Paper

Noise Reduction Methods for Distantly Supervised Biomedical Relation Extraction

Please cite the paper if you use the codes:

@INPROCEEDINGS{Li2017-zu,
  title           = "Noise Reduction Methods for Distantly Supervised
                     Biomedical Relation Extraction",
  booktitle       = "Proceedings of the {BioNLP} 2017 workshop",
  author          = "Li, Gang and Wu, Cathy H and Vijay-Shanker, K",
  pages           = "184--193",
  year            =  2017,
  conference      = "BioNLP 2017"
}

Requirements

• python 2.7 (should work with 3.6 but not tested)
• bash terminal

Setup

• Download data: wget http://biotm.cis.udel.edu/biotm/projects/ds/data.tar.gz
• Uncompress data folder: tar -zxvf data.tar.gz
• Create a python virtual environment: virtualenv env
• Activate virtual environment: . env/bin/activate
• Install python modules: pip install -r python_modules.txt

Run experiments

This is optional as all the result files are already included in this repo. Run the experiments if you want to regenerate them.

First activate the virtual environment: . env/bin/activate

Baseline and all the heuristics

sh script/single_instance.sh

Multi-instance learning baseline

sh script/multi_instance.sh

Evaluation results and figures

Figures are stored in eval/figures

Compute scores in Table 5

python src/compute_scores.py

Draw Figure 2

python src/draw_scores.py

Draw Figure 3

python src/draw_curves.py