This is a tf1.1.3 based repo which use PCNN+ONE to solve Relation Extraction problems. Most of the ideas borrow from OpenNRE.
NOTICE: This is actually PCNN+MAX instead of ONE, and this repo is archived and will not be updated.
Well, I have to say, Keras is a great framework and easy to use in model stacking. But, it is not enough flexible to some self-design and research project. Maybe it is because I have not got the spirit of Keras... So I decide to use tensorflow instead of keras in this repo. You may see the former commits contains keras code. But I have changed the code into tf already in the last commit.
This repo is not support for gpu yet, but I'm working on it, since I don't have a gpu yet...
I am a newbie and still learning, so feel free to raise some issues and make pull requests.
- download the data and organize data in a proper file dir structure
python singlefile.pytensorboard --logdir ./summary/pcnn_one_train_demo/
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.
We've provided a toolkit to convert the original NYT10 data into JSON format that
OpenNREcould use. You could download the original data + toolkit from Google Drive or Tsinghua Cloud. Further instructions are included in the toolkit.
- Zeng et al. Distant Supervision for Relation Extraction via Piecewise Convolutional Neural Network. 2015
- Zeng et al. Relation Classification via Convolutional Deep Neural Network. 2014