Four Approaches to Music Recommendations: Pandora, Mufin, Lala, and eMusic

ReadWriteWeb gives us some nice examples of the kinds of recommendation systems I wrote about in my previous post. Pandora is content-based, although the features are extracted by humans. The result is high-quality data, but poor scalability. Mufin is a classical example of content-based music recommenders, using a purely algorithmic approach. Lala seems to be old-fashioned word-of-mouth recommendations put on the Internet. eMusic is a hybrid system, but combines social with expert, and social with content-based like Oscar Celma proposes. Apple Genius is most likely a typical collaborative filtering recommender, based on artist (not song or album) similarity.

All hail the information triumvirate!

Wikipedia has come to dominate Google web search results. It often ranks #1 for searches on common topics like Internet and Evolution. Is it true that Wikipedia articles are the very best source of information for all of these topics? Or are we witnessing the effects of a popularity feedback loop, fueled by the principles of least effort, and our tendency to stick with the first and obvious answers? The web link graph is fundamentally a product of socialization, and Google is fundamentally a social search engine. A popularity bias in inherent in all social information systems, leading us all down the same well-trod path. Could it be that, counter to our expectations, the natural dynamic of the web will lead to less diversity in information sources rather than more?