pitasi / idem

"You may also like..."

Idem

A simple recommend engine for anything - based on likes, Facebook-style.

This basic implementation came out while I was having a shower. There's plenty of room for improvements - any suggestion is welcome, so please continue reading.

Rationale

The two entities are simple: users and things.

An user can like any number of things.

Case 1: Based on the things the other users you want to suggest to a certain user something it may like too.

Case 2: Based on a thing, you want to list other things that are similar and that the users may likes too.

Todo

• Case 1
• Case 2
• Reduce used memory by storing only half of the graph (instead of storing A->B and B->A we should exploit the symmetry of the problem to just store A-B).
• Build a simple webserver with a small set of HTTP endpoints (e.g. GET /suggestions/{user}, POST /like/{user}/{thing})
• Make some serialization/deserialization available for persistence. Still not sure if picking a dabatase, use something embedded like Bolt, or just be agnostic.

Note

I don't need this to be based on some NN or be super-fast. I plan to use this on a small set of available things and to update the cache of suggestions every once in a while.