LAUSR

Abstract Network embedding algorithms learn low-dimensional features from the relationships and attributes of networks. The basic principle of these algorithms is to preserve the similarities in the original networks as much as possible. However, in heterogeneous information networks, existing algorithms are not sufficiently expressive to capture detailed semantic patterns between nodes. In this paper, we propose a novel heterogeneous information network embedding algorithm called the relational triplet network (RTN). In the data sampling phase, meta-schema-based random walks are performed to extract semi-hard quadruplets based on the node type and degree. In the representation learning phase, a relational triplet loss is designed to optimize the distance of triplet embeddings on diverse heterogeneous relationships. The empirical results demonstrate that our algorithm can obtain multiple types of representations and outperform other state-of-the-art methods in node classification and link prediction.