Designing a System for Semi-Automatic Population of Knowledge Bases From Unstructured Text

Abstract

Important information from unstructured text is typically entered manually into knowledge bases, resulting in limited quantities of data. Automated information extraction from the text could assist with this process, but the technology is still at unacceptable accuracies. This task therefore requires a suitable user interface to allow for correction of the frequent extraction errors and validation of proposed assertions that a user wants to enter into a knowledge base. In this paper, we discuss our system for semi-automatic database population and how it handles the issues arising in content extraction and populating a knowledge base. The main contributions of this work are identifying the challenges in building such a semi-automated tool, the categorization of extraction errors, addressing the gaps in current extraction technology required for databasing, and the design and development of a usable interface and system, FEEDE, to support correcting content extraction output and speeding up the data entry time into knowledge bases. To our knowledge, this is the first effort to populate knowledge bases using content extraction from unstructured text.