Dynamic security log processing using deep learning techniques

Recently, the number of discovered cyber attacks increases rapidly. Tools for stealing personal data, destroying systems, or controlling infrastructure become continuously sophisticated to achieve malicious aims. Companies are trying to reduce the number of risks on their assets by using various security monitoring devices and tools. SIEM solutions are used for security monitoring, allowing different logs to be correlated. They offer visibility for security teams and allow early response to attacks. The main problem of SIEM software is the implementation of log parsing, which directly influences correlation rules efficiency. Usually, the biggest limitation is parsing dynamic log structures from different event sources. The main contribution of this paper is to apply advanced deep neural networks which use attention mechanisms for efficient log content parsing and its understanding. The proposed question answering model for feature extraction from raw logs should achieve automatic log procession. Obtained results show indisputable advantages of deep attention techniques compared to the common approaches