PA-LSTM-relation-extraction

Implementation of Position-aware Attention and Supervised Data Improve Slot Filling.

Environment Requirements

• python 3.6
• pytorch 1.3.0

Data

• TACRED
• glove.6B.200d.txt

Usage

1. Purchase the dataset and move it in the data folder.
2. Download the embedding and move it in the embedding folder. The download script is available in the embedding folder.
3. Run the following the commands to start the program.

python run.py

More details can be seen by python run.py -h.

Result

The result of my version and that in paper are present as follows:

paper	my version
0.651	0.6503

The training log can be seen in train.log.

Note:

• Some settings are different from those mentioned in the paper.
• If you cannot achieve the same result as mine, please check the version of Pytorch.
• Some errors exists in my code, but on the whole it is right.

Reference Link

• https://github.com/yuhaozhang/tacred-relation