Machine Learning-Based Harmful Website Classification Method

Method for Classifying Harmful Websites Using HTML Tokenization and Machine Learning

This technology relates to a method for classifying whether a website is harmful by tokenizing and vectorizing the HTML source code of the accessed website and processing it through a machine learning model.

Existing rule-based harmful site blocking methods struggle to detect new or modified sites and often result in high false-positive rates.

To address this, this technology tokenizes HTML source code and inputs it into a machine learning model, enabling accurate, learning-based classification of even newly emerging harmful sites.

This invention is the result of the 'AI Information Collection for Monitoring Illegal Content Distribution Sites' project, supported by the Ministry of SMEs and Startups.

‍

‍

Key Features:

• A step of accessing a specific website by extracting and modifying numbers from a domain address and repeatedly attempting connection
• A step of extracting and preprocessing the HTML source code of the website to perform tokenization
• A step of vectorizing each token according to a pre-set algorithm and inputting them into a machine learning model
• A step of classifying the website as harmful based on the output of the machine learning model