README.md

Automatic semantic relation extraction

^{This is the main repository for the NLP project on FRI of the group "Vesoljci". The repository was created for a project that was part of Natural Language Processing course at the University of Ljubljana, Faculty for computer and information science.}

To use this repository as intended you should have a NVIDIA GPU with appropriate NVIDIA driver and CUDA versions that are compatible with PyTorch and Tensorflow.

Repository folder structure

.
├── assets                      
├── config_files                
├── data                        
│   ├── experiments             
|   |   ├── experiment_1            # folder with dataset setup split into train and test data
|   │   │   ├── model_1         
|   |   |   ├── model_2
|   |   |   ├── test.csv
|   |   |   └── train.csv
│   │   ├── experiment_2               
|   |   |   ├── model_1             # model configuration, saved model, results, generated graphs
|   |   |   |   ├── annotaion.csv
|   |   |   |   ├── config_dict.json
|   |   |   |   ├── model.pt
|   |   |   |   ├── results.txt              
|   |   |   |   └── graph.png
|   |   |   └── ...
|   |   └── ... 
│   └── Termframe                   # Termframe dataset with its original folder structure
└── report                          

• assets: files used in README.md
• config_files: configuration parameters for each of the experiments
• data: folder with the Termframe dataset, all the pre-processed data is saved here after running the scripts as well as new experiments folder with different datasets setups and models that is created automatically by running the scripts.
• report: folder with the scientific paper

Dependencies, environment creation and activation

• python>=3.7
• pytorch>=1.8.0
• cudatoolkit>=11.1
• dependencies in the environment.yml can be installed automatically with the comand below (solving the environment might take a while) or manually (pip is recommended for installing transformers)
• pytorch and cudatoolkit versions have to be installed manually

# create new environment
conda create --name <env_name> python=3.9

# activate environment
conda activate <env_name>

# install dependencies
conda env update -f environment.yml

# don't forget to manually install pytorch with cudatoolkit

Preprocessing data

# convert data from .tsv to .csv format
python convert_data.py

# prepare and split data for training and testing
# for sequence tagging
python prepare_data.py

#for relation extracion
python prepare_data_regions.py

Train models

# train sequence tagger
python train_sequence_tagging.py

# train relation extractor
python train_relation_extraction.py

# run experimental relation extraction (after training relation extractor)
python relation_extraction_growing_window.py

Generate graphs

# install nltk packages (used only once, since the packages are saved locally)
python install_nltk_packages.py

# generate graphs
python generate_graphs.py

Example of a complete graph of relations

Example of a graph with largest connected componens