Hacker News Reader

here is my research on detecting communities in Twitter's Social Graph (36 Million users) using Label Propagation algorithm. http://www.akshaybhat.com/LPMR/ You can find both code as well as results.

View on HN

Akshay, great use of LP. Did you consider other functions for adsorption besides a max vote? It would be interesting to see a distribution of possible communities based on seeded community labels from your current results.

Also, thanks for releasing your code. I'm currently working with the Junto package http://code.google.com/p/junto/ but I'll give your map reduce implementation a try for the added benefit of running LP on larger datasets.

The LP algorithm which I have used [ http://pre.aps.org/abstract/PRE/v76/i3/e036106 ] is different from Label Propagation algorithms used for Semi Supervised learning. The algorithms implemented in Junto are Semi Supervised algorithms (generally used with Bi-Partite graphs) while the label propagation algorithm which I have used is used for unsupervised community detection in networks.

If you dont have access to a Hadoop cluster, I have some multiprocessor code, which is slightly faster, if you have enough RAM (32 Gb for twitter dataset). drop me mail if you need it.

Reply on news.ycombinator.com