Online social networks (OSNs) are trendy and rapid information propagation medium on the web where millions of new connections either positive such as acquaintance or negative such as animosity, are being established every day around the world. The negative links (or sometimes we can say harmful connections) are mostly established by fake profiles as they are being created by minds with ill aims. Detecting negative (or suspicious) links within online users can better aid in mitigation of fake profiles from OSNs. A modified clustering coefficient formula, named as Mutual Clustering Coefficient represented by M_cc, is introduced to quantitatively measure the connectivity between the mutual friends of two connected users in a group. In this paper, we present a classification system based on mutual clustering coefficient and profile information of users to detect the suspicious links within the user communities. Profile information helps us to find the similarity between users. Different similarity measures have been employed to calculate the profile similarity between a connected user pair. Experimental results demonstrate that four basic and easily available features such as work(w),education(e),home_town(ht)and current_city(cc) along with M_CC play a vital role in designing a successful classification system for the detection of suspicious links.
In the past decade, the connectivity within people has been spreading rapidly with the help of social networking sites. The connection (or link) of a person on OSNS can be either positive such as friendship or negative such as animosity. The negative links are mostly established by fake profiles as they are being created by minds with ill aims such as running spam campaigns [1], casting unfair online votes [2], accessing user personal information [3], etc. In order to fulfill aims, the fake profile users need to create as many links as possible with real profiles. The chances of friend request being accepted by a real user from the fake profiles are low as most of the connections are being established on a network if the two persons either know each other in offline or share some interests. Therefore, in order to increase the chance of friend requests being accepted, the fake users are nowadays targeting the communities 1 where users are connected and share a strong bond with each other. It is very frequent to have a high number of mutual friends for members of a group of connected people. Furthermore, it has been realized that more is the number of mutual friends, the more is the possibility that the friend request is being accepted by the users [31]. This particular feature of common friends is exploited by the fake profile owners to increase their coverage. Once these fake users succeed in bobbing few naïve users in a group (or on a page), their trust level gets increased among other members in the form of common neighbors which increases the acceptance chances of friend requests by others members in the group. Since these fake users are getting penetrated into the online communities in a clever way and get cloaked into the real public, therefore, it becomes challenging for researchers to identify and obliterate them from the social networks.
Research shows that users in OSNs connect with the people either they know in offline or met online. Social networking sites such as Facebook are being primarily used by people to maintain and strengthen the pre-existing offline social relations. It has been observed that if two persons have enough number of common friends, there are high chances that the two persons share some common offline entity such as same organization, school, course, etc., that cause them to befriend online. However, even if somehow the fake users managed to penetrate into the user groups by exploiting the mutual connections, on the other hand, there are least chances of similarity between two profiles. Based on this observation we proposed a novel approach to identify the suspicious links established by the fake users. In this paper, we present an approach to identify suspicious (negative) links established by the adversaries by exploiting the mutual friend feature in a group or a page on Facebook. Identifying suspicious links can better aid in designing the fake user detection system. The proposed approach is based on the combination of mutual clustering coefficient and profile information of a user which basically assists in detecting suspicious connections in a group or a page on Facebook. Clustering coefficient [22] is one of the topological measures used to study the structure of a graph. For a graph like Facebook-network, the clustering coefficient indicates to which extent people have mutual friends or how likely the friends of a user are connected to each other. High clustering coefficients signify a tightly connected community in which most of the friends of a user are themselves friends. In our work, we have modified the clustering coefficient as represented by , to measure the connectivity between the mutual friends of two connected users in a group. Profile information helps us to find the similarity between users. The similarity between two user profiles based on the selected attribute set can be calculated by several text-based similarity measures such as Ngram [23], Cosine similarity [24], Jaro [19]. Moreover, the authors in [20] have presented more than ten approaches to compare text documents. In this paper Fuzzy string based similarity measures profile similarity between connected friends. The main contributions of the paper are as follows:
The , a novel and a unique feature, discussed in section 3.3 have been introduced first time for the detection of suspicious (negative) links on Facebook.
The four basic and easily available features including , , ho along with have been extracted from users on Facebook network with the help of IMcrawler [43] in order to form the training dataset. The collected data set along with source code has been made available to help researchers of different domains.
Fuzzy string based similarity measures have been used to efficiently calculate the profile similarity between connected user pairs. Fake Identities have been manually designed and injected into the network to establish the links with real users of a community on Facebook
This content is AI-processed based on open access ArXiv data.
