TAP: A static analysis model for PHP vulnerabilities based on token and deep learning technology.

Area of Science:

• Computer Science
• Software Engineering
• Cybersecurity

Background:

• Increasing use of web applications leads to rising source code security issues.
• Existing vulnerability detection models often rely on complex graphs or expert-defined regex patterns.
• Vulnerabilities in web applications pose significant risks to both service providers and users.

Purpose of the Study:

• To propose TAP, a convenient and easy-to-use analysis model for discovering vulnerabilities in PHP web programs.
• To leverage token mechanism and deep learning for enhanced source code security analysis.
• To improve upon existing methods for web application vulnerability detection.

Main Methods:

• Developed a custom tokenizer based on PHP language tokens to unify and optimize parsing, incorporating parameter iteration for data flow analysis.
• Trained a deep learning model combining word2vec and Long Short-Term Memory (LSTM) network algorithms.
• Evaluated the TAP model on the Software Assurance Reference Dataset (SARD) and SQLI-LABS dataset, specifically focusing on CWE-89 vulnerabilities.

Main Results:

• TAP achieved an Area Under the Curve (AUC) of 0.9941 and an accuracy of 0.9787 on the CWE-89 dataset, outperforming other models.
• Demonstrated superior performance in multiclass classification compared to RIPS, achieving a Kappa of 0.8319 and a Hamming distance of 0.0840.
• The custom tokenizer effectively unified tokens and supported PHP features, optimizing parsing and enabling data flow analysis.

Conclusions:

• TAP offers a highly accurate and efficient solution for detecting PHP web application vulnerabilities.
• The proposed token-based deep learning approach surpasses existing methods in both detection accuracy and multiclass classification.
• TAP provides a valuable tool for enhancing the security of PHP web programs.