Debiasing Knowledge Graphs: Why Female Presidents are not like Female Popes.

Abstract

Bias, may it be in sampling or judgment, is not a new topic. However, with the increasing usage of data and models trained from them in almost all areas of everyday life, the topic rapidly gains relevance to the broad public. Even more, the opportunistic reuse of data (traces) that characterizes today's data science calls for new ways to understand and mitigate the effects of biases. Here, we discuss biases in the context of Linked Data, ontologies, and reasoning services and point to the need for both technical and social solutions. We believe that debiasing knowledge graphs will become a pressing issue as these graphs enter everyday life rapidly. This is a provocative topic, not only from a technical perspective but because it will force us as a Semantic Web community to discuss whether we want to debias in the first place and who gets a say in how to do so.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 23 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.