Detecting malicious tweets in twitter using runtime monitoring with hidden information

Abstract

Although there is voluminous data flow in social media, it is still possible to create an effective system that can detect malicious activities within a shorter time and provide situational awareness. This thesis developed patterns for a probabilistic approach to identify malicious behavior by monitoring big data. We collected twenty-two thousand tweets from publicly available Twitter data and used them in our testing and validation processes. We combined deterministic and nondeterministic approaches to monitor and verify the system. In the deterministic part, we determined assertions by using natural language (NL) and associated formal specifications. We then specified visible and hidden parameters, which are used for subsequent identification of hidden parameters in Hidden Markov Model (HMM) techniques. In the nondeterministic part, we used probabilistic formal specifications with visible and hidden parameters, used in HMM, to monitor and verify the system. An important contribution of the work is that we specified some event patterns indicating malicious activities. Based on these patterns, we obtained output to indicate the possibility of each tweet being malicious.