Dynamic Attribute Network Representation Learning and Visualization of Integrated Spatiotemporal Trajectory

Network representation learning is a research hotspot in information network data representation.Compared with traditional network analysis technology,it has shown its effectiveness and efficiency.At present,most researchers only treat the network as static state whose structure does not change over time.The rich node attribute information in the network is rarely considered,which is difficult to adapt to the dynamic characteristics of real-time networks.Considering the dynamic nature of the network and the attributes of nodes,a dynamic attribute network representation learning based on spatiotemporal path named DAWalk is proposed in this paper,which aggregates the structural feature and attribute feature into the embedded representation of nodes.The spatiotemporal trajectory sequence is introduced to capture the structural characteristics and dynamic evolution trend of the network.In the model learning,we use the improved auto-encoder to minimize the distance loss of pairs of nodes in the sequence to learn the highly nonlinear rule of hidden sequence node pairs,which makes the learned node representation more robust.Experimental results show that the performance of proposed DAWalk is better than other benchmark algorithms in visualization,link prediction and node classification tasks.