Refining the Auto-Generated KG

"Sales in Springfield" returns one tidy number, and that number is wrong. The dataset holds four different Springfields in four different states, silently summed into one. Auto-generation gives you a good draft, not a finished model.

The lecture walks through the refinement pass: fixing entity identity, hiding technical classes, curating what answers show, letting the agent review the whole graph while the Creator confirms each change, and finishing with names, synonyms and descriptions.

Key points

• Draft, not result: auto-generation works from your schema alone, so reviewing and cleaning that draft is the Creator's core job.
• Unique identity: four different cities are all called Springfield, so the sql is adapted to city plus state and Veezoo creates one entity per city.
• Hide the plumbing: hidden takes technical classes such as Row ID out of the user's view.
• Curate answers: display_with sets what a class is shown with, and KB_ShowOnlyRepresenters drops the identifier column.
• Let the agent sweep: it reviews every file and proposes names, synonyms and descriptions, and the Creator confirms each one.
• Judgment stays with you: an average is the right default for a discount rate, because summing rates means nothing.
• Name for the business: name each field for what it means to users, so a warehouse column called "created_at" becomes Order Date.

Check your understanding