An Estimation Method of Attributed Network Representation Spatial Dimension

Network representation learning has been demonstrated to be helpful in network analysis and achieves great success in recent years. However, the spatial dimension, which is a crucial parameter of representation space, rely on expert experience for manual configuration, lacking of theoretical model support. To address this issue, we propose a novel method called EMANRD for estimating the dimension of representation space that meets the goal of network representation learning based on the original network data. Specifically, EMANRD mixedly integrate topological structure and attribute information into a metrizable space and estimate the ideal representation spatial dimension by designing a ideal representation space generating model. Experimental results on three real-world datasets demonstrate that EMANRD can effectively estimate the ideal dimension of representation space.