lightfm
lightfm copied to clipboard
Implicit positive and negative feedback
Hello, Thanks for the great package!
I am new to recommendation systems, and I wonder how could I model and specify negative implicit feedback using LightFM:
- In my problem, customers receive personalized products every number of months; they can keep products they like (purchase) and return the ones don't want (negative feedback)
- The rest will be just missing data (unknown products for customers)
For now, I am just removed from each customer recommendation list previously sent products that were not purchased. There may be better ways to include implicit negative information in the model, but I don't know if that is possible.
Thanks!
When creating your interaction matrix, you can use a negative value for returned items, zero for neutral and a positive value for purchased items as your rating column.
Got it, thanks. I will try with a mock example to check how it goes.