UPDATE: This paper is part of the Multiple Relation Extraction task. More information can be found in Multiple Relational Facts Extraction.

This code is for ACL2018 paper "Extracting Relational Facts by an End-to-End Neural Model with Copy Mechanism"

Environment

• python2.7
• requirements.txt

Data

You need to modify the data path in const.py before running the code. The pre-processed data is released.

WebNLG:

• dataset
• pre-trained word embedding

NYT:

• dataset
• origin dataset
• pre-trained word embedding

Or, you can download them (without pre-trained word embedding) in Baidu SkyDrive with code "y286".

Train

python main.py -c config.json -t 0 -cell lstm

The cell can be "gru" or "lstm". You can specify the GPU card number by "-g". For exampe, "python main.py -c config.json -t 0 -cell lstm -g 0".

Test or Valid

You need to set the epochs of the model in main.py first. Then run the following commands: python main.py -c config.json -t 1 -cell lstm

"-t 1" means test and "-t 2" means valid.

Data process

The data_process can help you understand how did we process the data.