This project is a collaborative project between Emory NLP and Real Life Science, which develops annotation guidelines and address automatic extraction of medical entities (e.g., ‘Patient Condition’ and ‘Procedure’) and their relations in disease-related social media data. The paper will be published in Proceedings of the EMNLP Workshop on Health Text Mining and Information Analysis/2022.
The model employed in this project is adapted from this paper by @lxucs and @jdchoi77. Since our annotation does not include coreference, we skip the coreference evaluation part in the original model.
To use the trained model, configure the experiment settings in experiments_new.conf file and use the following commands to predict:
python run_med.py [config_name] [model_suffix] [gpu_id]
The results will be saved under 'extraction/[config_name]' folder.
To view and analyze the model results, configure the settings in postprocess.conf. And run the resultviewer.py file.