Accelerating the Update of Knowledge Base Instances by Detecting Vital Information from a Document Stream

With the growth of Linked Data, updating knowledge bases (KB) is becoming a crucial problem, particularly when representing the knowledge linked to permanently evolving instances. Many approaches have been proposed to extract new knowledge from textual documents in order to update existing KB. These approaches reach maturity but rely on the fact that the adequate corpus is already constructed. In most cases, the considered documents are manually selected which makes an exhaustive update of the KB difficult. In this paper, we propose an original approach to automatically detect from a document stream, pieces of information reporting new knowledge about instances of the KB.