Skip to content

Latest commit

 

History

History
104 lines (79 loc) · 3.41 KB

README.md

File metadata and controls

104 lines (79 loc) · 3.41 KB
Raw

CTI-Document-Analyzer

CTI-Document-Analyzer is a tool for automatically extracting ATT&CK Techniques from CTIs

Usage

Installing requirements

To use CTI-Document-Analyzer you must install Python requirements run:

python3 -m pip install -r requirements.txt

The models required for operation are provided in figshare repo, please download it into the models/data folder.

Ways of use

The framework is provided with the use_pipeline.py. It can be used by specifying in the -path option the folder where all the documents are to be analyzed (which must be provided in .html or .pdf format)

python3 use_pipeline.py -path [FOLDER]

In this case the results will be available into the output/ folder.

Moreover, 3 possible text chunking modes could be selected by specifying it with the option -scraper :

  • Adjacent clustering (default)
  • Sentence splitting
  • Paragraph splitting

More information could be retrieved by running

python3 use_pipeline.py -h 

usage: CTI Document Analyzer [-h] -path PATH [-so | -po]
                             [-scraper {semantic,naive,paragraph}]
                             [-summarize] [-tfv]

It retieves ATT&CK Techniques from the document given in input using multiple
DL-Models

options:
  -h, --help            
  -path PATH            Path of CTRs (if used in scrape only mode/default
                        mode), or path of CTI scraped (previous output
                        provided from so-mode) if used in predict only mode
  -so, --scrape-only    uses the script only as a scraper and chunker for
                        CTRs.
  -po, --predict-only   uses the script only for prediction, the output of
                        scrape-only must be provided
  -scraper {semantic,naive,paragraph}
                        What chunker should be used
  -summarize            Chose if should be provided the summarization of
                        scraped paragraph
  -tfv, --tf-verbose    Enable Tensorflow/Torch/CUDA Warming and info

The input should be provided into a folder in .pdf or .html format

Folder Organization

The framework is organized in many folders:

├── data
      ├── caldera
      ├── resources
          ├── apt29
          ├── carbanak
          ├── fin6
          ├── fin7
          ├── menuPass
          ├── oilrig
          ├── sandworm
          └── wizardspider
      ├── datasets
      	├── dataset_classifier.csv
        └── dataset_detector.csv
      └── sorter
├── finetune
├── models
├── output
├── sorter
├── utilities
├── requirements.txt
├── use_pipeline.py
├── LICENSE
└── README.md

  • The data/ folder contains many information unnecessary for the framework execution

    • In data/resources/ are listed all the resources used in our analysis
    • In data/datasets/ are present the dataset in .csv format

  • The finetune/ folder contains the script which could be used to recreate the models provided.

  • The models/ folder contains the wrapping class of the used model. Those models (downloadable in figshare must be provided in the models/data/ folder

  • The other folders are framework utilities.