New classification quality estimators for analysis of documentary information: application to patent analysis and web mapping

摘要

The information analysis process includes a duster analysis or classification step associated to an expert validation of the results. In this paper, we propose new measures for estimating the quality of cluster analysis. These measures derive form the Galois lattice theory and from the Infoimation retrieval (IR) domain. As opposed to classical measures of inertia, they present the main advantages to be both independent of the classification method and of the difference between the intrinsic dimensions of the data and those of the clusters. We present two experiments using the MultiSOM model, which is an extension of the Kohonen's SOM model, as cluster analysis method. Our first experiment on patents data shows how such measures can be used to compare viewpoint-oriented classification method, like MultiSOM, with a global cluster analysis approach, like Kohonen' s SOM. Our second experiment, which takes part in the EISCTES EEC project, highlights that break-even points between our different measures of Recall/Precision can be used in order to determine an optimal number of dusters for Web data classification. The contents of the dusters obtained when using different break-even points are compared in order to study the quality of the resulting maps. This optimisation seems to be mandatory when one want to classify documents issued from the Web, where sparseness is usually a blocking factor.