Attention-CNN-relation-extraction

Implementation of Attention-Based Convolutional Neural Network for Semantic Relation Extraction.

Environment Requirements

• python 3.6
• pytorch 1.3.0

Data

• SemEval2010 Task8 [paper]
• Google News - Mikolov et al.(2010) [paper]

Usage

1. Download the embedding in the embedding folder and use convert.py to convert it to the UTF-8 format.
2. Run the following the commands to start the program.

python run.py

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

3. You can use the official scorer to check the final predicted result.

perl semeval2010_task8_scorer-v1.2.pl proposed_answer.txt predicted_result.txt >> result.txt

Result

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

paper	my version
0.843	0.8156

The training log can be seen in train.log and the official evaluation results is available in result.txt.

Note:

• Some settings are different from those mentioned in the paper.
• No validation set used during training.
• Just complete the part of general Attention-CNN. WordNet and words around nominals are not used. More details are available in Section 4 in this paper.
• Although I try to set random seeds, it seems that the results of each run are a little different.
• The result of my version is not ideal. Maybe my understanding is wrong. If you find it, please let me know.

Reference Link