Here's a project I did for my information retrieval class. The code isn't quite open source/I haven't been maintaining it since I've jumped ship to Rails. If I get some time, I will patch it up and have it as a turn-key app for App Engine.
Of course, since then, I've had some more insights into better search techniques and algorithms. If I have to do it all over again, I'd have done it with Hadoop and Mr. Job (well, this was recently open sourced from Yelp!). And of course, this time around, I'd use Sphinx and Thinking Sphinx and use Amazon Map Reduce.
Regardless, here's a writeup of what went right and what didn't:
EDIT: I've had some sprinkles of insights into the travel recommendation space. I've always wanted to build a Pandora for travel and recently this startup came out of secrecy on TechCrunch yesterday - http://techcrunch.com/2010/11/16/gogobot/. It's called "gogobot" and I am pleased to say that my insights at the time of this project couldn't be more right. For one thing, the CEO of this startup explains why travel portals are link farms and SEO machineries that make it impossible to find good recommendations. gogobot also take the natural approach that I took with TheBonVoyage was to make it social from the get go. Like TheBonVoyage (my pet project I built while learning Rails 3), gogobot uses good visuals for exploration. In essence, I was happy to see that my instincts were right!
For what it's worth, here's the source code: https://github.com/samratjp/BayesianTravelSearch
