Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with
or
.
Download ZIP

Loading…

Improve search relevance #201

Closed
jeffhandley opened this Issue · 4 comments

5 participants

@jeffhandley

On the current nuget.org website, it's hard to search for packages by ID or Title. ID should be weighted greater than Title, which should be weighted greater than Tags, which should be weighted greater than Description.

If I search for 'jQuery' the official jQuery package should be the first result.

@xavierdecoster
Collaborator

Would be a very good improvement!
In addition, you might want to add some search filter options with proper defaults, e.g. include/exclude PreReleases in results.

@half-ogre

It's not a matter of searching by relevance so much as adding relevance based ordering to the list of current ordering options for search results (but keeping popularity, A-Z, recent, etc.). We would not want to lose the other ordering options, but adding relevance and making it the default would be great.

As I'm sure you know, relevance can be hard to get right. We spent lots of time on CodePlex's search, tweaking the weight of the relevance factors. I imagine they still do. For instance, how much does popularity count? Imagine I want to search on ORM, and there's a package with ID ORM, but what I really want is an ORM library. If EF and NHibernate are tagged ORM, and have ten of thousands of downloads, they should be pretty high in that list. But should the ORM package with 10 downloads be higher, or lower? Do you start tracking follow-through (i.e., what package I chose after a search) and giving that weight? Finding the right weights can be tough.

Going with just popularity (i.e., downloads) was a good (and cheap) first step.

This'll be fun to tackle, and we should look diligently at using a true search framework that meets our needs.

@pranavkm
Collaborator

I have a ongoing branch which includes using Lucene.Net to power both search results for the site as well as VS. One way to achieve @anglicangeek's popularity requirements is by adding a multiplier to a document (Lucene's equivalent of a row of data). If I get that working we could have both relevance working as necessary whilst having popular packages with similar relevance being ranked higher.

@TimLovellSmith

I think we can close this now that Lucene is in use?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Something went wrong with that request. Please try again.