Dimensionality reduction for graph of words embedding

  • Authors:
  • Jaume Gibert;Ernest Valveny;Horst Bunke

  • Affiliations:
  • Computer Vision Center, Universitat Autònoma de Barcelona, Bellaterra, Spain;Computer Vision Center, Universitat Autònoma de Barcelona, Bellaterra, Spain;Institute for Computer Science and Applied Mathematics, University of Bern, Bern, Switzerland

  • Venue:
  • GbRPR'11 Proceedings of the 8th international conference on Graph-based representations in pattern recognition
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

The Graph of Words Embedding consists in mapping every graph of a given dataset to a feature vector by counting unary and binary relations between node attributes of the graph. While it shows good properties in classification problems, it suffers from high dimensionality and sparsity. These two issues are addressed in this article. Two well-known techniques for dimensionality reduction, kernel principal component analysis (kPCA) and independent component analysis (ICA), are applied to the embedded graphs. We discuss their performance compared to the classification of the original vectors on three different public databases of graphs.