Spatial-Temporal Attention TCN-Based Link Prediction for Opportunistic Network

• DOI: 10.3390/electronics13050957
• Publication Date: 2024-03-01T16:19:24Z
• Authors: Jian Shu; Yunchun Liao; Jiahao Li
• Keywords: Convolution (computer science); Spatial correlation; Link (geometry)

Link prediction for opportunistic networks faces the challenges of frequent changes in topology and complex variable spatial-temporal information. Most existing studies focus on temporal or spatial features, ignoring ample potential In order to better capture correlations evolution explore their information, a link method based attention convolution network (STA-TCN) is proposed. It slices into discrete snapshots. A state matrix information attribute constructed represent Time convolutional mechanisms are employed learn Furthermore, improve performance, proposed converts auto-correlation error non-correlation error. On three real datasets, ITC, MIT, Infocom06, experimental results demonstrate superior predictive performance compared baseline models, as shown by improved AUC F1-score metrics.