Agrupamento automático baseado em autoridade e conteúdo

Abstract

This dissertation introduces a technique of clustering analysis that combines concepts of document similarities by contents with link information. The Classic Vector Model is used to carry out the calculation of the similarities between the documents and a link analysis algorithm that is used to get the value of the authority of each document. Calculating the values of the local authorities from the documents belonging to each one of the groups and employing the biggest local authority as the reassign the cluster, we have redistribution of the documents to the clusters. This combination provides clusters represented by the best authority in that subject. This algorithm, called Local Authority Clustering, was proposed, introduced and the quality of its results was evaluated through comparison with the traditional K-means. The AAL has the link structures of the Web as definite from the characteristics that will be used to clustering the documents with several applications in this environment, as the identification of the clusters in a large collection of pages to minimize the search or even to gather together the result of the search generating different clusters of documents.