INDUCTIVE RELATION PREDICTION FROM RELATIONAL PATHS AND CONTEXT WITH HIERARCHICAL TRANSFORMERS

Relation prediction on knowledge graphs (KGs) is a key research topic. The dominant embedding-based methods mainly focus on the transductive setting and lack the inductive ability to generalize to new entities during inference. Existing methods for inductive reasoning predict missing relations mainly by mining the connections between entities, i.e., relational paths. However, most of these methods do not consider the nature of head and tail entities contained in their relational context, which gives additional information for the reasoning. This paper proposes a novel method that captures both connections between entities and the intrinsic nature of entities, by simultaneously aggregating RElational Paths and cOntext with a unified hieRarchical Transformer framework, namely REPORT. REPORT relies solely on relation semantics and can naturally generalize to the fully-inductive setting, where KGs for training and inference have no common entities. Empirical experiments validate the superiority of REPORT on several datasets. It performs consistently better than all baselines on almost all the eight version subsets of two fully-inductive datasets. Moreover, REPORT can generate interpretable explanations for each of its predictions by providing the contributions of each element to the prediction results.