5. Aardvark
Aardvark is social search engine
$50Million
Reference :
- http://techcrunch.com/2010/02/11/google-acquires-aardvark-for-50-million
6. What is Social Search?
the incorporation of information about social
networks and relationships into the information
retrieval process
“Social search” is an umbrella term used to
describe search acts that make use of social
interactions with others.
Reference :
- Sebastian Marius Kirsch, “Social Information Retrieval”, Rheinische Friedrich-Wilhelms-Universität Bonn, 2005
- Einat Amitay et al, “Social Search and Discover Using Unified Approach”, ACM, 2009
- David Carmel et al, “Personalized Social Search Based on the User’s Social Network”, ACM, 2009
- Meredith Ringel Morris et al, “What Do People Ask Their Social Networks, and Why?”, CHI, 2010
- Brynn M Evans, Ed H. Chi, “Towards a Model of Understanding Social Search”, ACM, 2010
7. What is Social Network?
A social structure made of individual called
‘nodes’, which are connected by one or more
specific types of interdependency, such as
friendship, common interest, financial exchange,
dislike or prestige.
Reference :
- http://en.wikipedia.org/wiki/Social_network
- http://en.wikipedia.org/wiki/Social_networking
8. Goal of Social Search
To use the social information to improve
the user’s search experience over regular
full text search.
Social search(a.k.a social information retrieval)
= social networks + information search
Reference :
- Sebastian Marius Kirsch, “Social Information Retrieval”, Rheinische Friedrich-Wilhelms-Universität Bonn, 2005
- Einat Amitay et al, “Social Search and Discover Using Unified Approach”, ACM, 2009
9. Previous Studies
Social Network Based Search
- PeopleRank, SocialRank, SNDocRank, Aardvark
Social Annotation Based Search
- FolkRank, SocialBookmarkRank, SocialSimRank,
SocialPageRank
Reference :
- Andreas Hotho et al, “Information Retrieval in Folksonomies”, The Semantic Web, 2006
- Yusuke Yanbe et al, “Can Social Bookmarking Enhance Search in the Web?”, ACM, 2007
- Shenghua Bao et, al, “Optimizing Web Search Using Social Annotations”, WWW2007, 2007
- David Carmel et al, “Personalized Social Search Based on the User’s Social Network”, ACM, 2009
- Liang Gou et al, “SNDocRank: a social network-based video search ranking framework”, ACM, 2010
- Abderrahmen Mtibaa et al, “PeopleRank: Social Opportunistic Forwarding”, IEEE, 2010
10. Aardvark
Aardvark is social search engine
- Social Network Based Search
- Deployed and validated in large-scale
12. Which one is better?
Aardvark
vs
<Anonymous group> <My Friends (of the friends)>
13. Social network rocks!
• More Reliable
• More Satisfactory
• Even More Responsible
Reference :
- Damon Horowitz, “The Anatomy of a Large-Scale Social Search Engine”, WWW2010, 2010
- Meredith Ringel Morris et al, “What Do People Ask Their Social Networks, and Why?”, ACM, 2010
- David Carmel et al, “Personalized Social Search Based on the User’s Social Network”, ACM, 2009
16. Structure of Aardvark
Sign-up time
DB
Importer Index
User, Topic,
Graph, Content
at query time
Conversation Routing
Gateway
Manager Engine
Reference :
- Damon Horowitz, “The Anatomy of a Large-Scale Social Search Engine”, WWW2010, 2010
17. First solution :
We deal with user’s up-to-date topic
Your
up-to-date topic
does matter
Reference :
- Xuerui Wang, Andrew McCallum, “Topics over Time”, ACM, 2006
- Andreas Hotho et al, “Trend Detection in Folksonomies”, SAMT, 2006
- http://burak-arikan.com/growth-of-a-twitter-graph, “Growth of a twitter graph”, 2010
18. Structure of Aardwolf
Up-to-date Topics Up-to-date
Topic
Aggregator
Social Annotation Service
at query time Up-to-date Topics
Conversation Routing
Gateway
Manager Engine
19. Why ?
Social Annotation Service
• Tag(Annotation) as
topic
• Easiest way to get
user’s up-to-date
topic
Reference :
- Andreas Hotho et al, “Information Retrieval in Folksonomies”, The Semantic Web, 2006
- Yusuke Yanbe et al, “Can Social Bookmarking Enhance Search in the Web?”, ACM, 2007
- Shenghua Bao et, al, “Optimizing Web Search Using Social Annotations”, WWW2007, 2007
20. Ranking process of Aardvark
Gateway
Query : “what would be perfect for my girlfriend’s present?”
