Out of Many, One: Unifying Web-Extracted Knowledge Bases

Extracting knowledge from large text corpora and the world wide web is an important problem in artificial intelligence. Arguably, the majority of the world’s knowledge is contained in natural language text and as such needs to be brought into structured form to be accessible for automated reasoning. There are numerous information extraction (IE) projects that address this problem, such as YAGO, Freebase, and OpenIE [10, 11]. Each of these projects has its unique strengths and weaknesses. For instance, projects unconstrained by an ontology provide more coverage, but suffer from noise and ambiguities of the extracted facts. If these projects are categorized along dimensions such as extraction types, temporal and geographical attributes, events coverage, and schema language, one realizes that the projects would be highly complementary if their knowledge was integrated.