A hybrid soft computing with big data analytics based protection and recovery strategy for security enhancement in large scale real world online social networks

In recent years, Online Social Networks (OSN) have become an integral part of human society. OSN has become an important and crucial mode of interaction among people around the world. For the past two decades, the need for data privacy security among the participants of an OSN has been challenged as technological advances have provided a lot of benefits and exposure to all active members sharing content in various online communities. Large amounts of data are usually stored on different network nodes, which can lead to malicious data security and privacy issues. When OSN users share any private content with the other participants of same network like with their friends and colleagues, their information may be inadvertently leaked to various other individuals to whom it is not supposed to be shared. Those individuals may include one's enemies, social robots, and users with fake credentials, spammers or data cutters. In addition, the spread of malicious through OSN can lead to massive damage to the network and privacy leakage. To overcome this problem, we propose a hybrid soft computing with big data analytics based protection and recovery strategy (HSBD-PR) for security enhancement in large scale OSNs. First, we introduce a modified teaching-learning with fish swarm optimization (MTL-FS) technique to classify the abnormal network users as five classes are susceptive, contagious, doubt, immune and recoverable users. Second, we fuse a block window optimization with Bayesian belief network (BWB) for analyze the propagation mechanism of abnormal users in network. After that, deep belief neural network (DBNN) based protection and recovery strategy is developed to improve the immunity of abnormal users and reduce the count of it. Finally, the performance of proposed HSBD-PR approach evaluate by standard benchmark datasets is Enron email, Epinions and Facebook. The results of this proposed method are compared with the existing state-of-art methods in terms of different performance metrics are detection failure rate, detection success rate, marginal detection gain and detection accuracy respectively.