One of the main problems in online advertising is to display ads which are relevant and appropriate \wrt what the user is looking for. Often search engines fail to reach this goal as they do not consider semantics attached to keywords. In this paper we propose a system that tackles the problem by two different angles: help (i) advertisers to create more efficient ads campaigns and (ii) ads providers to properly match ads content to keywords in search engines.
We exploit semantic relations stored in the DBpedia dataset and use an hybrid ranking system to rank keywords and to expand queries formulated by the user. Inputs of our ranking system are (i) the DBpedia dataset; (ii) external information sources such as classical search engine results and social tagging systems.
We compare our approach with other RDF similarity measures, proving the validity of our algorithm with an extensive evaluation involving real users.
Semantic Tags Generation and Retrieval for Online Advertising - CIKM 2010
1. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
SEMANTIC TAGS GENERATION AND RETRIEVAL
FOR ONLINE ADVERTISING
1Politecnico di Bari
Via Orabona, 4
70125 Bari (ITALY)
2University of Trento
Via Sommarive, 14
38100 Trento (ITALY)
Roberto Mirizzi1, Azzurra Ragone1,2,
Tommaso Di Noia1, Eugenio Di Sciascio1
2. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
Outline
Tags in Web 2.0 → 3.0
Computational advertising
NOT (Not Only Tag): semantic tag cloud
generation
DBpediaRanker: RDF ranking in DBpedia
Conclusion and Future work
3. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
Who is using tags nowadays?
and many
more…
4. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
What about Tags in Online Advertising?
5. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
BigG (& co.) helps you… in half (i)
…nice, but there is no
“semantics” in it.
You can not expand your
keywords list exploiting the
meaning of a term
(keyword/tag/query)
https://adwords.google.com/select/KeywordToolExternal
Keyword Tool
Based on actual Google
search queries
Generates keywords
based on the content of a
URL, words or phrases
1
2
3
6. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
BigG (& co.) helps you… in half (ii)
…nice, but there is no
“semantics” in it.
You can not expand your
keywords list exploiting the
meaning of a term
(keyword/tag/query)
Keyword Tool
Based on actual Google
search queries
Generates keywords
based on the content of a
URL, words or phrases
7. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
Why not to use Semantic tags?
Plugged into the Web 3.0
Disambiguation
Relations among tags
Machine understandable
NOT: Not Only Tag
http://sisinflab.poliba.it/not-only-tag/
8. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
NOT: Not Only Tag
Objectives
Assist advertisers to
create more efficient ads
campaigns
Support ads providers to
properly match ads
content to keywords in
search engines
Improve
advertiser experience and ad selection
9. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
What is behind NOT? (i)
10. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
What is behind NOT? (ii)
Comments
DBpedia resources are
highly interconnected
in the RDF graph
Not all the relevant
resources for a given
node are its direct
neighbors
1. Explore the
neighborhood of a
resource to discover
new relevant
resources not
directly connected to
it
2. Rank the results
11. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
DBpedia graph exploration in NOT
Open_source_CMS Web_application_frameworks
Content_management_systems Free_business_software …
…
Web_development Web_applications
JavaServer_Faces Python_web_application_frameworks
Zend_Framework
Joomla_extensions
skos:subject skos:broaderCategoryArticle
Legend
…
……
Magento
…
PHP
Drupal
…
12. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
The functional architecture
Back-end
Query engine
Storage
Tag Cloud
Generator
GUI
Ext.InfoSources
DBpedia
Lookup
Service
Interface
Delicious
Yahoo!
