SPEAKERS: Sehar Shafique, PhD Fellow at ITU
ABSTRACT: “Knowledge graphs are increasingly constructed through declarative mapping languages, which allow users to define transformation rules from heterogeneous data sources into structured graph representation. As mappings evolve, assessing and maintaining their structural quality becomes a critical challenge. Yet no systematic framework for evaluating the quality of KG construction mappings exists. In this paper, we present the first empirical study of how YARRRML mappings evolve over time in open-source development contexts, combining commit classification with a longitudinal analysis of six structural quality metrics adapted from software engineering. We analyze 521 commits across 35 open-source repositories and find (1) that feature additions and bug fixes dominate mapping evolution, (2) that repositories do not follow uniform growth and correction trajectories, mostly driven by changes in data, and (3) URI and namespace errors are the most prevalent bug type. Our results indicate that support for dependencies in knowledge graph construction is lacking, and that declarative mappings must be seen as part of all data, software, and knowledge ecosystems and lifecycles during development.”