/ITDBERT

Insider threat detection via bert

Primary LanguagePython

ITDBERT: An insider threat detection Framework based on sequence semantics

Extracting user behaviour semantics with Temporal information embedded

Required packages

  • python 3.7
  • numpy
  • sklearn
  • torch 1.7.0+cu110
  • Keras 2.3.0

Usage

# Prd-Train and insider threat detection：
# MaskedLM +(emb/CLS)
python trans_corpus.py
python robertaLSTM.py

# word2vec + (TextCNN/BiLSTM_ATT/RCNN/Transformer)
python Mygensim.py
python run.py

#Autoregressive models(deeplog)
python ActModel_train.py

Files

  • data: the dataset after data preprocessing
  • deeplog: the implementation of deeplog model
  • model: models architecture and hyperparameter setting
  • preTrain: pretrain model
  • save_model: save place
  • 42cert2Corpus.py: Preprocessing of cert datasets
  • robertaLSTM.py: run MaskedLM+LSTM for insider threat detection
  • run.py: run training models for insider threat detection
  • utils.py: utils for training