search

graphaware / neo4j-reco

Neo4j-based recommendation engine module with real-time and pre-computed recommendations.
374 stars 77 forks source link

How to implement k-Nearest Neighbour similarity #11

Closed tolomaus closed 9 years ago

tolomaus commented 9 years ago

Hi,

I have been playing with this awesome recommenation engine (see my recommendations github repo) to calculate recommendations with a collaboration-based and a content-based strategy.

Now I'm wondering how I could implement a k-Nearest Neighbour similarity with the engine, as explained in this article.

bachmanm commented 9 years ago

This is something I would personally code as a "background" job using the GraphAware Framework's Timer-Driven functionality. A reference implementation of that is our NodeRank module. We may get around to implementing it at some point, but a pull request from you will be welcome of you feel like it! Cheers!

tolomaus commented 9 years ago

Hi Michal,

Indeed we could calculate the similarity in the background with some "NodeRankBykNN" module and then use this similarity index in a recommendation module.

Thanks for the pointer. If I get around to implement it I will definitely send it as a PR.

Niek.