An ensemble of SOM networks for document organization and retrieval
Apostolos Georgakis, Haibo Li, Mihaela L. Gordan · 2006
In this paper a variant of the well-known self-organizing map algorithm is exploited for document organization and retrieval. An ensemble of self-organizing maps are employed in an effort to boost the performance of the standard algorithm. In doing so, the feature space is uniformly sampled and the bootstrapped sets that are created are supplied to each of the constituent members of the ensemble. The proposed variant along with the standard selforganizing map algorithm are used to partition the document repository into clusters of semantically related documents. The quality of clustering offered by the proposed variant is accessed quantitatively via information retrieval experiments using a set of test documents to query the text clusters created by the proposed algorithm. The same experiment is repeated for the standard self-organizing algorithm. An improved performance with respect to the average recall-precision curves is achieved by the proposed variant compared to the standard algorithm.