Beneath the Cream: Unveiling Relevant Information Points from CrimeBB with Its Ground Truth Labels

Beneath the Cream: Unveiling Relevant Information Points from CrimeBB with Its Ground Truth Labels

연구 분야: Strategies

학회: International Symposium on Cyber Security, Cryptology, and Machine Learning

In the realm of cybersecurity, identifying and mitigating the exploitation of vulnerabilities is crucial. Building on prior research that analyzes underground hacking forums, this study refines methodologies for detecting vulnerability exploitation within underground discussion forums. Using the CrimeBB dataset, previous works employed machine learning approaches to extract insights, label textual information, build predictive models, and classify forum posts discussing Common Vulnerabilities and Exposures (CVE). Recently, the PostCog framework was released to facilitate navigation of the CrimeBB data. The current study integrates the PostCog extension with ChatGPT, enhancing the labeling of posts by type, intent, and crime category into new classifications such as Proof-of-Concept (PoC), Weaponization, and Exploitation. Additionally, using the SHAP explanation method, we uncover insights into the keywords frequently found in the text—such as “fud”, “sell”, “buy”, and “pm”—which have emerged as significant indicators of exploitation.