Analyzing Opinion Spammers’ Network Behavior in Online Review Systems

Review systems are indisputably important resource for users when making various decisions on products or services. Consequently, they become increasingly targeted by attackers who deliberately inject biased reviews, so called opinion spams, aiming to influence normal users' decisions for financial gain. In this paper, we perform an empirical analysis of opinion spammers on one of the famous online review systems, Yelp. Specifically, we analyze two different types of networks: implicit and explicit networks of opinion spammers and those of non-spammers. Through analyzing the network characteristics in different networks, we show similarities and differences between opinion spammers and non-spammers in terms of statistical characteristics and network properties. More specifically, through extensive analysis on Yelp dataset, we show that (i) the explicit network of non-spammers exhibits typical “small-world” properties of social networks. (ii) the implicit network of non-spammers is close to random networks. (iii) in both explicit and implicit networks, opinion spammers form near-isolated communities with dense inner connections among themselves, while exhibiting the lower level of “small-world” properties.