Dec 27, 2011: Junto has now moved to Github (https://github.com/parthatalukdar/junto)

This toolkit consists of implementations of various graph-based semi-supervised learning (SSL) algorithms. Currently, three algorithms are implemented: Gaussian Random Fields (GRF), Adsorption, and Modified Adsorption (MAD).

Starting with v1.2.0, Junto also contains Hadoop-based implementations of these three algorithms.

Why is the toolkit named Junto? The core code was written while Partha Talukdar was at the University of Pennsylvania, and Ben Franklin (the founder of the University) established a club called Junto that provided a structured forum for him and his friends to debate and exchange knowledge:

http://en.wikipedia.org/wiki/Junto_(club)

This has a nice parallel with how label propagation works: nodes are connected and influence each other based on their connections. Also "junto" means "along" and "together" in a number of Latin languages, and carries the connotation of cooperation---also a good fit for label propagation.