This is my third week listening and documenting last.fm recommendations. I am still facing the same problems of loosing all my recommendations at the end of every week and having to work the whole week to regain my preferences and straighten out my station to suit my preferences. I am tired of reworking the whole system … done it 3 times already!!! :(
I think I can safely conclude that basing recommendations on artists is not a very bright thing to do. In a social setting, where people recommend music to one another, maybe this makes sense, as people tend to talk more about an artist than a particular song. But I would personally like to listen to some good music than patronize my favorite artist by listening to all their music.
Last.fm could definitely improve their user rating system. When I say ‘don’t play a track’, the system should understand that I don’t want to listen to that type of music or artist, especially if I carry out this action multiple times over the same dataset. Last.fm does not seem to learn from its user actions, and the only way to refine ones recommendations is by deleting artists from the recommended artists list.
In my personal opinion, last.fm is all about pushing new music and artists to the end-users play list. This explorative nature of the recommendations can be really frustrating sometimes, as you expect certain consistency in your recommendations. I am not in the mood for new music everyday. Some days I just want to listen to my good-ol favorites and not worry about rating bad music and artist.
Taking a page out of Remi’s Pandora blog, I have contacted last.fm developers on their blog asking them to help me better analyze/understand last.fm recommendations. No replies yet. Wish there was a ‘Lucia’ in the last.fm team too ;)
No comments:
Post a Comment