Collaborative filtering Interesting Problem

Hi All,

I am on building my own recommendation system after going through lesson 4 more deeply. I have a hypothetical store which gives credits on paying a credit card bill. Customers can accumulate credits by paying credit card bill every month which can be used to purchase different merchandise(Gift Cards etc).

My goal is to come up with a model which can recommend existing and new merchandise based on the history data to both new and existing customer. But my concern is @jeremy taught us to recommend new products to customer which they have not already bought it. Is there any way to cater to this problem as to how we can also recommend existing products.

My second ques is as to how to use other features while building Collaborative filtering model via fastai.Like jeremy taught us to use movie,user and id. But I have got a rich data set which talks about customers attribute like Credits earned,Credits burned,Historical attribute etc. Its analogous to situation where i have got a lot of movies and users attribute.

Thanks in advance for your help.


Any help guys

You may want to consider using tabular data approach (as described in Lesson 4). In that case you can have many features. I think Jeremy recommended this approach for the cold-start problem where you do not have any data when new items are introduced but I think you might be able to use it here too.

If you found a way of including additional features with the collaborative filtering approach please share your findings!

Were you able to implement this using fastai? I trying to work on a similar problem and cant figure out how to apply the fastai approach.