Clustering with Expectation Maximisation





In December 2002, I was given a Summer Studentship Grant from Monash University faculty of Computer Science and Software Enginneering with Dr. David Albrecht as my supervisor. We looked at the way that various users had travased websites, and how to cluster users with similiar usage patterns. The software below was abstracted away from a particular problem. Given any graph, with each vertex labelled by a unicode character, and a sequence of unicode characters to represent a path which a user took, it can cluster the users any requested number of clusters. It is worth noting however that the Expectation Maximisation algorithm only finds local maxima - and it will almost always produce a different clustering every time it is run. We were requested to write the software in C#, so the Mono environment under Linux was used to write this software. It also works under Windoze. I had begun a C++ port of this, but never had time to finish it. For more information, see the documentation and sources.


Back | Home