23. WHY LIKE THIS?
• how does an expert select his music?
Saturday 26 February 2011
24. WHY LIKE THIS?
• how does an expert select his music?
• how does a bartender select his music?
Saturday 26 February 2011
25. WHY LIKE THIS?
• how does an expert select his music?
• how does a bartender select his music?
• making the expert data accessible in a usable way for
a bartender
Saturday 26 February 2011
26. WHY LIKE THIS?
• how does an expert select his music?
• how does a bartender select his music?
• making the expert data accessible in a usable way for
a bartender
A musical context is a musical description for situations based on
atmospheres and musical properties.
Saturday 26 February 2011
27. EXPERTS @ WORK
songs param. value matches
Saturday 26 February 2011
28. EXPERTS @ WORK
songs param. value matches
Saturday 26 February 2011
29. EXPERTS @ WORK
songs param. value matches
Saturday 26 February 2011
30. EXPERTS @ WORK
songs param. value matches
Saturday 26 February 2011
31. EXPERTS @ WORK
songs param. value matches
Saturday 26 February 2011
41. context
subcontext A
Saturday 26 February 2011
42. context
subcontext A
songs with
subgenre(easy listening
OR pop café)
Saturday 26 February 2011
43. context
subcontext A
songs with
subgenre(easy listening
OR pop café)
+
songs with
mood(relax OR tasteful
OR stylish)
Saturday 26 February 2011
44. context
subcontext A
songs with
subgenre(easy listening
OR pop café)
+
songs with
mood(relax OR tasteful
OR stylish)
+
songs with
popularity(5 UNTIL 7)
Saturday 26 February 2011
45. context
subcontext A subcontext B
songs with
subgenre(easy listening
OR pop café)
+
songs with
mood(relax OR tasteful
OR stylish)
+
+
songs with
popularity(5 UNTIL 7)
Saturday 26 February 2011
46. context
subcontext A subcontext B
songs with songs with
subgenre(easy listening genre(pop)
OR pop café)
+
songs with
mood(relax OR tasteful
OR stylish)
+
+
songs with
popularity(5 UNTIL 7)
Saturday 26 February 2011
47. context
subcontext A subcontext B
songs with songs with
subgenre(easy listening genre(pop)
OR pop café)
+
+
songs with
songs with
mood(intimate OR
mood(relax OR tasteful
OR stylish)
+ romantic
OR sensual)
+
songs with
popularity(5 UNTIL 7)
Saturday 26 February 2011
48. context
subcontext A subcontext B
songs with songs with
subgenre(easy listening genre(pop)
OR pop café)
+
+
songs with
songs with
mood(intimate OR
mood(relax OR tasteful
OR stylish)
+ romantic
OR sensual)
+
+
songs with songs with
popularity(5 UNTIL 7) popularity(6 UNTIL 7)
Saturday 26 February 2011
49. context
subcontext A subcontext B
songs with songs with
subgenre(easy listening genre(pop)
OR pop café)
+
75% 25%
+
songs with
songs with
mood(intimate OR
mood(relax OR tasteful
OR stylish)
+ romantic
OR sensual)
+
+
songs with songs with
popularity(5 UNTIL 7) popularity(6 UNTIL 7)
Saturday 26 February 2011
58. WHAT’S NEXT?...
• manually annotated metadata
•5 music experts at Aristo Music and
different consultants
• about 200,000 songs
• but, not enough...
Saturday 26 February 2011
59. THE LONG TAIL
http://en.wikipedia.org/wiki/Long_Tail
Saturday 26 February 2011
60. THE LONG TAIL
http://en.wikipedia.org/wiki/Long_Tail
Saturday 26 February 2011
61. THE LONG TAIL
http://en.wikipedia.org/wiki/Long_Tail
Saturday 26 February 2011
62. OTHER PROBLEMS...
• satisfyingthe music choice of all
customers
• retail and catering differ from you
and me!
• new markets
• react fast on emerging music trends
• adding the full Belgian library catalog
Saturday 26 February 2011
63. GENERATE THE METADATA
• from different sources:
• the audio signal
• web sources
• the Aristo database
• attention metadata
• using our metadata generation framework: SamgI
Saturday 26 February 2011
80. HOW TO EVALUATE THIS?
• run an experiment
• evaluate on 1995 artists and 9 genres.
• different search engines: Google,Yahoo! and Live! Search.
• over time: 8 times over a period of 36 days.
Saturday 26 February 2011
81. THE DATA SET
Blues Country Electronic
Folk Jazz Metal
Rap Reggae RnB
Saturday 26 February 2011
82. THE DATA SET
Blues Country Electronic
Folk Jazz Metal
Rap Reggae RnB
10% 9%
3%
2% 12%
13% 5%
4%
41%
Saturday 26 February 2011
83. THE DATA SET
Blues Country Electronic
Folk Jazz Metal
Rap Reggae RnB
Saturday 26 February 2011
88. MORE FINE-GRAINED...
• 18 artists
• more search engines: Google.co.uk/.fr/.be, uk/
fr.search.yahoo.com
• twice a day for 53 days
• 250,000 queries!
