关键词: 动态网络表示学习, 深度神经网络, 自编码器, 门控循环单元, 链路预测

Abstract: Learning the latent vector representation of nodes in the graph is an important and ubiquitous task, which aims to capture various attributes of the nodes in the graph. A lot of work proves that static graphs represent part of the node information that has been learned. However, real-world graphs evolve over time. In order to solve the problem that most dynamic network algorithms cannot effectively retain node neighborhood structure and temporal information, this paper proposes a dynamic network representation learning method based on deep neural network (DNN) and gated recurrent unit (GRU): DynAEGRU. The algorithm uses the autoencoder as the framework. First, the encoder uses a deep neural network to gather neighborhood information to obtain low-dimensional feature vectors. Secondly, the GRU network is used to gather node temporal information, and the decoder reconstructs the adjacency matrix and compares it with the real graph to construct the loss. Through the experimental analysis of this model and 8 graph representation learning algorithms on 3 datasets, the results show that the DynAEGRU model has a good performance.

Key words: Keywords: dynamic network representation learning, deep neural network, autoencoder, gated recurrent unit, link prediction