Jaume Gibert2011
Proceedings
2011
10th International Conference on Multiple Classifier Systems
6713
36-45
Jaume Gibert
Ernest Valveny
Oriol Ramos Terrades
Horst Bunke
MCS
Carlo Sansone
Josef Kittler
Fabio Roli
Multiple Classifiers for Graph of Words Embedding
LNCS
During the last years, there has been an increasing interest in applying the multiple classifier framework to the domain of structural pattern recognition. Constructing base classifiers when the input patterns are graph based representations is not an easy problem. In this work, we make use of the graph embedding methodology in order to construct different feature vector representations for graphs. The graph of words embedding assigns a feature vector to every graph by counting unary and binary relations between node representatives and combining these pieces of information into a single vector. Selecting different node representatives leads to different vectorial representations and therefore to different base classifiers that can be combined. We experimentally show how this methodology significantly improves the classification of graphs with respect to single base classifiers.