MediaDigits

Mediadigits Logo

Multimedia and web technologies

A short introduction to Last.FM

Le mercredi 2 janvier 2008 par Stéphane Turlier dans les catégories ,

Last.FM is a music web service providing a social approach to the music experience. As many social web services, each user has an account and a profile. Basically, the user can browse the Last.FM catalogue and listen to 30 seconds excerpts. For each artists different kind of information are given such as biography, album covers, concert dates,… For many artists it is not possible to listen to whole tracks, so Last.FM proposes a system of webradios. Those radios are basically made of a streamed playlist generated “on the fly” and delivered to the user. For instance, when a user selects the Tracy Chapman similar artists web radio, he/she can listen for free to a streamed audio content containing music of artists similar to Tracy Chapman. To generate this playlist Last.FM uses a system of similarity. The music similarity is a rather complex problem thus it will be discussed in an later article. To make it simple in a first approach, we can consider that two artists are similar if they belong to the same genre. However Last.FM has a much more refined way of computing artist similarity.

Audio scrobbling: an implicit relevance feedback method

One of the most interesting features of Last.FM is its scrobbling protocol. This technology aims to monitor the user behaviour so as to infer its likes and dislikes. The assumption is the following, if a user listens to a whole track (i.e. more or less till the end), that means that he/she likes it. Moreover, using the GUI, the user can give feedback by rating each song and each artist. The Last.FM player proposes by default a very simple feedback based on three interactions : Love, ban and skip. Based on these observations and on the similarity classification, Last.FM can build a user profile with user preferences. For each artist, Last.FM can provide a web radio and users can skip, ban or love the tracks when they listen to them.

Social filtering and user similarity

Now, Last.FM has built a whole user database with tons of user preferences that have been collected using the scrobbling protocol. The idea of social web is putting together user data, comparing them in order to extrapolate and create information. In matter of music recommendation, what is interesting is to give the user the possibility to discover music he may not have heard before. With the similar artist web radios, Last.FM already gives the possibility to listen to new music, however, the only common factor between the tracks is the similarity to a given artist and no personalisation to the inner particularity of the user profile is used in such a scenario. For each user, LastFM is also able to provide a personalised web radio, only by comparing the user profile with other user profiles. This webradio streams music that the user has already heard and listened to, but it also streams music completely new to the user.

In order not to constrain the users to the Last.FM player, Last.FM has documented its scrobbling protocol. As a result various plugins have been developed in order to be able to scrobble using players such as winamp, windows media player and even itunes. However, it is still possible to write from scratch add-ons to existing media software in order to have them scrobble. Thus, when a user listens to his/her own music, his/her music profil can be transparently and automatically updated provided that his/her library has accurate ID3 tags.

Bibliography

  • G. W. E. Pampalk, “Dynamic playlist generation based on skipping behavior.” 2005.
  • D. L. Chao, J. Balthrop, and S. Forrest, “Adaptive radio: achieving consensus using negative preferences.” 2005.

Comments

---