https://drive.google.com/file/d/1SZsz2tn8ku7wJCl_XgjSze3jANf-S1Kk/view?usp=share_link

A large portion of enterprise knowledge exists in unstructured documents such as reports, filings, contracts, and research papers. Transforming these documents into knowledge graphs is challenging because unstructured data lacks the rigor of structured data pipelines. Structured enterprise data collection is carefully designed and governed, with predefined schemas, validated entities, and controlled updates. Unstructured documents, by contrast, contain ambiguity, evolving terminology, and incomplete context. Directly mapping such information into rigid schemas often forces uncertain observations into seemingly certain facts, risking the loss of important context.

This workshop introduces Knowledge Maps, an evidence-first approach to building knowledge graphs from unstructured data while preserving context and provenance. Instead of collapsing extracted information into canonical facts, entities and relationships act as navigational handles linking directly back to supporting evidence in source documents.

Participants will work with a curated dataset and use modern AI tools to design extraction pipelines, develop graph schemas, and construct knowledge maps. The workshop will also explore deferred entity resolution, resolving entity identities during analysis—when richer context is available—rather than prematurely during ingestion. The session concludes with interactive graph exploration and analysis.

This workshop will be taught together with Mengjia Kang from JPMorgan Chase.