1. Understanding and Organizing User Generated Data
Methods and Applications

2. August 16, 1977
June 25, 2009

3. August 16, 1977

5. June 25, 2009

7. officially pronounced dead

8. Media
Social
In 1977, only few people had the possibility to address others by means of Radio and TV. They broadcasted to the public whatever they felt was interesting.
Nowadays things look different. People are networked and got directly involved in the information flow. Similarly, media items are interconnected, and they are both, produced and cosumed by people. A significant amount of the information flow nowadays occurs over the internet, using computers, and, more and more, mobile devices. The result are novel data structures that demand being investigated. In fact, this is the context this thesis lives in. We try to investigate and exploit the data and structures that have emerged from the use of social media.

9. Results that are directly applicable in end-user services
Part 1: Direct Links
This talk:
Results that are directly applicable in end-user services
Part 2: Similarity

10. Part 1: Direct Links

11. high clustering
Probability that two of my friends are (becoming) friends themselves is high!

12. VENETA: Friend Finding

13. I want to meet people!
same contact = friend of a friend
privacy preserving!

14. Cluestr: Contact Recommendation

15. Clustering Survey: Communities are often addressed as groups!
„There‘s no training tonight!“
„Let‘s have a BBQ tomorrow!“
„Our next meeting is at 2pm!“
Survey:
Communities are often addressed as groups!
Clustering

16. Recommend contacts from clusters of already selected contacts
Clustering
Communities can be identified using clustering algorithm

17. Considerable time savings possible!
updated group
recommended contacts
Group
(i.e. „invited“ contacts)
new recommendations
Considerable time savings possible!

18. Part 2: Similarity

20. Academic Conferences

21. Similarity between Scientific Conferences
publication
conference
author

22. Confsearch (Screenshot)
Highlight Ratings
Highlight Related Conference Search

23. Music Similarity

24. How similar is Michael Jackson to Elvis Presley?

25. Audio Analysis
Usage Data

26. „Users who listen to Elvis also listen to ...“
#common users
(co-occurrences)
Occurrences of song A
Occurrences of song B
Problem: Only pairwise similarity, but no global view!

27. 1 2 3 Getting a global view... d = ? pairwise similarities
graph for all-pairs
distances (shortest path) 2
MDS to embed graph into Euclidean space
while approximately preserving distances (=> global view) 3

28. Classical Multidimensional Scaling (MDS)
Principal Component Analysis (PCA):
Project on hyperplane that maximizes variance.
Computed by solving an eigenvalue problem.
Basic idea of MDS:
Assume that the exact positions y1,...,yN in a high-dimensional space are given.
It can be shown that knowing only the distances d(yi, yj) between points we can calculate the same result as applying PCA to y1,...,yN.
Problem: Complexity O(n2 log n)
use approximation: LMDS [da Silva and Tenenbaum, 2002]

29. estimator for distance: Remove edges that get
Use embedding as
estimator for distance:
Remove edges that get
stretched most and re-embed
Problem:
Some links erroneously shortcut certain paths
Example: Kleinberg graph (20x20 grid with random edges)
Original embedding
(spring embedder)
After 6 rounds
After 12 rounds
After 30 rounds
Kleinberg graph: grid edges + one random edge per node (follows 1/d^2 distribution, d=manhattan distance)

31. Play „random“ songs that match my mood!
skip =
listen =

32. which songs match the user‘s mood
Realization using our map?
After only few skips, we know pretty well
which songs match the user‘s mood

33. Browsing Covers
„In my shelf AC/DC is
next to the ZZ Top...“

34. Conclusion
„from users for users“

35. Thank you

36. List of Publications
Social Audio Features for Advanced Music Retrieval Interfaces
M. Kuhn, R. Wattenhofer, S. Welten
Multimedia 2010
Visually and Acoustically Exploring the High-Dimensional Space of Music
L. Bossard, M. Kuhn, R. Wattenhofer
SocialCom 2009
Cluestr: Mobile Social Networking for Enhanced Group Communication
R. Grob, M. Kuhn, R. Wattenhofer, M. Wirz
GROUP 2009
From Web to Map: Exploring the World of Music
O. Goussevskaia, M. Kuhn, M. Lorenzi, R. Wattenhofer
WI 2008
VENETA: Serverless Friend-of-Friend Detection in Mobile Social Networking
M. von Arb, M. Bader, M. Kuhn, R. Wattenhofer
WiMob 2008
Exploring Music Collections on Mobile Devices
O. Goussevskaia, M. Kuhn, R. Wattenhofer
MobileHCI 2008
The Layered World of Scientific Conferences
M. Kuhn and R. Wattenhofer
APWeb 2008
The Theoretic Center of Computer Science
M. Kuhn and R. Wattenhofer. (Invited paper)
SIGACT News, December 2007
Layers and Hierarchies in Real Virtual Networks
WI 2007