Text Classification Using Graph-Encoded Linguistic Elements.
Kevin R. Gee, Diane J. Cook · 2005
Inspired by the goal to more accurately classify text, we describe an effort to map tokens and their characteristic linguistic elements into a graph and use that expressive representation to classify text phrases. We outperform the bag-of-words approach by exploiting word order and the semantic and syntactic characteristics within the phases. In this study, we map tagged corpora into a placeholder graph structure and classify the phrases within, using the cross-dimensional linguistic characteristics of each token. Finally, we present heuristics for use in applying this method to other corpora.