Temporal networks are networks that edges evolve over time, hence link prediction in temporal networks aims at inferring new edges based on a sequence of network snapshots. In this paper, we propose a graph wavelet neural network (TT-GWNN) framework using topological and temporal features for link prediction in temporal networks. To capture topological and temporal features, we develope a second-order weighted random walk sampling algorithm. It combines network snapshots with both first-order and second-order weights into one weighted graph. Moreover, it incorporates a damping factor to assign greater weights to more recent snapshots. Next, we adopt graph wavelet neural networks to embed the vertices and use gated recurrent units for predicting new links. Extensive experiments demonstrate that TT-GWNN can effectively predict links on temporal networks.

中文翻译：

---

基于小波神经网络嵌入的时空信息有效链路预测

时间网络是边缘随时间变化的网络，因此时间网络中的链路预测旨在基于一系列网络快照来推断新的边缘。在本文中，我们提出了一种使用拓扑和时间特征的图小波神经网络（TT-GWNN）框架，用于时间网络中的链接预测。为了捕获拓扑和时间特征，我们开发了一种二阶加权随机游走采样算法。它将具有一阶和二阶权重的网络快照组合到一个加权图中。此外，它还结合了阻尼系数，可以为最近的快照分配更大的权重。接下来，我们采用图小波神经网络嵌入顶点，并使用门控循环单元来预测新链接。