Hi,
I have built two MVPs (Minimum Viable Product) with Meteor. One being a Ebay Clone for a specific country where the auction market is totally empty. I am planning to use one of the many services that track user Data like https://www.optimizely.com/ .
But beyond that. Should I use something like this http://www.seldon.io/ to emulate EBay’s recommendation engine ( they call it customer DNA).
Thanks in Advance.
quick search shows there are some npm packages as well as one of our own,
https://atmospherejs.com/samcorcos/recengine
Damn, its our @samcorcos.
Ha! Sadly it doesn’t work anymore. Fundamentally, you cannot build an
efficient recommendation engine with a non-relational database (as far as I
can tell).
I would suggest looking into Neo4j and keeping a second set of database
entries that you use just for your recommendation engine. If you’re looking
for something like ebay’s, check out collaborative filtering. Coursera has
a great course on recommendation systems.
Lots of companies keep more than one database for special use cases. I
happen to know that Medium has a neo4j database just for keeping track of
relationships between users.
I might be wrong, depend on how you implement it, you might still able to use NoSql. Lookup the term collaborate filtering for recommend system, and it does require a extra db to store the calculated relationship
So, I have been looking into this recently and it seems the best available solution is PredictionIo as it is like meteor for AI stack. This is a well tested stack which is used by companies like dropbox in production which means your not going to face scaling issues. But, the bigger question, IMHO is more about architecture and the best way to integrate it with meteor so that the whole setup can be scaled and this requires some thought.
Hopefully this is helpful, also PredictionIo also has recommendation template for e-commerce, so you can probably use them out of box with little to no tweaking.