ICML 9

9º World Congress on Health Information and Libraries

Salvador, Bahia - Brazil, September, 20 to 23 - 2005

BVS4

4th Regional Coordination Meeting of the VHL

September, 19 to 20 - 2005

Finding related articles by a bibliometric approach

Participants:
  • Departamento de Ciências da Informação, Universidade Federal do Espírito Santo  - Brasil
  • Departamento de Ciências da Informação, Universidade Federal do Espírito Santo  - Brasil

Finding related articles by a bibliometric approach

The increasing amount of electronic documents and the diversity of scientific areas are turning the traditional manual work in libraries of clustering documents an impratical job. Hence, an automatic solution is needed in order to keep the same pace as the new documents are generated. The increase of the digital library services, in particular in Brazil, is an example of this growth. Nevertheless, when looking at the provided metadata struture offered by some of the management content software used in this area, we notice that important features are negleted. In this paper we are going to present a factorial analysis on the bibliographical references of a set of articles. By this analysis we are going to show a novel way of creating a web of related subject papers. Therefore, this paper presents a methotology for scientific document clustering based on their citations. This methodology is based on the decomposition of the citation vs. document matrix by the Singular Values Decomposition. By this decomposition we calculate the degree of similarity between any two documents grouping them by their similarities with respect to their citations. We carried out an experiment in order to validate our assumptions. The experiments was performed on a set of 70 articles from an on-line Brazilian journal: Datagramazero. The results showed that this methodology can well be used as an extra tool for a digital library for locating, searching and classifying documents, as we could create a graph of the most related papers of this journal only by looking at their citations frequencies.