There is an interesting article by Clive Thompson in this Sunday NY Times Magazine on the progress of the $1 million Netflix challenge to improve their recommender results by 10%. The article does not shy away from the mechanics of search algorithms, including a discussion of how Singular Value Decomposition is useful for summarizing the search space. This level of detail often left out of the popular press when describing algorithms.
(By the way, one of the earliest and most cited papers on SVD is Deerwester, S., Dumais, S. T. , Furnas, G. W., Landauer, T. K., & Harshman, R., Indexing by Latent Semantic Analysis, JASIS, 1990.)
Thompson goes to talk about the “Napoleon Dynamite” problem and Netflix’s internal debate of whether hiring cinephiles to tag all 100,000 movies would help. The article ends with Pattie Maes questioning whether computer search alone with ever be enough. Instead, future recommender systems might need to a mixture of algorithmic and social networking tools.
In all, it the article provides a great discussion of rather complex issues about recommender systems. He makes it clear how progress can measured empirically, how difficult it is to improving algorithms even by a fraction of a percent, and science can advance through an open dialogue among highly competitive research teams.