Detecting Social-Network Bots Based on Multiscale Behavioral Analysis

Social network services have become one of the dominant human communication and interaction paradigms. However, the emergence of highly stealth attacks perpetrated by bots in social-networks lead to an increasing need for efficient detection methodologies. The bots objectives can be as varied as those of traditional human criminality by acting as agents of multiple scams. Bots may operate as independent entities that create fake (extremely convincing) profiles or hijack the profile of a real person using his infected computer. Detecting social networks bots may be extremely difficult by using human common sense or automated algorithms that evaluate social relations. However, bots are not able to fake the characteristic human behavior interactions over time. The pseudo-periodicity mixed with random and sometimes chaotic actions characteristic of human behavior is still very difficult to emulate/simulate. Nevertheless, this human uniqueness is very easy to differentiate from other behavioral patterns. As so, novel behavior analysis and identification methodologies are necessary for an accurate detection of social network bots. In this work, we propose a new paradigm that, by jointly analyzing the multiple scales of users’ interactions within a social network, can accurately discriminate the characteristic behaviors of humans and bots within a social network. Consequently, different behavior patterns can be built for the different social network bot classes and typical humans interactions, enabling the accurate detection of one of most recent stealth Internet threats.