Wednesday, March 7, 2012

Factorized association

Hello,

Let’s say (for simplicity), in my site you can do one of two things – look at products and buy products.

I want to build an association structure between my products based on those two actions, but(!) when a user looks at two products it creates less important association than when the user actually bought those two products.

So basically, I want to give a different factor base on different actions occurred on my products.

How do I build my structure? How do I query it?

You could also build a single model with two nested tables, one for purchased products and one for browsed products. The browsed products would be input and the purchased input and predict. This would let you answer the question "people who browsed these products bought these products" and "people who purchased these products also purchased these products".

There currently is no way to set a weighting on this, however.

Another option is to simply build two models and interpret the results as you need to on your site.

|||

New, but somewhat similar problem-

Let's say - I’m running a music sharing site where users can rate the songs from 1 to 5.

How do I use this rating in my calculations for associating songs?

Meaning – the association between two songs is stronger when a user rate them both with 5, and weaker (even negative) when the rate is low.

No comments:

Post a Comment