I'd like to train a system that takes text and predicts IAB classes.

Are there any public datasets available for this?

  • I am planning to create a dataset for this and make it public - can you let me know your requirements if you still have them 4 years on :)
    – dendog
    Commented Mar 5, 2021 at 21:03
  • Awesome! I would still welcome a dataset for this. I think ideally it would be a a multi-label dataset and (in a perfect world) would be for paragraph-sized text or smaller since aggregating up would be easier than detangling down. As you get larger chunks the odds of needing multi-label would seem to increase too. Good luck and let me know if there's a repo or something to track.
    – Jim K.
    Commented Mar 9, 2021 at 11:26
  • @dendog Is your dataset available? Thanks a lot!
    – Nicolas Raoul
    Commented Mar 14 at 4:13
  • Sorry I never got round to this, plus I feel now with LLMs and synth data - this is sort of solved.
    – dendog
    Commented Mar 24 at 17:03

2 Answers 2


I think this is what you are looking for


if you want to validate your model you can use this free api to validate results of your model


you just need to provide attribution to https://front-page.com

  • Also take a look at commonscreens.com, they have the already categorized data of 77 million domains and provide a training data set also.
    – Bpm Tips
    Commented Jan 18, 2023 at 17:18

I was looking for that as well and did not find any so far. But this paper describes approach to use Wikipedia data for this.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.