Stephan Baumann
(DFKI, ArtScientist, Ph.D Artificial Listening
Systems, Music Retrieval & Social Media Experte)
Inhaltsbasiert, kollaborativ, sozial, egal wie man es dreht, zeitgenössische
Aggregations- und Filterungsmechanismen, die uns zum Musikgenuss- gar
Kauf bewegen sollen, erzeugen E!ekte, die man mögen oder hassen kann.
Ginge es auch anders? Hochselektiv, reduziert, emotional und
wieviel Technologie ist hierbei im Spiel.
16. Collaborative Filtering:
Implizit/Explizit per User
Stephan Martin Dominik!
Song Nr.1 60mal / Tag - 99mal / 2005!
Song Nr.2 3 mal / Woche 4mal / 2005 -!
Song Nr.3 1 mal / 2006 20mal / Woche -!
HORST 17.09.2009
23. Empfehlungsmaschinen
funktionieren ... gut genug?
[MagnoSable ISMIR2008] A Comparison of
Signal-based Music Recommendation to Genre
Labels, Collaborative Filtering, Musicological
Analysis, Human Recommendation, and
Random Baseline
!!! 15 Probanden !!! YES
HORST 17.09.2009
24. Empfehlungsmaschinen
funktionieren ... gut genug?
[PachetRoy ISMIR2008] Hit Song Science is not
yet a Science
!!! 32.000 Songs, HIFIND Experten Daten!!! NO
HORST 17.09.2009
25. Empfehlungsmaschinen
funktionieren ... gut genug?
[GodfreyChordia ISMIR2008] Hubs and
Homogenity: Improving Content-Based Music
Modeling
!!! 617 Songs, OpenNap Experten Daten!!! ?
HORST 17.09.2009
26. Empfehlungsmaschinen
funktionieren ... gut genug?
[Salganik et al SCIENCE2006] Experimental
Study of Inequality and Unpredictability in an
Artifical Cultural Market
!!! 14341 Probanden !!! NO
HORST 17.09.2009
27. Empfehlungsmaschinen
funktionieren ... gut genug?
[FlederHosanagar ManagementScience2009]
Blockbuster Culture‘s Next Rise or Fall: The
Impact of Recommender Systems on Sales
Diversity
!!! Mathemat.Simulation !!! ?
HORST 17.09.2009
28. Empfehlungsmaschinen
funktionieren ... gut genug?
[BergerLeMens PNAS2009] How Adoption
Speed affects the Abandonment of Cultural
Tastes
!!! Verwendung statistischer externer Daten !!!
„adopt
quickly=die
fast“
HORST 17.09.2009
29. Empfehlungsmaschinen
funktionieren ... gut genug?
[Celma PH.D THESIS2009] Music
Recommendation and Discovery in the Long Tail
!!! 288 Probanden !!!
!!! 3 verschiedene Empfehlungssysteme!!! CF even
better than
HYBRID
HORST 17.09.2009
32. Was ist damit möglich?
Konzepte Virgin Suicides
Alltogether Miner‘s Son
Sofia Coppola
Beth Hirsch Rhodes
J.B. Dunkel Thomas Mars Lost in Translation
Etienne de Crecy PHOENIX Too Young
AIR
ORANGE Pharell Williams Angel
Eyewater
CASSIUS Feeling for you
Gwen McRae
Philippe Zdar All this love that I‘m giving
HORST 17.09.2009
33. Was ist damit möglich?
Beziehungen
directed
composed contains
composed played married
friends directed
member
lead singer
tour manager contains soundtrack
composed
friends 2006 charts 65
related performed
member composed
worked with
composed performed
member
contains a sample
HORST 17.09.2009
34. Was ist damit möglich?
Jeder Pfad eine Story! Virgin Suicides
Alltogether Miner‘s Son
composed contains directed
Sofia Coppola
composed Beth Hirsch Rhodes married
friends played directed
J.B. Dunkel Thomas Mars Lost in Translation
member lead singer
Etienne de Crecy PHOENIX Too Young contains soundtrack
tour manager composed
AIR friends 2006 charts 65
ORANGE related Pharell Williams Angel
member composed Eyewater
worked with performed
CASSIUS Feeling for you
composed performed
member contains a sample Gwen McRae
Philippe Zdar All this love that I‘m giving
HORST 17.09.2009
35. Jeder Pfad eine Story!
Persönliche Aspekte Virgin Suicides
Alltogether Miner‘s Son
composed contains directed
Sofia Coppola
composed Beth Hirsch Rhodes married
friends played directed
J.B. Dunkel Thomas Mars Lost in Translation
member lead singer
Etienne de Crecy PHOENIX Too Young contains soundtrack
tour manager composed
AIR friends 2006 charts 65
ORANGE related Pharell Williams Angel
member composed Eyewater
worked with performed
CASSIUS Feeling for you
composed performed
member contains a sample Gwen McRae
Philippe Zdar All this love that I‘m giving
HORST 17.09.2009
37. „Fresh Brainfood“
Algorithm Design
[Wilcock 200x] Talking OWLs: Towards an
Ontology Verbalizer
[Galanis et al 2009] An Open-Source Natural
Language Generator for OWL Ontologies and its
Use in Protégé and Second Life
[Górka et al 2007] Information System Based on
Natural Language Generation from Ontology
HORST 17.09.2009