Knowledge base completion by learning pairwise-interaction differentiated embeddings

摘要: A knowledge base of triples like (subject entity, predicate relation,object entity) is a very important resource for management. It useful human-like reasoning, query expansion, question answering (Siri) and other related AI tasks. However, such often suffers from incompleteness due to large volume increasing in the real world lack reasoning capability. In this paper, we propose Pairwise-interaction Differentiated Embeddings model embed entities relations low dimensional vector representations then predict possible truth additional facts extend base. addition, present probability-based objective function improve optimization. Finally, evaluate by considering problem computing how likely triple true task completion. Experiments on WordNet Freebase show excellent performance our algorithm.