Deep learning-based method for mobile social networks with strong sparsity for link prediction

Abstract

Link prediction, the process of uncovering potential relationships between nodes in a network through the use of deep learning techniques, is commonly applied in fields such as network security and information mining. It has been utilized to identify social engineering attacks, fraudulent activities, and privacy breach risks by predicting links between nodes within a network. However, the topology of mobile social networks is subject to change over time, and the sparsity of links affects the accuracy of predictions. To address the issue of strong sparsity in link prediction for mobile social networks, a deep learning-based prediction method named DLMSSLP (deep learning-based method for mobile social networks with strong sparsity for link prediction) was developed. This method was designed to employ a combination of a Graph Auto-Encoder (GAE), feature matrix aggregation, and multi-layer long short-term memory networks (LSTM). It aimed to reduce the learning cost of the model, process high-dimensional and nonlinear network structures more effectively, and capture the temporal dynamics within mobile social networks, thereby enhancing the model’s predictive capability for the generation of existing links. When compared to other methods, DLMSSLP demonstrated significant improvements in AUC and ER metrics, showcasing the model’s high accuracy and robustness in predicting uncertain links.

Keywords

• link prediction;mobile social networks;strong sparsity;deep learning