Data mining on orkut
I've long had an interest in social networking and have been happy to work on orkut at Google. I finally got permission from ACM to post a paper I co-authored about orkut: Evaluating Similarity Measures: A Large-Scale Study in the Orkut Social Network, which describes research I did with fellow Googlers Mehran Sahami and Orkut Buyukkokten. While the paper would only be of interest to computer scientists (especially working in recommender systems, data mining, and collaborative filtering), I also described the highlights in a non-technical manner in Too Much Information, an occasional column for orkut media:
Last spring, orkut began providing related community recommendations, which are displayed on the bottom-right side of community pages. They were generated automatically based on common community membership. For example, since many of the same people belonged to both The Simpsons! and South Park, links to each community were displayed on the other community's page. Some of the associations were amusing. For example, there was a link from C++ (a computer programming language) to What's she trying to say? (a community for men who don't understand women). Another interesting relation was between Chocolate and PMS.