Elasticsearch Learning to Rank Plugin Released


(Doug Turnbull) #1

Hey everyone, I wanted to let you know we've been working on an Elasticsearch Learning to Rank plugin as announced in this blog post

What is Learning to Rank? Learning to Rank uses machine learning to build self-learning search systems. With enough work, a learning to rank solution can self-learn relevance based on user behavior. You may have heard recently that Bloomberg released a learning to rank plugin for Solr. This plugin brings comparable functionality to Elasticsearch. It comes in two parts:

  • A script plugin for accepting learning to rank models (in this case models generated by the library Ranklib)
  • An ltr query used to rescore top N results using the learning to rank model

There's a full example included in the scripts directory, as walked through in detail in the blog post.

As always, feedback on our contribution would be very valuable. We'll be writing a lot more about using Learning to Rank with Elasticsearch in the weeks and months ahead.


(Ashley Mannix) #3

A couple of guys were talking about this internally and I haven't really seen a proper LambdaMART implementation so I'll definitely check it out :+1:

Thanks for sharing!


(system) #5