Abstract: The frequent node mobility of opportunistic networks leads to the challenges of link prediction. To better reflect the temporal changes in topology, an opportunistic link prediction method based on temporal generative adversarial networks (ONLP-TGAN) is proposed. The network volatility is defined for calculating the network segmentation time slot so that an opportunistic network is sliced into fine-grained snapshots. Information matrices are extracted from these snapshots， and a fusion matrix is obtained by integrating the information from both the spatial and temporal dimensions. Network feature vector matrices are constructed by graph embedding methods. Combining gated recurrent units and generative adversarial network, a temporal generative adversarial network model is constructed. It learns the evolutionary features of network topology, so as to achieve link prediction in networks at future time. Experimental results on three real datasets， ITC、Infocom06, and MIT， demonstrate that the proposed mothed ONLP-TGAN outperforms the baseline model in terms of Precision， Accuracy， AUC, and GMAUC metrics.

Key words: Opportunistic Network, Link Prediction, Generative Adversarial Networks, Gated Recurrent Units, Graph Embedding