Ideas on Conduct Collaborative Filtering for Recommendation Systems

Recommendation engines are becoming more and more useful for organizations in retail, education, and entrainment. There are two general types of recommendation engine:

1. Content-Based Filtering: an item is recommended based on a user’s own viewed and purchased histories. This requires tags or features to be created/obtained for different items.
2. Collaborative Filtering: an item is recommended based on similar users’ viewed and purchased histories.

A creative way to use collaborative filtering is to find similar users by matching user view-purchase groups rather than all views and purchases, and then recommend the current items, as well as past items and future items. This way, the time variable is considered. For feature engineering, the variables should be set up as below:

1. Contacts: Gender, Current Age, Location.
2. Histories: One variable will be items purchased by the user, and other variables are the number of views by items viewed BEFORE this purchase, and AFTER the previous purchase. Each row for the dataset would be such a viewed item + purchase combination (multiple for each user)

An example dataset:

					Product Viewed Before This Purchase & After Last Purchase
User	Date	Purchased	Gender (1/0)	Age	A1	A2	A3	A4	A5	A6	B1	B2	B3
User 1	2/2/20	A2	1	10	7	8	6	1	6	4	0	0	0
User 2	2/4/08	B3	0	20	0	0	0	0	0	0	6	1	8
User 2	2/8/20	B1	1	28	7	0	3	0	0	8	8	2	1
User 3	2/8/20	A6	0	20	8	6	8	7	0	5	8	8	5
User 3	2/1/20	A3	1	20	7	5	2	8	10	2	2	7	8