Bing
Google
Graph
Explorer
SPARQL
Context
Analyzer
Ranker
Offline computation
Linked Data graph
exploration
Rank nodes exploiting
external information
Store results as pairs of
nodes together with their
similarity
Runtime Search
Start typing a query
Query the system for
relevant tags
(corresponding to DBpedia
resources)
Show the semantic tag
cloud
1
2
3
1
2
3
OfflinecomputationRuntimesearch
1
2
3
1
2
3
13. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
DBpediaRanker: ranking
?r1 ?r2
isSimilar
v
hasValue
einfo_sourc2
21
1
21
einfo_sourc21
)(
),(
)(
),(
),(
rf
rrf
rf
rrf
rrsim
viceversaandrandrbetweenwikilink,2
saor viceverrandrbetweenkwikilin,1
randrbetweenwikilinkno,0
),(
21
21
21
21 rrorewikilinkSc
)(
),(
),(
2
12
21
rl
rrl
rroreabstractSc
Graph-based and text-based ranking
Ranking based on external sources
14. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
DBpediaRanker: an example (i)
wikilinkScore(Zend_Framework, PHP) = 2 abstractScore(Zend_Framework, PHP) = 1.0
15. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
DBpediaRanker: an example (ii)
sim(Zend_Framework, PHP)Google = 1.53e6 / 2.96e6 + 1.53e6 / 1.71e9 ≈ 0.52 + 0
delicious
16. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
DBpediaRanker: context analysis
The same similarity measure is used in the context analysis
?r1
?c1
belongsTo
v
hasValue
?c2
?c…
?cN
C
Example:
C = {Programming Languages, Databases, Software}
Does Dennis Ritchie belong to the given context?
Algorithm:
If(v>THRESHOLD) then
r1 belongs to the context;
add r1 to the graph exploration queue
Else
r1 does not belong to the context;
exclude r1 from graph exploration
EndIf
17. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
Evaluation (i)
http://sisinflab.poliba.it/evaluation
Comparison of 5 different algorithms
50 volunteers
Researchers in the ICT area
244 votes collected (on average 5 votes for each users)
Average time to vote: 1min and 40secs
18. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
Evaluation (ii)
http://sisinflab.poliba.it/evaluation/data
3.91 - Good
19. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
Conclusion
NOT: a prototype system for tag cloud generation in
semantic advertising
DBpediaRanker: ranking algorithms for resources in
DBpedia
Future work
Use the back-end of the system to develop new interfaces
for exploratory browsing
Improve ranking algorithms
Combine a content-based recommendation and a
collaborative-filtering approach
Develop a platform to test our system with real ads about
different domains
20. CIKM 2010 – 19th ACM Internation Conference on Information and Knowledge Management
October 29, 2010 – Fairmont Royal York, Toronto, Canada
Q&A
Thanks for your attention!
SEMANTIC TAGS GENERATION AND RETRIEVAL FOR ONLINE ADVERTISING (CIKM 2010)
If you're interested in learning more…
1. Roberto Mirizzi, Tommaso Di Noia. From Exploratory Search to Web Search and back. 4th Workshop for Ph.D. Students in Information
and Knowledge Management (PIKM 2010)
2. Roberto Mirizzi, Azzurra Ragone, Tommaso Di Noia, Eugenio Di Sciascio. Ranking the Linked Data: the case of DBpedia. 10th
International Conference on Web Engineering (ICWE 2010)
3. Roberto Mirizzi, Azzurra Ragone, Tommaso Di Noia, Eugenio Di Sciascio. Semantic tag cloud generation via DBpedia. 11th International
Conference on Electronic Commerce and Web Technologies (EC-Web 2010)
4. Roberto Mirizzi, Azzurra Ragone, Tommaso Di Noia, Eugenio Di Sciascio. Semantic tagging for crowd computing. 18th Italian
Symposium on Advanced Database Systems (SEBD 2010)
5. Roberto Mirizzi, Azzurra Ragone, Tommaso Di Noia, Eugenio Di Sciascio. Semantic Wonder Cloud: exploratory search in DBpedia. 2th
International Workshop on Semantic Web Information Management (SWIM 2010) - Best Workshop Paper at International Conference on
Web Engineering (ICWE 2010)
Roberto Mirizzi - mirizzi@deemail.poliba.it
See you tomorrow at PIKM 2010 in Room Alberta at 4pm with…
From Exploratory Search to Web Search and back