Marc
78 Beiträge
schrieb am 03.03.13 um 06:26 Uhr
Link zu diesem Post
Abstract:
In the Internet music scene, where recommendation technology is key for navigating huge collections, large market players enjoy a considerable advantage. Accessing a wider pool of user feedback leads to an increasingly more accurate analysis of user tastes, effectively creating a "rich get richer" effect. This work aims at significantly lowering the entry barrier for creating music recommenders, through a paradigm coupling a public data source and a new collaborative filtering (CF) model. We claim that Internet radio stations form a readily available resource of abundant fresh human signals on music through their playlists, which are essentially cohesive sets of related tracks. In a way, our models rely on the knowledge of a diverse group of experts in lieu of the commonly used wisdom of crowds. Over several weeks, we aggregated publicly available playlists of thousands of Internet radio stations, resulting in a dataset encompassing millions of plays, and hundreds of thousands of tracks and artists. This provides the large scale ground data necessary to mitigate the cold start problem of new items at both mature and emerging services.
Furthermore, we developed a new probabilistic CF model, tailored to the Internet radio resource. The success of the model was empirically validated on the collected dataset. Moreover, we tested the model at a cross-source transfer learning manner – the same model trained on the Internet radio data was used to predict behavior of Yahoo! Music users. This demonstrates the ability to tap the Internet radio signals in other music recommendation setups. Based on encouraging empirical results, our hope is that the proposed paradigm will make quality music recommendation accessible to all interested parties in the community.

Source:
http://dl.acm.org/citation.cfm?id=2187838

Very interesting article, unfortunately the abstract only is available. This has gotten me started, how cool and how possible would be to use the power of internet radio to create music recommendations for your own music library?
Some thoughts on the matter:
-> If a log is kept of all played tracks by any given radio, relationships can be established given whioch track was played before and after which track. Let's say that after two days a radio played track A 4 times before track B and 2 times before track D and C.
What are your thoughts?
 
Kaefer
179 Beiträge
schrieb am 03.03.13 um 09:54 Uhr
Link zu diesem Post
Hi!
I think most internet radio stations use random playback, so I don't think you will find a special statistical relationship between 2 successively played songs. It is more about the whole playlist which represents a specific genre.
Do you plan to develop a software program?
Viele Grüße,
Käfer
 
Marc
78 Beiträge
schrieb am 03.03.13 um 16:05 Uhr
Link zu diesem Post
Ahh you are right. Thanks Kaefer for making the point.
I'm not a developer, so the aim here was/is to discuss whats presented in the article; how good
data gathered from internet radios would make for music recommendatios.
 
Marc
78 Beiträge
schrieb am 03.03.13 um 16:05 Uhr zuletzt bearbeitet von Marc am 03.03.13 um 16:10 Uhr
Link zu diesem Post
(Duplicated) - I'd rather have this one deleted… :)
 
alex
2470 Beiträge
schrieb am 05.03.13 um 21:18 Uhr
Link zu diesem Post
I think it will be possible because much data about streams is collected. But it's hard to get it working without "false positives" I think. Anyway, it would be a nice experiment:-)
LG/Best regards, Alex

"Journalism is printing what someone else does not want printed. Everything else is public relations."
- George Orwell

D1734FA178BF7D5AE50CB1AD54442494
 
alex
2470 Beiträge
schrieb am 07.03.13 um 20:55 Uhr
Link zu diesem Post
I bought the article, read it and only understand it partially. I think I have to study mathematics6-)
LG/Best regards, Alex

"Journalism is printing what someone else does not want printed. Everything else is public relations."
- George Orwell

D1734FA178BF7D5AE50CB1AD54442494
 
Marc
78 Beiträge
schrieb am 08.03.13 um 08:20 Uhr
Link zu diesem Post
Ahh same here, I find it very interesting but I will need some assitance to understand the model.6-)
I don't know but, would the publishers be kind enough to share the code used for the model analysis?
 
Marc
78 Beiträge
schrieb am 25.03.13 um 04:55 Uhr
Link zu diesem Post
I've done some progress with the article but I'm half way still. Perhaps a stupid question, but in order to run an experiment and see what happens is it necessary to fully understand the model?
Or for example the source code could be simply used to work with information stored from radio stations?

I don't have Linkedin but if* is a good idea I may create an account and contact one of the developers/authors of the publication.
Profile of Natalie Aizenberg
http://il.linkedin.com/in/natalieaizenberg
Profile of Yehuda Koren
http://il.linkedin.com/pub/yehuda-koren/7/614/856
Profile of Oren Somekh
http://il.linkedin.com/pub/oren-somekh/28/257/b8
 
alex
2470 Beiträge
schrieb am 03.04.13 um 23:10 Uhr
Link zu diesem Post
If you get information I would take a look into that in the future. I'm sorry but I cannot tell if you need the full model or not. AFAIR the document describes simple models and describes the "full" model at the end… I really cannot help out here, the only thing I could do is to give you collected data (SQL dump).
LG/Best regards, Alex

"Journalism is printing what someone else does not want printed. Everything else is public relations."
- George Orwell

D1734FA178BF7D5AE50CB1AD54442494