Active knowledge graph completion

Enterprise and public Knowledge Graphs (KGs) are known to be incomplete. Methods for automatic completion, sometimes by rule learning, scale well. While previous rule-based methods learn closed (non-existential) rules, we introduce Open Path (OP) rules that constrained existential rules. We present a novel algorithm, OPRL, learning OP Closed complete KG answering queries of unclear origin, usually derived from holdback test set in experimental settings. However, can generate relevant completion. OPRL generates even when there is no answer the query, or correct missing entity not KG. For well, propose embedding-based fitness function efficiently estimate quality. Additionally, novel, efficient vector computation formally assess evaluate using adaptations Freebase, YAGO2, Wikidata, synthetic Poker find mines hundreds accurate massive KGs with up 8 M facts. The precision as high 98% recall 62% on KG, demonstrating first solution active knowledge graph