Using machine learning to rank search results (part 1)

A large catalog of products can be daunting for users. Providing a very fine grained filtering of search results can be counter-productive: it leads them from information overload to lack of choice.

On e-commerce sites, this results in poor conversion—users leaving the site without checking out.

The key is obviously to provide relevance and choice, which is much more complicated than it sounds, as different users may have very different tastes.

This describes how I explored a machine learning, neural networks based solution to relevance ranking.

We’ll demonstrate that an ANN (artificial neural networks) based approach can provide better ranking than our historical ranking heuristics, and still provide good performance.

Read the complete article on my blog.

Credits: Thanks to Alfredo Motta and Andy Shipman for their helpful review of this article.

photo of Julien Letessier

Software Engineer
comments powered by Disqus