Building A Relation Knowledge Base For Open Information Extraction

In recent years, many knowledge bases, such as DBpedia, YAGO, Freebase and Zhishi.me, have been published on the Web in the form of linked data, which are very useful for both human reading and machine consumption. However, comparing with the number of different entities in these knowledge bases, there are only a few dis-tinct relations. Furthermore, these knowledge bases only extract data from structured or semi-structured data sources without considering implicit knowledge from unstruc-tured text, which is in a huge and increasing amount on the Web. On the other hand, open information extraction, such as Machine Reading and Never-Ending Language Learning, focuses on extracting entities and their relations from text at the Web scale. In this background, building a relation knowledge base for open information extraction is complementary to the existing work well.In this paper, we defined the basic structure of a relation knowledge base for open information extraction and designed the overview architecture to build such a knowl-edge base. Moreover, for each relation, our relation knowledge base contains not only subject-object pairs as relation examples but also high level relation restrictions such as the domains, ranges and dependency path patterns. Both information are quite useful to describe relations, which can be used as complement to the entity-based linked data, as well as further natural language processing training data or high-quality ontologies for additional extraction. The process of building the relation knowledge base is to acquire the relation examples and restrictions from text automatically, which was in-spired by open information extraction as well. Because the acquisitions to the relation examples and restrictions are highly coupled, after we extracted the candidates from text, we adapt a novel algorithm based on Expectation Maximization (EM) to estimate the scores for them. And an experimental relation knowledge base from the Chinese encyclopedias and linked data is built to show the effectiveness and efficiency of our algorithm.