my problem is the following: I have purchase probabilitiy estimations of some products. The model behind don't take care of any inter-correlations through these products. So my task is to re-calibrate the probabilities and take inter-correlations into account:
Example of problem discription, there are for exmple two products recommanded for the same user:
Product 1: purchase prob. for user 1 is 50%
Product 2: purchase prob. for user 1 is 30%
PROBABLY: for user 1 product 2 is better to advertise him because of the underlying inter-corrleations AND NOT product 1 as it is recommanded through the model because of the higher probability.
My raw data consists of:
-> 1. matrix of useritems: consists of purchase probabilities
-> 2. matrix of useritems: consists of real pruchases (binary, but really sparse)
-> 3. matrix of user*items: consists of real click data (counts how often a user clicked on the product-page; also very sparse)
My question is now:
how to combine the data and is there any similar approch maybe you have seen in some papers? Until now my research was very unlucky. There are some ideas about scoring rules or hybrid recommander systems? But until know, I had not seen any similar task...
Kind regards
Dear check this factors
awareness (“Cognitive stage”) and preference (“Affective stage”) and purchase (“Conative stage”) towards a product or a brand, which could be clarified by below figure (Lavidge & Steiner,1961; Hutter et al., 2013)
If you are looking for a third factor to extract intercorrelation AND if the Nr.2 and Nr. 3 raw data are "sparse", an additional study for controlling them seams to be the only solution (besides poking about in a fog by datamining). Good luck!
- Badges
- Science topic
- Similar topics
- Mathematics
- Algebra
- Matrix