Link Prediction in Dynamic Social Networks Using Deep Learning

Understanding the dynamics inherent in social networks, driven by their evolutionary trajectories, presents a complex challenge due to the multitude of variable factors at play. Nonetheless, a comparatively simpler task involves grasping the connection between two specific nodes within such networks. Typically, problems that evolve over time manifest as intricate structures depicted as networks that are dynamic, with contents and relationships appearing and disappearing over time. The task of effectively inferring dynamic connectivity poses significant challenges, particularly in large dynamic networks characterized by nonlinear transmission patterns and scattered connections. To tackle this issue, this study employs a Deep Belief Network (DBN) for deep feature representation of nodes and utilizes a Restricted Boltzmann Machine (RBM) for link prediction. The proposed RBM-DBN model, capitalizing on dimensionality reduction, offers more precise predictions. The effectiveness of the suggested method is assessed using two genuine, publicly available datasets from online sources, namely Facebook and Epinions. Experimental results demonstrate that the proposed model surpasses baseline models in terms of precision and recall, establishing its utility as an effective model for link prediction in social networks.