Data Structure and approach for recommendations system
Posted on 2015-01-26
I am working on a simple recommendations system. Conceptually the bare bones structure (using relational database terms because thats what I am familiar with) is:
I then want to, for a given recommender_id rank all other recommender_ids by their similarity (e.g. if recommender_id 1 scored item_id 1 as 10 and item_id 100 as 0 then another recommender ranking 1 as 9 and 100 as 1 will rank higher than a recommender scoring 10 for 1 and 8 for 100).
I expect this to be a quite large sparse matrix, possibly tens of thousands on each axis.
I would like to get the approach right from the start and would like recommendations for how to best structure the data and which language / tools to use.