Knowledge Mining in Cybersecurity: From Attack to Defense

Khandakar Ashrafi Akbar, Sadaf Md. Halim, Yibo Hu, Anoop Singhal, Latifur Khan, Bhavani Thuraisingham

Abstract: In the fast-evolving world of Cybersecurity, an analyst often has the difficult task of responding to new threats and attack campaigns within a limited amount of time. If an analyst fails to do so, this can lead to severe consequences for the system under attack. In this work, we are motivated to aid the security analyst by introducing a tool which will help to produce a swift and effective response to incoming threats. If an analyst identifies the nature of an incoming attack, our system can produce a ranked list of solutions for the analyst to quickly try out, saving both effort and time. Currently, the security analyst is typically left to manually produce a solution by consulting existing frameworks and knowledge bases, such as the ATT &CK and D3FEND frameworks by the MITRE Corporation. To solve these challenges, our tool leverages state-of-the-art machine learning frameworks to provide a comprehensive solution for security analysts. Our tool uses advanced natural language processing techniques, including a large language model (RoBERTa), to derive meaningful semantic associations between descriptions of offensive techniques and defensive countermeasures. Experimental results confirm that our proposed method can provide useful suggestions to the security analyst with good accuracy, especially in comparison to baseline approaches which fail to exhibit the semantic and contextual understanding necessary to make such associations.