1. Music discovery on the net
Barcamp3, Berlin
Petar Djekic
October 18th, 2008
2. From phonograph to widgets
Widget
Mobile
Web Web
PC PC PC
Portable Portable Portable Portable
TV TV TV TV TV
Car Audio Car Audio Car Audio Car Audio Car Audio
Radio Radio Radio Radio Radio Radio
Phono Phono Phono HiFi system HiFi system HiFi system HiFi system
1890 1920 1930 1950 1980 1990 2000
Source: own, Wikipedia ’08
3. Yet still..
„iPod classic can „There is are an average of
hold up to 30,000 700 songs stored on a U.S.
songs“ music downloader’s
player.“
„Average MP3 player only
57% full“
Source: Apple 2008, Forrester Research 2008, IPSOS 2006
4. Music Discovery 4
“The only bad thing about “A wealth of information
MySpace is that there creates a poverty of
are 100,000 bands and no attention”
filtering.
I try to find the bands I Herbert A. Simon, Nobel prize
winning economist
might like but often I just
get tired of looking.”
15 year old student, IFPI focus
group research, July 2007
6. Recommendation technologies: Overview
Human behaviour: Recommendations are based on
behaviour, e.g., Collaborative filtering using listening or
purchase habits
Human annotation: Recommendations are based on
annotations and expertise, e.g., ratings, tags, classification
into genres, editorial content
Content analysis: Recommendations are based on
characteristics of the content itself, e.g., sound density,
vocals, tempo, sound color, instruments, volume, dynamics
11. Recommendation technologies: Issues
Relevance: How good does Variety: Variety of
the content suit my taste? recommendations (Beatles-
How about mood and problem); connection
expectations? between variety and content
available
Scalability: Indexing of
existing content libraries Privacy: Who owns YOUR
and new releases (cold- data?
starts)
Explanation: Why was
Objectivity: Manipulation of something recommended?
rankings, consistency of
Portability: How about
recommendations
mobile devices, MP3
players
12. Mash it up now! <resources>
Human annotation/behaviour
MusicBrainz: similar artists, tags, meta data, CC / PD
license
Yahoo! Music: similarities, charts, ratings, meta data,
REST webservice, max. 5000 queries/day
Last.fm: similarities, tags, ratings, meta data, REST
webservice, free for non-commercial use
Content analysis
Echo.nest: sound analysis, recommendations, custom
HTTP webservice,
audiobaba: similarities, custom HTTP webservice, max. 1
query/sec
13. Mash it up now! <resources>
Matching Full-track
Identifier: MusicBrainz, Youtube
ISRC, All music guide
Imeem Media
Meta data: G’n’R, Platform, yahoo
GunsNRoses, Guns N’
Seeqpod, skreemr
Roses…
Acoustic fingerprints: Radio stream
Standards?
14. Recommendations, again
Books
David Jennings (2006) Net, Blogs, and Rock‘n‘Roll
David Huron (2008) Sweet Anticipation: Music and the Psychology of Expectation
Papers
Kim, J., and Belkin, N. J. (2002). Categories of music description and search terms and phrases used by non-music experts, http://
ismir2002.ismir.net/proceedings/02-FP07-2.pdf
Tintarev, N. (2007), A Survey of Explanations in Recommender Systems, http://www.csd.abdn.ac.uk/~ntintare/
TintarevMasthoffICDE07.pdf
Mobasher, B. et al (2007) Trustworthy Recommender Systems: An Analysis of Attack Models and Algorithm
Robustness, http://maya.cs.depaul.edu/~mobasher/papers/mbbw-acmtoit-07.pdf
Conferences
The International Conferences on Music Information Retrieval and Related Activities, ISMIR, http://www.ismir.net/
ACM Recommender Systems, RecSys, http://recsys.acm.org
Blogs
Duke Listens!, http://blogs.sun.com/plamere/