Community-based features for identifying spammers in online social networks

Classifier Ensembles Using Structural Features For Spammer Detection In Online Social Networks

“…In order to, more strictly, characterize the spammers from legitimate users, the methods proposed in [1] and [13] exploit the existence of community structures in social networks [2] to extract novel structural features for the task. Communities resemble groups of nodes that are relatively densely connected to each other but sparsely connected to other dense groups in the network.…”

“…Identifying community structure in social networks is important as it reveals the functional groups in a system and thus provides information about the role of individual nodes. In the context of spammer detection, a node at the boundary of a community which has out links to nodes that belong to other distinct communities may be considered suspicious, as legitimate users tend to show high interactivity within their respective communities [1]. For spammer detection, these methods split the interaction network of OSN users into communities and then extract community-based features of network nodes (users) to classify them as spammer or legitimate.…”

“…In traditional e-mail networks, the most common form of spamming involves the Random Link Attack (RLA) where a small number of spammers send spam to a large number of randomly selected victim nodes. Spammers tend to be senders of spam messages to a socially un-related set of receivers, unlike legitimate senders whose receivers tend to cluster or form communities [1]. Many spam/spammer detection methods have been proposed in literature, which are based on the content analysis (keywords-based filtering) of the interactions between users.…”

“…Many spam/spammer detection methods have been proposed in literature, which are based on the content analysis (keywords-based filtering) of the interactions between users. However, many counter-filtering techniques based on the usage of non-dictionary words and images in spam objects are often employed by spammers [1]. Content-based spam filtering systems also demand higher computations.…”

“…Incorporating additional sociological characteristics (like interaction behavior of nodes within and across network community structures) in the classification models can improve their performance for identifying spammers who incorporate sybil and cloning attacks. In [1], we have identified some novel community-based structural features, based on inter-and intracommunity users interaction patterns, to learn improved classification models for spammer classification in online social networks. However, the results only spanned over single classifiers and the significance of using ensemble learning methods has not been extensively evaluated yet.…”