Implicit Recommender model

User 3400 | 3/10/2016, 3:26:13 AM

I am presently working on Recommender systems based on click and purchase history. Basically I have the following information:

• User id • Merchant id • Number of times viewed • Number of times purchased

Descriptive stats:

Total users = 16376 Total merchants = 135 Total transactions = 28,383

An interesting observation to note is that 57% of the entire dataset is covered by top 10% merchants. So it’s highly skewed.
As of now I have defined an “Affinity” as the ratio of Purchase/View and tried to come up with a recommender model. Is it possible to include Click and Purchase details to account for negative affinity?

Any suggestions would be very helpful

Comments

User 1774 | 3/12/2016, 5:45:10 PM

Hi, 1. What is your goal? To recommend a user which merchant it should consider purchasing at? 2. Do you have the transaction timestamps, or simply how many time each user viewed/purchased from each merchant?