Routing
Engine
Rank users by algorithm of Aardvark
Users : Ranked by their social information
Conversation
Manager
21. Ranking Algorithm
User’s answerable probability
Intimacy Expertise
Topic Assignment
Topic Expertise
Reference :
- Damon Horowitz, “The Anatomy of a Large-Scale Social Search Engine”, WWW2010, 2010
22. Ranking process of Aardwolf
Gateway
Query : “what would be perfect for my girlfriend’s present?”
Up-to-date Topic
Aggregator
Up-to-date Topics Social Annotation Service
Routing
Engine
Rank users by algorithm of Aardwolf
Users : Ranked by their social information
Conversation
Manager
23. Ranking Algorithm
Aardvark Topic Expertise
Aardwolf
Topic Up-to-date Topic
Reference :
- Damon Horowitz, “The Anatomy of a Large-Scale Social Search Engine”, WWW2010, 2010
24. Second solution :
We deal with user’s knowledge of location
• User’s knowledge of location
25. Structure of Aardwolf
Location History Location
Manager
Location sharing SNS
at query time
Conversation Routing Up-to-date
Gateway Topic
Manager Engine Aggregator
26. What is the ?
Location Sharing SNS
User’s
knowledge
of Location
27. Ranking process of Aardvark
Gateway
Query : “where is the best restaurant in Kang-Nam, Seoul ?”
Routing
Engine
1. Select users if their profile mentions Seoul
2. Rank users by algorithm of Aardvark
Users : Ranked by their social information
Conversation
Manager
28. Re-ranking process of Aardwolf
Gateway
Query : “where is the best restaurant in Kang-nam, Seoul ?”
Location
Location history Manager
Routing
Engine
1. Select users if they have been Kang-nam recetly.
2. Rank users by algorithm of Aardwolf
3. Re-rank users by their location score
Users : re-ranked by their location score
Conversation
Manager
31. Scenario
and his girlfriend have been seeing
B A A
each other for 3 years.
Me has been restaurant reviewer for several
C
years.
D C Me a big fan of Aardwolf.
“where is the best restaurant for Japanese
Ramen in Kang-nam, Seoul?”
32. Ranking process of Aardwolf
Gateway
Query : “where is the best Japanese Ramen restaurant in Kang-nam, Seoul ?”
Location
Location history Manager
Routing
Engine
1. Select users if they have been Kang-nam recetly.
Users : re-ranked by their location history
Conversation
Manager
33. Select users
by their location, Kang-nam, Seoul
Profile Foursquare
B A
A Seoul
Me
B Suwon
Kang-nam, Seoul
C Seoul
D C
D Seoul
Aardvark Aardwolf
A C D A B C D
34. Ranking process of Aardwolf
Gateway
Query : “where is the best Japanese Ramen restaurant in Kang-Nam, Seoul ?”
1. Select users if they have been kang-nam recetly.
Routing Up-to-date Topic
Engine Aggregator
Recent User Topics Social Annotation Service
2. Rank users by algorithm of Aardwolf
Users : re-ranked by their location history
Conversation
Manager
35. Rank :
Consider Change of Topic , Ramen
Up-to-date Aardvark Aardwolf
B A All Topic
Topic Score Score
A 26 8 0.5 0.30
Me
B 5 2 N/A 0.06
D C C 17 26 0.33 0.39
D 8 20 0.17 0.25
Aardvark Aardwolf
A C D C A D B
36. Re-ranking process of Aardwolf
Gateway
Query : “where is the best Japanese Ramen restaurant in Kang-Nam, Seoul ?”
Location
Location history Manager
Routing
Engine
1. Select users if they have been kang-nam recetly.
2. Rank users by algorithm of Aardwolf
3. Re-rank users by their location use count
Users : re-ranked by their location history
Conversation
Manager
37. Re-Rank :
Consider Location History, Kang-nam
Location Use
Rank score Re-rank score
B A Count, Kang-nam
A 5 0.3 0.18
Me
B 20 0.6 0.13
C 50 0.39 0.44
D C
D 120 0.25 0.74
Ranked Re-ranked
C A D B D C A B
38. Aardwolf rocks!
Up-to-date Aardvark Aardwolf
Profile Foursquare Topic LUC
Topic score score
A Seoul Kang-nam 26 8 5 0.5 0.18
B Suwon Kang-nam 5 2 20 N/A 0.13
C Seoul Kang-nam 17 26 50 0.33 0.44
D Seoul Kang-nam 8 20 120 0.17 0.74
Aardvark Aardwolf
A C D D C A B
40. Work in Progress
• Modified ranking algorithm to consider user’s
up-to-date topic
• Added re-ranking algorithm to deal with user’s
knowledge of location
Future work
• Implement the Aardwolf social search engine
• Experiment with the real data of Twitter,
Delicious and Foursquare