Clustering of biomedical scientific papers

In this paper we present a methodology for document clustering based on non-negative matrix factorization (NMF) and ensemble clustering. Thanks to the ensemble clustering the algorithm is less prone to get into a local minimum caused by the initialization of the NMF. Despite the ensemble clustering, the algorithm keeps the semantic interpretability of the NMF and constructs a coocurrence matrix that allows the projection of the documents onto a two-dimensional space suitable for visualization. The algorithm is freely available for the information retrieval community from the bioengineering laboratory Web page.