Knowledge Graph Programming with a Human-in-the-Loop: Preliminary Results

Abstract

In this paper we introduce knowledge graph programming, a new method for writing extremely succinct programs. This method allows programmers to save work by writing programs that are brief but also underspecified and underconstrained; a human-in-the-loop "data compiler" then automatically fills in missing values without the programmer's explicit help. It uses modern data quality mechanisms such as information extraction, data integration, and crowdsourcing. The language encourages users to mention knowledge graph entities in their programs, thus enabling the data compiler to exploit the extensive factual and type structure present in modern KGs. We describe the knowledge graph programming user experience, explain its conceptual steps and data model, describe our prototype KGP system, and present some preliminary experimental results.