We consider knowledge base (KB) completion for KBs rich in facts about common nouns (generics), such as "trees produce oxygen" or "dogs have tails". While KB completion has received much attention for named entity KBs, little emphasis has been placed on generics despite their importance for capturing general knowledge. Compared with named entity KBs such as Freebase, KBs about generics have more complex underlying regularities, are substantially more incomplete, and violate the commonly used locally closed world assumption (LCWA). Consequently, existing completion methods struggle with this new task, and the commonly used evaluation metrics become less meaningful. To address these challenges, we make three contributions: (a) a tensor factorization approach that achieves state-of-the-art results by incorporating external knowledge about relation schema and entity taxonomy, (b) taxonomy guided submodular active learning to efficiently collect additional annotations to address KB incompleteness, and (c) a more appropriate metric (yield at high precision) along with a constant-factor deterministic approximation algorithm to compute it cost-effectively with only a logarithmic number of human annotations. An empirical evaluation shows that our techniques achieve state-of-the-art results on two novel generics KBs about elementary level science.

通过使用外部信息，如关系模式和实体分类法，在基础知识库或KB中推断出更多与通用名词或总称有关的事实，这是以前的KB补全方法所无法处理的。同时，在收集有关罕见实体的注释方面，我们提出了一种新的方法，可以比多个常规方法更有效地推断出这些实体的信息。

引导张量因子分解的通用知识补全