Investigating cyber threats on social media using machine learning

"Internet platforms are places where data can be found easily and communication is fast. Thus, in this information age, knowledge can be used as a dangerous weapon for terror attacks in the cyber world. Such an attack can threaten the systems of a company or sometimes even the safety of society. Besides protecting individuals and systems from such attacks, the attack risks must also be determined because an unknown threat cannot be precluded. New threats such as malware, exploit, and fraud are often noticed too late. When these threats are first discovered, they are defined as zero-day; the weaknesses are then exploited due to the vulnerability of the computers and major damages arise. No manufacturer of security products such as IPS or IDS can succeed without fixing these risks immediately. To address these issues, this thesis employs social media resources (i.e. Twitter) to identify these weaknesses as quickly as possible. The aim is to determine zero-day attacks by gathering information published by people who are not aware that they are facing a known zero-day attack. The study uses data from social media and processes the data by implementing machine learning principles such as word classification. As a result, a program is developed that identifies vulnerabilities and notifies the relevant manufacturers or vendors. In this way, it is possible for vendors to remove the vulnerabilities in a more efficient manner by identifying zero-day attacks via social media data. In this thesis, a literature review has been undertaken to address the basic concepts of risk, security and social media, and a new-generation program has been developed for finding probable attacks that have been alluded to on social media.