The End !!!

This blog was a term-long project wherein I was to interact with a recommender system and document my analysis of its working. I chose last.fm, and over the past few months, I have tried my best to analyze every aspect of this system.

Before I present a summary of my findings, I would like to thanks Dr. W for providing us with this opportunity to work with and learn from these recommender systems. A special thanks to Remi and her analysis of Pandora, which has on many occasions, proven to be the guiding light in directing me in analyzing last.fm. And to Lucia, a member of the Pandora team whose comments have provided wonderful insight into the world of recommendation systems, especially in the domain of music.

My analysis of last.fm began on an exploratory note. Unlike Pandora which I use on a daily basis, last.fm was totally new to me. I spent the first few weeks trying to get a grip on the countless number of features last.fm has.

So what is last.fm? It’s a music recommendation system, which uses a collaborative approach to recommending artists. The music of these artists is recommended on the user-profile radio station. Last.fm learns of your preferences based on the music you listen to on a daily basis (collecting info from you music players) and on the friends you make on last.fm. Eventually, the system learns your preferences, and correlates that to the hit-charts to find the most popular music that match the user’s preferences.

Now I summarize my analysis by listing my finding for 3 different aspects of the system:

1. User Interface: the first thing that you see and that makes you like or hate a system is the user interface. Last.fm does a poor job in this department. Too many options and too many screens showing the same data, a first time user will tear his hair out trying to work his way through the maze of options to get to his recommendations.
The premise of this system is to provide an effective social networking tool that brings people of like minded musical preferences together. But the system’s bread and butter is its music recommending system, Audioscrobbler. If you hide your main functionality behind a veil of other features, sooner or later, people will move away to less complicated systems.
I guess in this department, Pandora wins hands down as having the best GUI for a music recommender system.

2. Recommendations: based on all my interactions, I have come to the conclusion that last.fm gives more preference to the popularity of music, than to its correlation to a user’s tastes. Popularity of an artist dictates if it recommended to a particular user or not, even though that artist might be less correlated to the user’s tastes when compared to some other non-popular artists.
This means good news for those who are in tune with the current trends in music, and who dig the latest developments in field of music. But what if I am looking for my favorite artists of tracks, which are not in the top 100 list? This question was partly responsible for my frustration with last.fm, about which I have blogged numerous times.
Another aspect of recommendations is the number of times it changes. If I like an artist, I should have a say in how long that artist is allowed to stay in my recommendation list. With popularity deciding the recommendation content, no matter what ratings the user gives to an artist, his appearance on the recommendation list will be solely dictated by how popular he is. This means the recommendation list fluctuates very frequently, making it very frustrating to see new music in your profile every time you login!!!!

3. Scrobbling: Scrobbling is the process by which audioscrobbler can listen in on the music you are playing on your local mp3 player on your desktop, or your iPod. For some reason, scrobbling never worked for me. No matter what kind of music I would listen to on my scobble-enabled winamp, it would have no effect on my recommendation list. I had to resort to listening music on the last.fm web player.


Last.fm has an underlying theme of social networking, which might explain the ton of features available on their webpage. But sorting their UI would help attract more music lovers to their site, who are just looking to listen to some good personalized music!!! Last.fm pushes a lot of new artists to the forefront, which helps users to explore a lot of new music. But this same explorative approach can sometimes be very irritating, having to listen to music you REALLLYYY don’t like.

I have played with last.fm long enough to know this – if u spend enough time on last.fm, you can get it to understand your tastes. Even though you might get new recommendations every week, they will at least align with your tastes.

If given a choice, I would go for Pandora any day because of its ease of use. But last.fm gives an innovative and fresh perspective to finding new music on the world wide web. So if you are looking to explore new musical avenues, and want to learn more from the popular tastes of today, and don’t mind a few rotten recommendations in your path to listening bliss, last.fm is the place for you.

I would like to end this post and this blog with the following: given enough time, any system can recommend the right kind of music. The system that does this in the shortest period of time with the least amount of effort from the user side, wins the race to become the ultimate music recommender. And in my books, Pandora sits at this sweet spot, and looks like last.fm has a long way to go before it can beat Pandora at its game.

On that note, I sign off. Hope you all find that perfect recommender system. Happy Searching and Happy listening !!!

Links:
1. my last.fm profile
2. excel sheet documenting my listening sessions
3. Remi's Pandora Blog !!!!