Previous cross-lingual knowledge graph (KG) alignment studies rely on entity
embeddings derived only from monolingual KG structural information, which may
fail at matching entities that have different facts in two KGs. In this paper,
we introduce the topic entity graph, a local sub-gra