Amazon Customer Insight Brainstorm (1)
I read about the Amazon applied scientist position related customer analytics. They would like to recruit someone who can help develop or evaluate algorithm that generates the recommendations for their customers. As a six-year faithful customer of Amazon, the customer recommendation problem is intriguing to me, which I have a few thoughts about.
- The first algorithm comes into mind is collective filtering, which was applied to move rating and recommendation as well.
- However, that simple thought ignores the fact Amazon has much richer data, both product and customer feedback wise than the simple movie rating
- product: Amazon have countless categories of goods, from physical to digital, from amazon own stock to third-party seller, from prime to non-prime etc.
- customer: Various age, gender, prime v.s. non-prime, subcriber v.s non-subscriber
- customer feedback: rating, text comment, return or not, purchase frequency, search history, what they put into wishlist/saveforlater, their last order date, which card they were using and etc. which reflects the significantly dynamic and versatile nature of customer behaviors
- So, re-considering this question, some sort of complex graph forms in my mind, with vertices of customer and product and product charateristics, and with various edges indicating relationship between the customer and product and common characterics shared by various product
- Then, I put myself as customer (yeah, indeed, I purchase from time to time from Amazon) who is recently searching a lot for baby stuff to think through the whole process, when I was trying to by Pampers Sensitive Wipes for my little girl
- I searched “Pampers wipe” first, the first page did not show the results I need
- then I changed the key word to “Pamperps sensitive wipes” and pick the first one to have closer look and choose a proper size to order
- as subcription saves money, I decided to add this to my subscription list, using my husband’s card as the payment
- I saw the “pampers diaper” in the recommendation list, which at once was added to my subscription as well
- before I signed out, I also checked my wishlist and saveforlater, making sure I did not leave any small wanted item there for combined free shipping.
- …this list can go on for a while, depending on how well Amazon is predicting my need
- What is the final solution for this problem? Deeper and more quantitative measures about customer behavior, reasonal metrics derived from graph, sensitive detection of temporally and spacilly variated customer behavior, willgenerate greater insights for product recommendation
- Will revisit this problem when I finish my graphy theory reading