A Brief Survey of Last.fm for Music Information Retrieval


 According to Wikipedia (2017a):

  • Originally an Internet Radio service by Felix Miller, Martin Stiksel, Michael Breidenbruecker, Thomas Willomitzer in 2002
  • Audioscrobbler by Richard Jones in UK, merged with last.fm in 2005
  • Acquired by CBS in 2007
  • Felix, Martin and Richard leave in 2009


What is Last.fm?

  • Music Recommender System
  • Two broad types of Recommender Systems:
    • Collaborative systems use info from similar users to make recommendations, as opposed to,
    • Content-based recommender systems like Netflix, which need some kind of content priming (Leskovec, Rajaraman, and Ullman 2014)
  • Social tagging in Last.fm
    • Collaborative filtering: “users annotate items that are relevant for them, so the tags they provide can be assumed to describe their interests, tastes and needs… The more users annotate a certain item with a particular tag, the better that tag describes the item contents.” (Cantador, Bellogín, and Vallet 2010)
  • By those definitions, the last.fm system would be considered a hybrid, since scrobbling may be regarded as adding content-based information. However, Wikipedia (2017a) classifies it is as using a collaborative filtering system


Features of Last.fm

  • Tracks various kinds of user data and listening habits
    • Country, age, favourites (Vigliensoni and Fujinaga 2016)
  • Audioscrobbler API: Integrates with various media players




“Scrobbling is a way to send information about the music a user is listening to”

– Last.fm (2017c)

  • It can only be performed when a track is at least 30s in duration, and has been played through for at least half its total duration
  • Scrobbles can be cached and resent, up to as many as 50 scrobbles in one request
    • Sometimes a scrobble might fail due to bad authentication, or other reasons
    • See Flowchart
  • Used to stream data, but now it’s linked in to youtube radio/Spotify (Wikipedia 2017)


What does such a system offer?

  • Folksonomy (Lamere 2008): Social tagging brings contextual information that computers can’t normally comprehend
  • Genre
    • Post activation analysis paralysis: Being unable to categorize a track for fear of wrongly categorizing it
    • Solution: Apply all the applicable tags
  • Mood (Laurier, Grivolla, and Herrera 2008)
    • Happy, sad, angry, relaxed can now be interpreted through the folksonomy


The Last.fm API

As per the guidelines on Last.fm (2017a):

  • Developer accounts: API Key


Application name

Research App
API key 17xxxxxxxxxxxxxxxxxxxxxxxxxxxx2
Shared secret cxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx6
Registered to adityatb



  • REST API (Representational State Transfer) (Last.fm 2017a, Wikipedia 2017b):
    • Supports HTTP Requests – GET, POST, PUT, DELETE


  • Corrects misspelled metadata information (Last.fm 2017c)
  • Directory of well documented methods (See Last.fm API page)
  • Brief Survey of Terms of Use (Last.fm 2017b):
    • Data is Any information procured through the API: images, text, etc.
    • Non-Commercial Purposes: They reserve “… the right to share in revenue generated… to be negotiated in good faith between Last.fm and You.”
    • No sub-licensing the data
    • Always credit Last.fm when using their data.


Some examples of how the API has been used

  • Comparing Social Tagging systems: Delicious vs. Last.fm (Cantador, Bellogín, and Vallet 2010):
    • Identified popular tags across different genres
    • Find artists that fall under those tags
    • Fans of those artists, and friends of those fans
  • Context-awareness in Recommender systems (Vigliensoni and Fujinaga 2016)
    • Demographics, age, gender can show patterns
    • Long term data to be able to observe usage patterns
      • Last.fm provides full listening histories + demographics
      • Integrated with about “600 media players” (Scrobbler API)
    • Data is used to define an extra parameter called ‘exploratoryness’ that enhances the performance of the system by 12%


Challenges/ Related Issues

Lamere (2008) discusses some of the known challenges when using systems like Last.fm:

  • Cold Start
  • Unpopular/New artists aren’t tagged as often
  • Dishonest tagging – Unpopular artist might apply popular tags to his work
  • Synonymy
    • The tag ‘love’ can be used for a romantic song and also a song that the user ‘loves’
  • Noise
    • “Some examples from Last.fm are Asdf, Random, Lazy eye, d and Sh-t that my sister listened to on my pc grrrrr’’.
  • Some such challenges have been circumvented and discussed by Lamere (2008), but are worth anticipating when using such an API for MIR research.


Additional Finds

 Graham Jenson’s GIT page

  • A compilation of Recommender systems and APIs.
  • Cited: Article surveying recommender systems since conceptualization in 1970s (Martin et al. 2011)



Cantador, Iván, Alejandro Bellogín, and David Vallet. 2010. “Content-Based Recommendation in Social Tagging Systems.” Proceedings of the fourth ACM conference on Recommender systems, Barcelona, Spain.

Lamere, Paul. 2008. “Social Tagging and Music Information Retrieval.”  Journal of New Music Research 37 (2):101-114. doi: 10.1080/09298210802479284.

Last.fm. 2017a. “API – Last.fm.” Last Modified 26 November 2013, accessed 15 January 2017. http://www.last.fm/api.

Last.fm. 2017b. “API Terms of Service – Last.fm.” accessed 18th January 2017. http://www.last.fm/api/tos.

Last.fm. 2017c. “Scrobbling 2.0 Documentation – Last.fm.” Last Modified 26 November 2013, accessed 15 January 2017. http://www.last.fm/api/scrobbling.

Last.fm. 2017d. “Track My Music – Last.fm.” accessed 17 January 2017. http://www.last.fm/about/trackmymusic.

Laurier, C., J. Grivolla, and P. Herrera. 2008. “Multimodal Music Mood Classification Using Audio and Lyrics.” 2008 Seventh International Conference on Machine Learning and Applications, 11-13 Dec. 2008.

Leskovec, Jure, Anand Rajaraman, and Jeffrey David Ullman. 2014. Mining of Massive Datasets. 2nd Edition ed: Cambridge University Press.

Martin, Francisco J, Justin Donaldson, Adam Ashenfelter, Marc Torrens, and Rick Hangartner. 2011. “The Big Promise of Recommender Systems.”  AI Magazine 32 (3):19-27.

Vigliensoni, Gabriel, and Ichiro Fujinaga. 2016. “Automatic Music Recommendation Systems: Do Demographic, Profiling, And Contextual Features Improve Their Performance?” 17th International Society for Music Information Retrieval Conference (ISMIR), New York City, USA.

Wikipedia. 2017a. “Last.fm – Wikipedia.” Last Modified 6 January 2017, accessed 14 Jan 2017. https://en.wikipedia.org/wiki/Last.fm.

Wikipedia. 2017b. “Representational State Transfer – Wikipedia.” accessed 14 January 2017. https://en.wikipedia.org/wiki/Representational_state_transfer.


(Visited 196 times, 1 visits today)