CROWDSOURCED TAXONOMY VERIFICATION: A FEEDBACK-DRIVEN FRAMEWORK FOR REFINING KNOWLEDGE GRAPH RELATIONSHIPS VIA ONLINE SEARCH INTERACTIONS

Summary from the TLDR Data newsletter

Grab treats knowledge-graph relationships as hypotheses that can be validated with live search behavior. Candidate parent, child, and sibling relationships are injected into suggestions, then clicks, dwell time, scrolls, and conversions are aggregated into confidence scores that promote or prune taxonomy edges.