A reliable friendship mechanism for online social network exploiting pre and post-filtering approach

Online social networks are becoming increasingly popular, and their uses are growing day by day. It is an integral part of our daily lives and an incomparable medium to communicate with family, friends, and professionals in the interest of personal and professional purposes. Because of these features, the online social network contains a great deal of information that individuals can share with one another. Therefore, personal information is easily disclosed online and is misused by unreliable friends or associates without the users’ awareness. Unfortunately, certain functional issues have not been addressed regarding the automatic filtering approach in initiating friendships and users’ interactions with their online associates. Furthermore, a user cannot properly scrutinize the anomalous behaviour of other users over the time variant, which can clearly engage them in malpractices. In order to address these issues, the proposed study develops a reliable friendship mechanism for online social networks by utilizing automated two-phased (pre and post) filtering approaches to determine reliable friends and monitor their behavioral activities. In the prefiltering approach, a user can select a friend using a reliable mechanism, which consists of two choices: attribute-based and model-based. A machine learning (ML) enabled postfiltering approach is designed to determine suspicious and unwanted behavioral activities. Finally, the proposed mechanism incorporates pre and post-filtering approaches, resulting in a novel hybrid approach that can accomplish the purpose of the study. The empirical analysis shows some significant comparison data towards the hybrid approach (after incorporating pre and post-filtering approaches), where the users’ perceptions of the proposed approach exceed the other competing approaches significantly. As shown in the proportionate mean values, the proposed hybrid approach achieved the highest ratio of 90.64%, with pre-filtering accounting for 69.76% and post-filtering standing for 70.56%, while the existing approach had the lowest proportionate mean value at 47.60%. The outcomes of this study are expected to assist OSN providers, research communities, and ICT authorities in providing a standard solution for selecting reliable friends and avoiding their malpractices in OSN