Skip to main content

MAP-1 — Map columns to concepts & give feedback (UI)

Task Library primitive. Route A of mapping — interactive training in the UI. Client-neutral.

ModuleClassify
Screen pathData Sets → [Set] → Attributes / Train Object Model; and Data Catalogs → ad-hoc mappings (Object & Concept level)
Type[repeatable] until the model is trusted
PrerequisitesA catalog (MOD-*) and registered data sets (ACQ-2/ACQ-3); classification must have run at least once to surface predictions.
ProducesConfirmed/curated mappings between data-set columns and catalog concepts, feeding the supervised classifier.
Primary roleData Solution Analyst / SME (feedback).
StatusDoc

Purpose​

Train the catalog by confirming, correcting, or rejecting the system's predicted mappings between columns and concepts. This is the no-file route — appropriate when the source-to-target map is small or still being discovered. (For a known, offline-analysed map, import it instead — MAP-2/MAP-3.)

Mapping types you'll see​

  • Predicted (ad-hoc) — tenant-wide model predictions; lower trust.
  • Training (ad-hoc) — provided specifically for training.
  • Manual — created when a user upvotes or links a concept/object to a column/data set.

Steps — give feedback at concept level​

  1. Open the catalog and view ad-hoc mappings at the Concept level (or use MOD-4 Technical View).
  2. For an accurate prediction, upvote. For a wrong one, click the downvote; the popup lets you: keep the existing mapping, downvote and suggest an alternative (counts as an upvote for that alternative), or downvote with no alternative (drops the prediction's score to zero).
  3. Repeat at the Semantic Object level for data-set-to-object mappings (same mechanics, coarser grain).
  4. Click Run Model in the footer to incorporate pending feedback and re-predict across the tenant. Downvoted mappings fall off after this run.

How the score works​

A mapping's score = latest unique upvotes ÷ latest total votes (or the system-generated score when no feedback exists yet). Counts next to the percentage show how many users agreed.

Gotchas​

  • Feedback isn't live until Run Model. Votes are pending until you run the tenant classification (MAP-4); only then do downvoted mappings drop and scores recompute.
  • You can't vote twice — the most recent feedback per user is the one counted; history is viewable from the vote counts.
  • Run Model is tenant-wide — it re-predicts everything and may move unrelated scores; expect a job, not an instant update.
  • PII-tagged concepts mask data for Read-Only users in samples/profiles (tagging), which can affect what reviewers see.
  • MAP-2 / MAP-3 — the import route (offline-analysed maps).
  • MAP-4 Run the classification model — applies pending feedback.
  • MOD-4 View object model — an alternate surface for the same feedback.
  • NRM-4 Train in UI — the project-level analogue of this interactive training.