Open Information Extraction for the Web

1 3 , 8 1 0 , 0 0 0 T u p l e s ? P r i m a r y E n t i t i e s ? R e l a t i o n s F i l t e r i n g Figure 4.2: Open Extraction from Wikipedia: TextRunner extracts 32.5 million distinct assertions from 2.5 million Wikipedia articles. 6.1 million of these tuples represent concrete relationships between named entities. The ability to automatically detect synonymous facts about abstract entities remains an open problem. Open Extraction from The General Web What happens when we augment the size of TextRunner’s input corpus by several orders of magnitude? In addition to processing Wikipedia, we added 500 million Web pages to the set of documents processed by TextRunner. This combination of Wikipedia and the Web is thus referred to as General-Web. After eliminating extractions found only in a single sentence, TextRunner was found to extract approximately 850 million raw tuples from General-Web, with 218 million tuples representing unique facts. Of these 218 million, 16.5 million tuples represent concrete facts; 14 million concrete facts remained after applying the aforementioned distributional The author wishes to thank Google Inc. for providing the corpus.