Skip to main content

Topic Classification — Reference

The Topic Classification model reads the pages behind the URLs in one of your event tables and labels each URL with a topic from an interest taxonomy. It is the classification step that IAB and Custom Interest rely on: Interest then scores each user's affinity for the topics of the pages they visited.

Unlike the other models, Topic Classification is not configured with JSON. When you pick it in the model setup, the Configure AI Model step opens a guided wizard, Configure Topic Classification, with three steps: Source data, Filters and Taxonomy. Your progress is saved as a Draft while you go.

Step 1 — Source data​

Website Access​

If the URLs you want classified sit behind a login, tick Website requires a login and fill in:

FieldWhat to enter
Login page URLThe full address of the page where you normally enter your credentials, starting with https:// (for example https://www.example.com/login). The model opens this page first and signs in, just like a visitor would, before reading the protected pages.
UsernameThe username or email used to sign in.
PasswordThe password for that account.

The step cannot be completed while a login is required and the login page URL is missing or not a valid http(s):// address, or no password has been provided.

:::info How BPP stores the credentials The password is stored encrypted and is never shown back — not in the setup, not in the model's configuration. When you edit a model that already has a password, the field is empty and reads Leave blank to keep the current password:

  • leave it blank to keep the stored password;
  • type a new one to replace it;
  • untick Website requires a login to remove the login, credentials included. :::

Use an account dedicated to BPP, with read-only access to the pages to classify, rather than a personal one.

Source Data​

Pick the source table and, from it, the URL column to classify and the event date column.

Step 2 — Filters​

Both filters are optional.

  • URL Filters — tick Would you like to apply some filters to the processed URLs?, choose a rule — Contains or Does not contain — and add the URL patterns (for example /blog/ or /cart). All patterns use the same rule: keep only the URLs that match, or skip them.
  • Source Filters — restrict which rows of the source table are processed, built like an audience filter. With more than one filter, Match ALL requires every filter (AND), ANY at least one (OR).

Step 3 — Taxonomy​

The Interest Taxonomy is the set of topics used to label your URLs. Either select an existing taxonomy table, or upload a new taxonomy:

  1. Give it a Taxonomy name (for example Retail interests 2026); BPP generates the table name from it.
  2. Use Download CSV template to see the expected format: a CSV with a category and a url column.
  3. Upload the .csv file (up to 10 MB, UTF-8 encoded). Each URL can appear only once — BPP checks the file and lists any duplicates to remove.
  4. Click Create table.

Next steps​

Once Topic Classification has run, configure an Interest model with the IAB or Custom interest type on the same page URLs.

Last updated on