README.md

metadata

license: cc-by-nc-4.0
task_categories:
  - text-classification
  - image-classification
  - multimodal
tags:
  - web-security
  - phishing-detection
  - html-analysis
  - screenshot
  - manually-verified
dataset_info:
  features:
    - name: url
      dtype: string
      description: Website URL and folder name for resource access
    - name: html_content
      dtype: string
      description: 'Path to HTML file (format: {url}/page.html)'
    - name: screenshot_content
      dtype: image
      description: 'Path to screenshot image (format: {url}/screenshot.png)'
    - name: label
      dtype: string
      description: Classification label from metadata.txt (phishing/legitimate)
  splits:
    - name: train
      num_bytes: 1024000000
      num_examples: 1000
  configs:
    - config_name: default
      data_files:
        - split: train
          path: data/dataset.csv

Web Content Classification Dataset with HTML and Screenshots

Dataset

Description: This dataset contains 1,000 carefully curated examples of web content designed for phishing detection classification tasks. Each example consists of a website URL along with its complete HTML content, a visual screenshot, and a manually verified classification label.

Important Note: This entire dataset was meticulously created through manual processes. Each website was individually visited, analyzed, and verified to ensure the highest quality and accuracy of labels.

Dataset Structure

The dataset is organized with a main CSV file that references all resources using a consistent folder structure:

CSV Columns:

• url: Website URL and folder name for resource access
• html_content: Path to HTML file ({url}/page.html)
• screenshot_content: Path to screenshot image ({url}/screenshot.png)
• label: Classification label (phishing/legitimate)

Directory Structure:

dataset: files: - dataset.csv folders: - name: example-domain-1.com files: - page.html - screenshot.png - metadata.txt - name: example-domain-2.com files: - page.html - screenshot.png - metadata.txt

File Reference System

The dataset uses a consistent referencing system where each URL serves as both the website address and folder name:

• HTML Content: "{url}/page.html (e.g., example.com/page.html)"
• Screenshot: "{url}/screenshot.png (e.g., example.com/screenshot.png) "
• Label: Value contained in {url}/metadata.txt

Collection Methodology

Manual Curation Process

This dataset was created through an extensive manual verification process:

1. Individual Website Visits: Each of the 1,000 websites was personally visited and analyzed
2. Manual Verification: Every classification label was manually assigned after careful examination
3. Quality Control: Each entry was individually checked for accuracy and completeness
4. Content Capture: HTML and screenshots were captured while ensuring proper rendering

Data Points Collected:

data_points: URL: "The exact folder name and web address" HTML content: "Full source code saved as {url}/page.html" Visual representation: "High-quality screenshot saved as {url}/screenshot.png" Classification label: "Value stored in {url}/metadata.txt (verified as phishing/legitimate)"