Adaptive, Personalized Diversity for Visual Discovery
Choon Hui Teo
Amazon
choonhui@amazon.com
Houssam Nassif
Amazon
houssamn@amazon.com
Daniel Hill
Amazon
daniehil@amazon.com
Sriram Srinavasan
UC Santa Cruz
ssriniv9@ucsc.edu
Mitchell Goodman
Amazon
migood@amazon.com
Vijai Mohan
Amazon
vijaim@amazon.com
S. V. N. Vishwanathan
Amazon & UC Santa Cruz
vishy@amazon.com
ABSTRACT
Search queries are appropriate when users have explicit in-
tent, but they perform poorly when the intent is difficult
to express or if the user is simply looking to be inspired.
Visual browsing systems allow e-commerce platforms to ad-
dress these scenarios while offering the user an engaging
shopping experience. Here we explore extensions in the di-
rection of adaptive personalization and item diversification
within Stream, a new form of visual browsing and discovery
by Amazon. Our system presents the user with a diverse set
of interesting items while adapting to user interactions. Our
solution consists of three components (1) a Bayesian regres-
sion model for scoring the relevance of items while leverag-
ing uncertainty, (2) a submodular diversification framework
that re-ranks the top scoring items based on category, and
(3) personalized category preferences learned from the user’s
behavior. When tested on live traffic, our algorithms show
a strong lift in click-through-rate and session duration.
Keywords
Machine Learning; Submodular Functions; Diversity; Per-
sonalization; Explore-Exploit; Multi-Armed Bandits
1. INTRODUCTION
The brick-and-mortar shopping experience is character-
ized by visual browsing where the user is able to quickly
scan a large number of potential purchases. The user has
a high potential to discover new items while maintaining
the ability to focus attention on items of particular inter-
est. This problem of discoverability is more challenging in
e-commerce where it can be difficult to expose the entirety
of an online retailer’s catalog. The in-store browsing expe-
rience is not well-replicated by search engines that restrict
item discovery to items relevant to an explicit search query.
Permission to make digital or hard copies of part or all of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full citation
on the first page. Copyrights for third-party components of this work must be honored.
For all other uses, contact the owner/author(s).
RecSys ’16 September 15-19, 2016, Boston , MA, USA
c
2016 Copyright held by the owner/author(s).
ACM ISBN 978-1-4503-4035-9/16/09.
DOI: http://dx.doi.org/10.1145/2959100.2959171
Therefore, an online visual browsing experience may greatly
aid users in item discovery.
One effort in this direction is Amazon Stream (figure 1),
a new website for fashion discovery developed by Amazon
(www.amazon.com/stream). Stream enables users to eas-
ily discover popular, new, and relevant fashionable items
without the need for search queries or to sieve through less
relevant items. Toward this end, we have built a personal-
izable system that ranks and diversifies items scored by an
explore-exploit algorithm. These items are presented to the
user as an infinite scroll where each item can be interacted
with by clicking. This paper outlines elements of our diverse
and personalized visual shopping experience approach.
Figure 1: Screenshot of www.amazon.com/stream.
2. SCORING ITEM RELEVANCE
As the first step in generating the stream, we score each
item in the Stream catalog for relevance to our users. We
use click probability P (click | item is viewed) to quantify
relevance. Click refers to any of the following activities:
save item as favorite, visit the item’s detail page, or open
the modal window for the item.
We learn this probability distribution by using a Bayesian