semrep

---

PROJECT TITLE

Factuality check of the SemRep Predications

---

Description

The project deals with a Transformer based Language Model to filter predications belonging to the following subset of predicates from SemMedDB, informally called the "Substance Interactions" group:

• INTERACTS_WITH
• STIMULATES
• INHIBITS

---

AUTHOR INFORMATION

Md Rakibul Islam Prince Graduate Research Assistant Department of Electrical and Computer Engineering Indiana University-Purdue University Indianapolis Email: [email protected]

---

PACKAGE INSTALLATION

To reproduse the results at first all the necessarry packages are needed to be installed. the "semrepenv.yml" YAML file encapsulates the conda environment I used.

run

conda env create -f semrepenv.yml
conda activate semrepenv
or,
pip install -r requirements.txt

to install the environment before running any scripts or notebook. Or, you can manually install the packages from the "requirements.txt" file

---

PROJECT STRUCTURE AND DESCRIPTIONS

/semrep
├── /data
│ ├── substance_interactions.csv
│ └── substance_interactions_cleaned.csv
├── /logs
│ ├── bert_logfile.log
│ ├── biobert_logfile.log
│ └── ...
├── /models
│ ├── semrep_simple_bert_model
│ ├── semrep_simple_biobert_model
│ └── ...
├── /plots
│ ├── bert_cat_arg_dis_impact_all.png
│ ├── bert_cat_arg_dis_impact_verbal.png
│ ├── bert_cum_arg_dis_impact_all.png
│ ├── bert_cum_arg_dis_impact_verbal.png
│ ├── bert_precision_recall_curve_all.png
│ ├── bert_precision_recall_curve_verbal.png
│ ├── bert_roc_curve.png
│ ├── bert_sub_obj_heatmap_all.png
│ ├── bert_sub_obj_heatmap_verbal.png
│ └── ...
├── /results
│ ├── bert_test_set_0_results.csv
│ ├── val_bert_results.csv
│ ├── test_bert_results.csv
│ └── ...
├── /src
│ ├── semrep_model.ipynb
│ └── utils.py
├── README.txt
├── requirements.txt
└── semrepenv.yml

Below is an overview of the key files and folders in this project:

• `data/': Directory where the raw and processed data files are stored.

• `data/substance_interactions.csv': raw data file

• `data/substance_interactions_cleaned.csv': processed and clean data file

• logs/: Directory containing the logs for each model.

• logs/<model_name>_logfile.log: logfile for model <model_name>

• models/: Directory containing the finetuned checkpoints of the models.

• plots/: Directory containing all the generated plots during analysis.

• results/: Directory where the test and validation results are installed.

• src/: Directory containing the model notebooks and scripts.

• src/semrep_model.ipynb: Notebook detailing the full implimentation of the project

• src/utils.py: Scripts used for data analysis visualization tasks

• `README.txt': File detailing the description of the codebase.

• `requirements.txt': File detailing necessarry packages.

• `semrepenv.yml': File for recreating the environment.