.tar.gz -C wikisection  
  
CHOI:  
手动下载 https://github.com/koomri/text-segmentation/tree/master/data/choi  
mkdir choi  
放入文件夹  
  
MANIFESTO:  
手动下载 https://github.com/koomri/text-segmentation/tree/master/data/manifesto  
mkdir manifesto  
放入文件夹  
  
CLINICAL:  
手动下载 https://github.com/pinkeshbadjatiya/neuralTextSegmentation/tree/master/code/data/clinical  
mkdir clinical  
放入文件夹    
  
CITIES:  
wget http://groups.csail.mit.edu/rbg/code/mallows/data/wikicities-english.tar.gz  
tar -zxvf wikicities-english.tar.gz  
rm -r wikicities-english/test  
mv wikicities-english/training/* wikicities-english/  
rm -r wikicities-english/training  
mv wikicities-english wikicities

ELEMENTS:  
wget http://groups.csail.mit.edu/rbg/code/mallows/data/wikielements.tar.gz  
tar -zxvf wikielements.tar.gz

# 代码运行  
[查看代码结构](https://github.com/beiweixiaoxu/DSG-SEG/blob/master/doc/code_tree.md)  
两个修改:  
1. 永久修改parameters.py中，各个资源的路径，如`parser.add_argument("--wiki_path",type=str, default="/data/wiki_727")`  
2. 永久修改parameters.py中，保存dataset和encoder的路径(用于避免重复读取)，如`parser.add_argument("--dataset_dir",type=str, default="/data/saved_dataset")`  
样例:  
```python
python run.py [--etype {one_hot,glove,w2v,bert}] [--encoder_fine_tune]
              [--sr_choose {f_model,l_model,s_model,t_model,g_model,b_model,random_baseline}]
              [--tr_choose {balanced,left,right}]
              [--gr_choose {texting,DSE_SEG}]
              ...
```
DSG_GCN:
```python
python run.py --sr_choose=g_model --gr_choose=DSE_SEG
```
-SENTNODE:
```python
python run.py --sr_choose=g_model --gr_choose=texting
```
-PMI:
```python
python run.py --sr_choose=g_model --gr_choose=DSG_SEG --pmi
```
Bert+Bi-LSTMnft:
```python
python run.py --etype=bert --sr_choose=b_model --length_filter=64
```
TextSeg:
```python
python run.py --sr_choose=s_model
```
Bert+Bi-LSTM:
```python
python run.py --etype=bert --sr_choose=b_model --length_filter=40 --sent_num_filter=60 --train_bs=3 --encoder_fine_tune
```
tree-sr-left
```python
python run.py --sr_choose=t_model --tr_choose=left --length_filter=40 --sent_num_filter=60
```
freq-sr:
```python
python run.py --etype=one-hot --sr_choose=f_model
```
latent-sr:
```python
python run.py --etype=one-hot --sr_choose=l_model
```
random-baseline:
```python
python run.py --sr_choose=random_baseline --infer --train_bs=1 --dev_bs=1 --test_bs=1  
```

DSG-SEG

DSG-SEG Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

DSG-SEG

DSG-SEG Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README