Local and Global Triple Embedding Network

his paper introduces an end-to-end learning framework called LoGNet (Local and Global Triple Embedding Network) for triple-centric tasks in knowledge graphs (KGs). LoGNet is based on graph neural net- works (GNNs) and combines local and global triple embedding informa- tion. Local triple embeddings are learned by treating triples as sequences. Global triple embeddings are learned by operating on the feature triple line graph GL of a knowledge graph G. The nodes of GL are the triples of G, edges are inserted according to subjects/objects shared by triples, and node and edge features are derived from the triples of G. LoGNet brings a refreshing triple-centric perspective in learning from KGs and is flex- ible enough to adapt to various downstream tasks. We discuss concrete use-cases in triple classification and anomalous predicate detection. An experimental evaluation shows that LoGNet brings better performance than the state-of-the-art.