by Ruth Suehle
Today was the first official day of JavaOne. I visited a couple of non-JBoss sessions that sounded really interesting, and they were–so now I share them with you. The first is about an improved web recommendation system, and the second is for improving collaboration with your off-site coworkers.
But first, if you’re here with us…
Come visit us at the JBoss booth in the Pavilion. We’re straight back and on the right when you come through the door. Every day, we’re holding 15-minute mini-sessions in the booth on the hour. You can meet the core developers and ask your questions in person. We’ve also got JBoss t-shirts and free entitlements of JBoss Developer Studio.
Also, the JBoss technical sessions are all still ahead of us. So if you’re here at JavaOne, be sure to check them out. If you’re not, keep reading this week to hear more about what’s going on.
Recommendation systems, or, what’s that about my nose hair?
The first session I heard this morning was Stephen Green and Paul Lamere from Sun talking about an improved recommendation system for websites like Amazon or iTunes or any that uses recommendations. They had some impressive statistics about the power of recommendations to drive sales and downloads. NetFlix is even offering a million-dollar prize to the first team that can improve their recommendation accuracy by 10%, which, it turns out, is a tougher task than you might think.
They had quite a few amusing examples of bad recommendations they got from various sites. If you own “Baby One More Time,” you might like Report on Pre-War Intelligence. Because you bought The Chronicles of Narnia, you might like this nose and ear hair groomer.
Project Aura is based on collaborative filtering, which means grouping similar items based on a halo of users and terms. It creates a text aura from content analysis of things like reviews, blogs, and tags. But those things only work when that content and those users exist for comparison and analysis. So Aura solves the “cold start” problem for music through feature analysis.
It extracts information from the song file about things like rhythm and harmonic content. That data is then fed through a trained system that tags that unknown song with all the relevant categories, based on what it has learned from all the previous input. In the example they showed, an analysis of Queen’s “Bohemian Rhapsody” resulted in tags like “classic rock,” “happy,” “England,” “70s,” and “favorite artist.” In the end, general tags lead to more specific tags, which lead to recommendations.
Project Aura uses openid for users and Attention Profile Markup Language (APML) to attach a concept a user likes, such as an artist name, along with a degree to which the user likes it, to that user’s profile for their recommendations. It’s not available for use yet, but they did suggest checking out taste.sourceforge.net for a look at open source collaborative filtering.
Project Wonderland, or as I’m thinking of it, Second Work
Sun noticed that on any given day, half of their employees are out of the office. And as anyone who’s experienced it knows, the personal interaction you lose when you’re off site, especially long term, can be really important. Conference calls just aren’t the same as being in the meeting yourself.
Enter Project Wonderland. It’s an open source, Java-based, 3D, API-based graphics engine that manages the world, animations, and avatars for all the sorts of employee interactions that you’d have in real life. Think of it as going to work in Second Life. It’s built on top of Project Darkstar, which was created for MMORPGs, and is extensible with customizable worlds.
The real bonus over other options (like just having a meeting on IRC) is that it allows employees to interact in a more real-world fashion. You don’t have to train people how to use a system, how to open a private chat, etc., because the interactions are natural. If you’re in a group and want to have a private conversation, you simply walk away from the group, just like you would if you were all standing together in real life.
Project Wonderland is approaching version 0.5. They’re aiming for a 1.0 in 2009.
Personal recommendation of the day
For those of you here braving the crowds, relentless badge checkers, and $7 hot dogs, I found a great secret today. The one thing I’ve always hated about conferences at the Moscone Center is the lack of good, quick restaurants nearby. Then this morning, somebody told me about Organic Coffee. Go straight upstairs to the street, turn right out the doors (toward the carousel), then right at the corner. It’s a block or two up the street on the left, near the Metreon. They have great sandwiches and a tasty-looking hunk of quiche I intend to try tomorrow. Ever better, it’s inexpensive, and there was no line, even at lunchtime.
JBoss session schedule