MuSe Client
MuSe Server
Music Repository
UserProfile
Music Retrieval Engine
Wrapper/ Connector 1
User Profile Engine
Social Connector 1
HTML
XML
XML
HTML
Recommender Manager
REST Webservice
EvaluationFramework
Recommendation List Builder
Pluggable Recommender 1
Recommender
Re
com
me
nd
er
Inte
rfac
e
Recommender 1
Web-based User Interface
Recommendation List
AdministrationSpotify
Connector
AJAX
Spotify APIJSON
Data Repositories
Co
ord
ina
tor
User Context
Trac
k M
ana
ger
Use
r M
ana
ger
Charts
Last.fm API
We
b S
erv
ice
s
Freebase
DBpedia
Sem
an
tic
We
b
Last.fm
Soci
al
Ne
two
rks
RDF
RDF
Music platforms such as Spotify, Google Play Music or
Pandora make millions of songs accessible
Social Networks and the Semantic Web describe
users and items in an unprecedented granularity
New opportunities and challenges for
Music Recommenders
Challenges for development
Prototypes programmed from scratch, again and yet again!
Immediate feedback available
Manifold sources of information to combine
Challenges for evaluations
Thorough evaluations are repetitive, yet non-trivial, tasks
Many aspects are inherently subjective, e.g. serendipity
can be only investigated in vivo (with real users)
Legal issues: (1) user data is highly sensitive
(2) content-based approaches require data
Real-time monitoring Results are plotted
and can be exported
New recommender algorithms can be
plugged in comfortably
Users provide profileinformation or social identities
Play and rate recommended songs from Spotify
Conforms with state-of-the-art privacy standards
Variety of baseline algorithms for comparison included
MUSE: A Music Recommendation Management SystemMartin Przyjaciel-Zablocki Thomas Hornung Alexander Schätzle Sven Gauß Io Taxidou Georg Lausen
Department of Computer Science, University of Freiburg, Germany{zablocki, hornungt, schaetzle, gausss, taxidou, lausen} @informatik.uni-freiburg.de
MUSE (Music Sensing in a Social Context)
Music Recommendation Management System
(MRMS) for developing and evaluating
music recommendation algorithms
Takes care of all regular tasks required for
in vivo evaluations
Simple API enables support for arbitrary
music recommenders
Connectors to social networks & semantic
web sources are added continuously
Open sourced under Apache License 2.0:
dbis.informatik.uni-freiburg.de/MuSe
An public instance of MUSE with a small user
community is online:muse.informatik.uni-freiburg.de
Web-based User Interface
Register & interconnect social identities
Listen & rate recommended songs
Manage new recommenders & evaluations
Data Repositories
3 shared data structures (restricted access!)
MUSIC RETRIEVAL ENGINE collects new songs
and retrieves meta information
USER PROFILE ENGINE retrieves information
from linked social profiles
Recommender Manager
Coordinates interaction of recommenders
with users and the data repositories
Forwardes user request & receives
generated recommendations
Composes list of recommendations
Recommendation Model
A Music Recommendation Management System
MUSE simplifies development and evaluation of music
recommendation algorithms
Provides many tools for developing recommenders
Takes care of typical tasks of in vivo evaluations
Open sourced under Apache License 2.0
Future Work
Increase flexibility of evaluations
Further connectors for social networks
New algortihms for music recommendation
Case Study
48 participants
1567 rated song
recommendations
Overall, 73% of all
songs positively
rated
dbis.informatik.uni-freiburg.de/MuSe
Configure & manage evaluations Browse & analyze results
Compose recommendations
Country ChartsInter-country view on songs
Social Tags & Social NeighborhoodSocial Networks and Semantic Web as source of information
City Charts Explore upcoming city trends
Annual Charts Taste in music evol-ves between the age of 14 and 20 [11]
Administrative access gives control and insign
DevelopersUsers
Compose recommendation lists out of diverse algortihms
MUSE
PLAYING
SONGS
USER MANAGEMENT
EXTENSIBILITY
EVALUATIONSPRIVACY & SECURITY
MUSIC RECOMMENDERS
Listen & rate recommended songs
Quality score per recommender type {love 2, like 1, dislike -1}
Let s stop reinventing the wheel and put the funback in developing great new algorithms!
-
Top Related