Previously, I wrote about one of my favorite web 2.0 apps, Pandora.

After I wrote that post, I read through a chapter in my new favorite book Programming Collective Intelligence by Toby Segaran.. The chapter was about searching and ranking using a neural net. It greatly reminded me of Pandora. The user inputs the name of an Artist or a song. Each artist/song are entries in Pandora's Music Genome Project which associates each artist/song with different musical characteristics. Pandora will initially suggest music based on these characteristics. That seems like straight-up database to me...bfd. What's interesting is that once the music starts playing, the user has the option to give each song a "thumbs-up" or "thumbs-down." That's training! For most of the semester, I assumed that Pandora must be using a neural network for this training. Then today, I start reading the chapter in Programming Collective Intelligence on filtering documents. There are other ways to train that are not as computationally expensive. This really leaves me to wonder. Aside from pondering the ways in which Pandora is suggesting tunes, I'm very curious about how one would unit test and system test these types of things.

It's been so easy to stick with Web 2.0 concepts this semester. The information I've found about the Semantic Web has been fairly sparse. From what I can tell, it's all about filtering from an extremely wide net. For example,a few days ago, I found out about Zemanta It's a semantic application that suggests content for blogs. The user types along and while they are writing their blog, Zemanta suggests content for the user. The technologies I've looked at for the Semantic web seem to revolve around markup and data representation.

T