Abstract: Some of the existing research on link disappearance only focused on discovering and analyzing the reasons for link disappearance, while others only used static network representations for prediction, rarely analyzing this problem from the network dynamic evolution. In response to the above research status, a pre-trained dynamic graph neural network academic cooperative behavior disappearance prediction model PreDGN was proposed. The temporal information of dynamic networks was captured by PreDGN through dynamic graph generation pre-training tasks, and the topological information of the network was supplemented with edge features constructed by temporal motifs. Then, combined with attention node embedding based on time encoding, node representations were learned more accurately. The historical information of the dynamic graph has been learned by the pre-trained model and can be fine-tuned in specific tasks for predicting the disappearance of academic cooperation behaviors. Experiments were conducted using data from the publicly available academic cooperation dataset HepTh with different time spans and scales. On the 1996, 1997, 94-96, and 97-99 sub datasets, compared to the second best method of dynamic graph neural networks DyRep, the Area Under the ROC curve (AUC) increases by 10.47, 8.16, 13.41, and 3.27 percentage points respectively, while Average Precision (AP) increases by 5.87, 2.15, 8.26, and 3.01 percentage points, respectively.

Key words: dynamic graph, pre-training, graph representation learning, graph neural network, network motif 

摘要： 现有链接消失问题研究工作一部分只停留在发现和分析链接消失的原因上，一部分仅使用静态网络表示进行预测，很少从网络动态演化的角度分析链接消失预测问题。针对以上研究现状，提出一种预训练动态图神经网络学术合作行为消失预测模型PreDGN(Pre-trained Dynamic Graph Neural Network)。PreDGN通过动态图生成预训练任务捕捉动态网络的时间信息，同时利用时序模体构造的边特征补充网络的拓扑信息，再结合基于时间编码的注意力节点嵌入，能够更精准地学习节点的表征。经过预训练的模型学习了动态图的历史信息，而且可以在特定的学术合作行为消失预测任务中进行微调。使用公开学术合作数据集HepTh中不同时间跨度、不同数据规模数据进行实验，在1996、1997、94-96和97-99子数据集上，相较于次优的动态图神经网络方法(DyRep)，AUC指标分别提高了10.47、8.16、13.41和3.27个百分点，AP指标分别提高了5.87、2.15、8.26和3.01个百分点。

关键词: 动态图, 预训练, 图表示学习, 图神经网络, 网络模体