Resolving Reality: The Hard Problem of Entity Resolution in Knowledge Graphs

Session track: Data Intelligence

Session time: 04:30 pm - 05:00 pm UTC November 12

Session description:

With Neo4j Document Intelligence, turning unstructured documents into a knowledge graph has never been easier. But getting from raw extraction to a graph you can actually trust hinges on a question that sounds trivial and rarely is: when do two extracted nodes refer to the same real-world entity? This is entity resolution, and it's one of the hardest problems in the entire pipeline - and it's never truly "done." In this session, we'll unpack the real challenges: ambiguous and conflicting evidence, the precision-versus-recall tension, scale that makes naive comparison impossible, and the cases where the "right" answer depends on a judgment call. Using entity resolution within Neo4j's Document Intelligence as our running example, we'll walk through the pipeline layer by layer: blocking to cut the comparison space down to plausible candidates, matching to score candidate pairs, clustering to turn pairwise matches into coherent groups, and merging to produce the canonical entities your graph can rely on. You'll leave with a clear picture of what each layer does, where it tends to break, the remaining challenges, and a closer look at ongoing work.