2. Purpose
This presentation is a short introduction
to the different components of an ads
server that leverages semantic analysis
to segment and target audience.
Patrick Nicolas Copyright 2009-2011 - All rights reserved.
3. Semantics
A few definitions…
Contextual targeting is the process of
inserting the most appropriate advertising
into a published content (web pages, social
network, tweets, blogs..)
Taxonomy is the study or science of
classification of concept or concrete
items, in a logical and repeatable manner.
Patrick Nicolas Copyright 2009-2011 - All rights reserved.
4. Conundrum
Should targeting relies on
● audience preferences & behavior history
● content topic and style
● both?
Publisher
Content
User
Market
Advertiser
Promotion
Patrick Nicolas Copyright 2009-2011 - All rights reserved.
5. Taxonomy-based Targeting
It is assumed that the content
consumed by a visitor is reflective of
his/her
interests,
tastes
&
demographic characteristics.
Therefore, targeting (yield) consists of
analyzing content & extracting
context.
Patrick Nicolas Copyright 2009-2011 - All rights reserved.
6. Architecture
An ads targeting
components:
engine
two
key
● Optimizer
to
balance
objectives
(budget,
volume)
&
constraints
(placements, frequency, exclusivity,..)
● Dispatcher to select the ad with the
highest predicted yield according to
content
Patrick Nicolas Copyright 2009-2011 - All rights reserved.
7. Use Case
1. Campaign manager defines the objectives &
constraints
2. Optimizer computes the best promotion in the
inventory that satisfy constraints
3. Dispatcher formats & dispatches the promotion
Ads.
inventory
Campaign 1
1
Manager
Optimizer
1
2
1
Dispatcher 3
Content
Patrick Nicolas Copyright 2009-2011 - All rights reserved.
8. Semantic Analysis
1. The optimizer performs a semantic analysis of
both the content & promotional material.
1. The analysis generates taxonomy or semantic
classification graphs.
1. Finally, the promotion with the taxonomy
graph which is the closest to the content
graph is selected
Patrick Nicolas Copyright 2009-2011 - All rights reserved.
10. Test Results
The taxonomy match algorithm was
evaluated against a rule-based targeting
engine
for
consumer
discretionary
products.
The Click Through Rate (CTR) increased
from 0.8% to 0.193% (handbags) and
0.052% to 0.98% (upscale pen).
Patrick Nicolas Copyright 2009-2011 - All rights reserved.
11. References
● How much can Behavioral Targeting Help Online
Advertising? J. Yan, N. Lu, G. Wang, W. Zhang, Y Jiang
http://www2009.eprints.org/27/1/p261.pdf
● Introduction to Semantic Analysis
http://www.cs.tut.fi/sgn/arg/klap/introduction-semantics.pdf
Patrick Nicolas Copyright 2009-2011 - All rights reserved.