A novel model for Sybil attack detection in online social network using optimal three-stream double attention network

Online social networks (OSNs) have gained popularity as platforms for professional, personal, and social networking. However, they are also vulnerable to fraudulent events such as rumors or fake news, which can mislead users and have serious consequences. The dissemination of misinformation on social networking sites has become a global threat. To address this, a novel model GO-3DANB is proposed to utilize a three-stream, double attention network-modified BiLSTM (3S-A2 DenseNet-modified BiLSTM) with Gannet optimization algorithm (GOA) to predict Sybils in OSNs. The GO-3DANB model extracts network weights and crucial characteristics using the 3S-A2 DenseNet-modified BiLSTM to automatically extract higher and lower features from social network fake account (SNFA) input data. The SNFA dataset, containing 17 metadata features from fake and real profiles and a total of 921 data points, is collected from Kaggle. The modified BiLSTM is used to establish forward correlation with input sequences and learn dependencies with input sequences in backward sequence, thereby completely extracting context information among various channels. We utilize GOA optimization to assess each individual's fitness for identifying Sybil's attacks. Our proposed method is evaluated using various metrics, including precision, sensitivity, accuracy, specificity, F1-score value, and receiver operating characteristic curve (ROC). In comparison to existing methods, our proposed method, GO-3DANB, achieves 96% accuracy, 96.34% precision, 95.31% specificity, 96.5% sensitivity, 0.96 F1-score value, and 0.97 ROC. Thus, our proposed method demonstrates the potential to address the problem of Sybil detection in OSNs and provides a valuable contribution to the field of Sybil security.