Problem Statement 5:
Amazon Advertising
Challenge
Advertising is an ever-growing business, forming the core revenue of multiple organizations.
Despite its ubiquitous presence the decision of choosing the perfect advertisement has
always been a challenge. There are multiple tenets to target this problem and at Amazon we
have various kinds of Ads such as targeted and contextual ads which cover different grounds.
In this challenge, we target
contextual ads
which are displayed on websites ranging from
product reviews to personal blogs. The main principle behind this category of ads is that they
are relevant to the content of the website they are rendered on and thus the measure of this
relevancy comes into picture.
We present the following challenges which are key to solving the bigger problem of
relevancy of ads.
1)
Determining the product centricity/specificity of a webpage
Owing to the myriad category of pages we serve at Amazon, the spectrum of content
we deal with is ultra-wide. A page may be focused on anything from a cryptic personal
blog with no mention of a single product to a world-famous tech blog that talks about
nothing but a product. This poses a problem for Amazon since our ads are almost
always about a product. The products themselves may be one of the 12 million+
products available on Amazon from various categories or GLs as we like to call them.
It is thus, desirable to have a system which is able to determine the nature of a website
with respect to this 'product centricity'.
The primary
input
to this system will be the URL/Page source of a webpage.
Amazon will also provide a set of URLs which match the description.
The
output
should at the least contain information about how much this page is
centered around a product.
( [0,1]; 1 being highly product centric and 0 being not at all product centric )
Any
other
information apart from product centricity that aids in determining the
sought after perfect and relevant ad is appreciated.
Example:
Input:
http://booksanddell.blogspot.ie/2017/03/samsung-galaxy-review.html
Output:
Higher score on product centricity scale; maybe 0.98
Input:
https://www.onlinebigbrother.com/big-brother-18-week-5-summary-and-live-
eviction-results/
Output:
Lower score on product centricity scale; maybe 0.04
