jho12/multitask-cyberthreat-detection

This fork is created for a project report for CS541 at Stevens Institute of Technology. This repository holds the data, source code and resulting model weights for the paper "Towards end-to-end Cyberthreat Detection from Twitter using Multi-Task Learning" to be presented at IJCNN 2020.

Requirements

• 3.6.1 <= python <= 3.8.x
• torch == 1.5.1
• scikit-learn == 1.0.2
• tqdm == 4.63.0
• pandas == 1.0.5
• pytorch-crf == 0.7.2

Running the Binary Classification Model

python bin_classifier.py -conf conf/bin/cnn.json -save true
python bin_classifier.py -conf conf/bin/cnn_rnn.json -save true
python bin_classifier.py -conf conf/bin/rnn_cnn.json -save true
python bin_classifier.py -conf conf/bin/rnn.json -save true

Running the Named Entity Recognition Model

python ner_classifier.py -conf conf/ner/cnn.json -save true
python ner_classifier.py -conf conf/ner/cnn_rnn.json -save true
python ner_classifier.py -conf conf/ner/rnn_cnn.json -save true
python ner_classifier.py -conf conf/ner/rnn.json -save true

Running the Multi-task Model

python mt_classifier.py -conf conf/mt/cnn.json -save true
python mt_classifier.py -conf conf/mt/cnn_rnn.json -save true
python mt_classifier.py -conf conf/mt/rnn_cnn.json -save true
python mt_classifier.py -conf conf/mt/rnn.json -save true

Finding Model Results

All model results will be stored in ckpts. bin corresponds to the binary classifier, ner corresponds to the named entity recognition model, and mt corresponds to the multi-task model.

Running Grid Search

Running grid search will take an extremely long amount of time. Results for grid search will be stored in results. If the folder does not exist, run the script and the folder will be created automatically.

python grid_search.py -task bin -arch cnn
python grid_search.py -task bin -arch cnn_rnn
python grid_search.py -task bin -arch rnn_cnn
python grid_search.py -task bin -arch rnn

python grid_search.py -task ner -arch cnn
python grid_search.py -task ner -arch cnn_rnn
python grid_search.py -task ner -arch rnn_cnn
python grid_search.py -task ner -arch rnn

python grid_search.py -task mt -arch cnn
python grid_search.py -task mt -arch cnn_rnn
python grid_search.py -task mt -arch rnn_cnn
python grid_search.py -task mt -arch rnn