next up previous
Next: Test over a subset Up: Clustering Abstracts of Scientific Previous: Performance Measurement

Experimental Results

Our main concern was to evaluate the term selection methods described above, in the clustering of abstracts task, specifically in a narrow domain. Thus, we have used only one clustering method, $ k$-NN using Jaccard similarity function [18], which is consider as unsupervised, and therefore it complies with our requirements.


David Pinto 2006-05-25