An open-source framework for neural relation extraction implemented in PyTorch.
Contributed by Shulin Cao, Tianyu Gao, Xu Han, Lumin Tang, Yankai Lin, Zhiyuan Liu
It is a PyTorch-based framwork for easily building relation extraction models. We divide the pipeline of relation extraction into four parts, which are embedding, encoder, selector and classifier. For each part we have implemented several methods.
- Embedding
- Word embedding
- Position embedding
- Concatenation method
- Encoder
- PCNN
- CNN
- Selector
- Attention
- Maximum
- Average
- Classifier
- Softmax loss function
- Output
All those methods could be combined freely.
We also provide fast training and testing codes. You could change hyper-parameters or appoint model architectures by using Python arguments. A plotting method is also in the package.
This project is under MIT license.
- Python (>=2.7)
- PyTorch (==0.3.1)
- CUDA (>=8.0)
- Matplotlib (>=2.0.0)
- scikit-learn (>=0.18)
- Install PyTorch
- Clone the OpenNRE repository:
git clone https://github.com/ShulinCao/OpenNRE-PyTorch
- Download NYT dataset from Google Drive
- Extract dataset to
./raw_data
unzip raw_data.zip
NYT10 is a distantly supervised dataset originally released by the paper "Sebastian Riedel, Limin Yao, and Andrew McCallum. Modeling relations and their mentions without labeled text.". Here is the download link for the original data. You can download the NYT10 dataset from Google Drive. And the data details are as follows.
Training data file and testing data file, containing sentences and their corresponding entity pairs and relations, should be in the following format
[
{
'sentence': 'Bill Gates is the founder of Microsoft .',
'head': {'word': 'Bill Gates', 'id': 'm.03_3d', ...(other information)},
'tail': {'word': 'Microsoft', 'id': 'm.07dfk', ...(other information)},
'relation': 'founder'
},
...
]
IMPORTANT: In the sentence part, words and punctuations should be separated by blank spaces.
Word embedding data is used to initialize word embedding in the networks, and should be in the following format
[
{'word': 'the', 'vec': [0.418, 0.24968, ...]},
{'word': ',', 'vec': [0.013441, 0.23682, ...]},
...
]
This file indicates corresponding IDs for relations to make sure during each training and testing period, the same ID means the same relation. Its format is as follows
{
'NA': 0,
'relation_1': 1,
'relation_2': 2,
...
}
IMPORTANT: Make sure the ID of NA
is always 0.
python gen_data.py
The processed data will be stored in ./data
python train.py --model_name pcnn_att
The arg model_name
appoints model architecture, and pcnn_att
is the name of one of our models. All available models are in ./model
. About other arguments please refer to ./train.py
. Once you start training, all checkpoints are stored in ./checkpoint
.
python test.py --model_name pcnn_att
Same usage as training. When finishing testing, the best checkpoint's corresponding pr-curve data will be stored in ./test_result
.
python draw_plot.py PCNN_ATT
The plot will be saved as ./test_result/pr_curve.png
. You could appoint several models in the arguments, like python draw_plot.py PCNN_ATT PCNN_ONE PCNN_AVE
, as long as there are these models' results in ./test_result
.
Not only could you train and test existing models in our package, you could also build your own model or add methods to the four basic modules. When adding a new model, you could create a python file in ./models
having the same name as the model and implement it like following:
import torch
import torch.autograd as autograd
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.autograd import Variable
from networks.embedding import *
from networks.encoder import *
from networks.selector import *
from networks.classifier import *
from .Model import Model
class PCNN_ATT(Model):
def __init__(self, config):
super(PCNN_ATT, self).__init__(config)
self.encoder = PCNN(config)
self.selector = Attention(config, config.hidden_size * 3)
Then you can train, test and plot!