FactRunner: Fact extraction over Wikipedia

The increasing role of Wikipedia as a source of human-readable knowledge is evident as it contains an enormous amount of high quality information written in natural language by human authors. However, querying this information using traditional keyword based approaches requires often a time-consuming, iterative process to explore the document collection to find the information of interest. Therefore, a structured representation of information and queries would be helpful to be able to directly query for the relevant information. An important challenge in this context is the extraction of structured information from unstructured knowledge bases which is addressed by Information Extraction (IE) systems. However, these systems struggle with the complexity of natural language and produce frequently unsatisfying results. In addition to the plain natural language text, Wikipedia contains links between documents which directly link a term of one document to another document. In our approach for fact extraction from Wikipedia, we consider these links as an important indicator for the relevance of the linked information. Thus, our proposed system FactRunner focusses on extracting structured information from sentences containing such links. We show that a natural language parser combined with Wikipedia markup can be exploited for extracting facts in form of triple statements with a high accuracy.