Role-based Representation Learning of Multi-relational Graph with Node Attributes

Multi-relational graph representation learning aims to obtain low-dimensional vector representations of nodes and edges. Researchers have paid significant attention to representation learning of multi-relational graphs in recent years. However, most of the existing methods overlook the node role information, which could be utilized to determine the similarities between nodes, often leading to sub-optimal performance. To address this issue, we propose a novel Role-based Multi-relational Graph Representation Learning method (RMGRL), which leverages node role information to capture the similarities between nodes across the entire graph. More specifically, we first define the roles of each node based on its structural information and attribute features. Then, by integrating both intra-graph and inter-graph structural information, we unifies information from different levels into a single optimization framework to obtain role-based node representations. Finally, the proposed method is evaluated on node classification and link prediction tasks. Experimental results on two real multi-relational graphs demonstrate that the proposed method outperforms the state-of-the-art methods.