Most of previous work in knowledge base (KB) completion has focused on theproblem of relation extraction. In this work, we focus on the task of inferringmissing entity type instances in a KB, a fundamental task for KB competitionyet receives little attention. Due to the novelty of this task, we construct alarge-scale dataset and design an automatic evaluation methodology. Ourknowledge base completion method uses information within the existing KB andexternal information from Wikipedia. We show that individual methods trainedwith a global objective that considers unobserved cells from both the entityand the type side gives consistently higher quality predictions compared tobaseline methods. We also perform manual evaluation on a small subset of thedata to verify the effectiveness of our knowledge base completion methods andthe correctness of our proposed automatic evaluation method.
展开▼