Text Mining With Hybrid Clustering Schemes

Hybrid information retrieval (IR) schemes combine dierent normalization techniques and similarity functions. Hybrid schemes provide an eÆcient tech-nique to improve precision and recall (see e.g., [4]). This paper reports a hybrid clustering scheme that applies a singular value decomposition (SVD) based algo-rithm followed by a k{means type clustering algorithm. The output of the rst algorithm becomes the input of the next one. The second algorithm generates the nal partition of the data set. We report results of numerical experiments performed with three k{means type clustering algorithms. Those are: the classi-cal k{means (see e.g., [9]), the spherical k{means (see [7]), and the information{ theoretical clustering algorithm introduced recently by [8], and [1]. A comparison with the results reported by [7] is provided. Clustering is an important, old and very diÆcult problem. While a wide variety of clustering algorithms are already available in the literature in many cases a singl