Saturday 26 February 2011
89. 2 Pac Rap
Alan Lomax Folk
Art Pepper Jazz
Cradle of Filth Metal
David Parsons Electronic
Desmond Dekker Reggae
Downpour Metal
IceT Rap
Jerry Butler RnB
Joy Lynn White Country
Louisiana Red Blues
Lou Rawls RnB
LTJ Bukem Electronic
Peter Tosh Reggae
Pinetop Smith Jazz
Robert Johnson Blues
Roy Rogers Country
Steeleye Span Folk
Saturday 26 February 2011
94. WHAT TO USE?
• use Google when it’s stable else rely on Yahoo!
• when is it stable? test with a small set
• some artists get classified incorrectly on bad days
• compare the accuracy achieved with the test set to the
average.
Saturday 26 February 2011
95. DEMO METADATA
GENERATION
• http://ariadne.cs.kuleuven.be/samgi-service/
Saturday 26 February 2011
98. MASHUPS
• mashups in music: re-mixing multiple existing songs to
create a new one.
Saturday 26 February 2011
99. MASHUPS
• mashups in music: re-mixing multiple existing songs to
create a new one.
•a mashup is an application that combines data from multiple
online sources to create a new result which was not the
original intend of the data.
Saturday 26 February 2011
100. MASHUPS
• mashups in music: re-mixing multiple existing songs to
create a new one.
•a mashup is an application that combines data from multiple
online sources to create a new result which was not the
original intend of the data.
• data is key!
• tweaking and enriching data is important
• interesting data makes an interesting mashup
Saturday 26 February 2011
101. MY MUSIC MASHUP.
• http://www.cs.kuleuven.be/~sten/lastonamfm/
Saturday 26 February 2011
102. COUNTRY & CONTINENT
• why is it useful?
• subgenres
• popularity
• recommendations
• expensive to annotate
• very few existing research
• very hard with signal processing
Saturday 26 February 2011
103. COUNTRY & CONTINENT
• why is it useful?
• subgenres
• popularity
• recommendations
• expensive to annotate
• very few existing research
• very hard with signal processing
Saturday 26 February 2011
104. COUNTRY & CONTINENT
• why is it useful?
• subgenres
• popularity
• recommendations
• expensive to annotate
• very few existing research
!
• very hard with signal processing
Saturday 26 February 2011
105. IN THE BACKGROUND...
google maps
freebase
last.fm google app
engine
twitter last.fm
last.fm yahoo! pipes last on am/fm
website
dapper youtube
Saturday 26 February 2011
113. FREEBASE MQL
{
"type" : "/music/album",
"name" : "Synchronicity",
"artist" : "The Police",
"track" : [
{"name":"Synchronicity II", "length":305.066},
{"name":"Every Breath You Take", "length":254.066},
{"name":"King of Pain", "length":299.066},
{"name":"Wrapped Around Your Finger", "length":313.733},
{"name":"Tea in the Sahara", "length":255.44},
{"name":"Walking in Your Footsteps", "length":216.773},
{"name":"Miss Gradenko", "length":120},
{"name":"Murder by Numbers", "length":276.8},
{"name":"O My God", "length":242.226},
{"name":"Synchronicity I", "length":202.866},
{"name":"Mother", "length":185.64}
]
}
Saturday 26 February 2011
114. BIOGRAPHIES
• from Last.fm (most of them are
coming from Wikipedia)
• using demonyms.
• “Anders Trentemøller is a
Danish electronic musician.”
• List of demonyms from Wikipedia
http://en.wikipedia.org/wiki/
List_of_adjectival_and_demonymic_forms_of_place_names
Saturday 26 February 2011
116. The Beatles
computer
Saturday 26 February 2011
117. The Beatles
Michael Jackson
computer
Saturday 26 February 2011
118. ABBA
The Beatles
Michael Jackson
computer
Saturday 26 February 2011
119. ABBA
The Beatles
1
Michael Jackson
computer
Saturday 26 February 2011
120. ABBA
The Beatles
1
Michael Jackson
2
computer
Saturday 26 February 2011
121. ABBA
The Beatles
3
1
Michael Jackson
2
computer
Saturday 26 February 2011
122. RESULTS
• with country we also know continent
• Last.fm and Freebase only contains data for small part
• accuracy:
• Last.fm: 91%
• Freebase: 90%
• Bio: 66%
• Combination: 77%
• more advanced possibilities
Saturday 26 February 2011
123. THANK YOU!
QUESTIONS?
slides will appear on http://www.slideshare.net/stengovaerts
Saturday 26 February 2011