Learn to rank with prodigy

1

Dear support,
I would like to use prodigy for “learn to rank” pairwise, but I’m not totally sure to do it in the right way. I think the A/B Evaluation is the correct procedure to follow, isn’t? My data is results from a search engine and I would like to create a model to rank better the SERP against a query. I have some features like the page title, the heading text, the main content text and also the sentiment score of each page.

I would like to rank the pages by all these fields and I’m wondering how to do it, if it is feasible with prodigy as well?

Any suggestions are really appreciated.
Thanks in advance.

C.

2

I could think of a couple of ways to do it — it depends on how your model works probably?

The pairwise classification is a good approach. You could also suggest more candidates (like 3 or 4), and mark the best one with a multi-select.

I haven’t tried this yet though, so I’d be interested to hear how you go, especially if you get the active learning working as well.