Glossary · Categories & taxonomy

What is a category confidence score?

Every category SellerSort suggests comes with a confidence score between 0 and 1. The score is banded — strong at 0.90 and above, fair from 0.85, weak below — and the same two boundaries decide which rows go into the ready-to-upload feed and which go into the file for review.

What this comes down to

  • Strong (≥ 0.90): safe to put straight into a feed.
  • Fair (0.85–0.90): probably right; worth a glance before submitting.
  • Weak (< 0.85): check it — a description or product type usually fixes it.
  • The runner-up is always shown, so a near tie looks like one.

SellerSort Team · Updated

The three bands

Confidence bands and what they route to
BandScoreAdviceBulk output
Strong0.90 and aboveSafe to put straight into a feedsupplemental-feed.csv
Fair0.85 to 0.90Worth a glance before you submitfull.csv only — held out of the feed
WeakBelow 0.85Check it; add a description or product typeneeds-review.csv

The boundaries are the constants the pipeline itself uses, so what the finder tells you and what the files contain cannot disagree. Fair rows sit in between on purpose: not confident enough to upload unseen, not doubtful enough to join the review queue. Filter full.csv on the confidence column to see them.

Separation from the runner-up

A score alone can hide a coin flip. The Category Finder also reports how far the answer is ahead of the next-best category — “clearly ahead”, “ahead, but not by much”, or “nearly tied” — and lists the alternatives. A nearly tied answer between two sibling categories is a prompt to decide, not a verdict.

The score is not a guarantee of what Google will do with the value; it measures how sure the classifier is that this taxonomy entry describes the product.

Questions

How do I raise a weak score?

Give the classifier more to go on. A short or vague title is the usual cause; adding the description, brand or your product_type typically moves the answer into the strong band.

Are fair matches wrong?

Usually not — they are probably right. They are kept out of the upload-ready feed because a glance is cheap and a wrong category at scale is not; you will find them in full.csv with their score.