Anvilogic/URL-Guardian-Dataset

URL Guardian Dataset

Dataset Summary

This dataset is designed for training classification model discriminating safe URLs from malicious ones. It consists of samples associated to their labels. The dataset is formatted for binary cross-entropy training.

Supported Tasks and Leaderboards

Languages

This dataset includes a multilingual set of domains, reflecting the diversity of internet domains globally.

Dataset Structure

Data Instances

Each instance in the dataset consists of two domains:

• value: An URL .
• label: Its classification label, where: 0 indicates a safe URL,1 indicates a malicious URL.

An example from the dataset is as follows:

{
  "value": "touro.edu/about",
  "label": "0"
}

Data Splits

The dataset is structured to be used for embedding model training and evaluation:

• Split: Train Number of Instances: 279,405
• Split: Validation Number of Instances: 59,883
• Split: Test Number of Instances: 60,712

Dataset Creation

Data Generation

Dataset is a mixture of malicious data sampled from following sources:

• URLHaus
• VT
• phishtank
• phishstats
• feodo

In order to obtain FQDN format for url we apply the following steps :

• Drop
  • non http:// and https:// URLs
  • URLs that don’t match RFC3987
  • URLs with invalid/empty hostname
  • local URLs
  • URLs that are actually comma-separated URLs
  • bare domains
• Processing steps
  • fix malformed schemes that contain http or https
  • remove URL scheme
  • remove www. only if is a subdomain
  • remove trailing slash / from path
  • remove URL query string
  • remove URL fragment
  • perform duplicate URL removal before and after data processing

Dataset Usage

This dataset is designed for malicious URL detection, aiding in phishing prevention, threat intelligence, and cybersecurity monitoring. It supports training models to classify URLs as safe or malicious, enhancing online security.