Classifying Case Relations using Syntactic, Semantic and Contextual Features

José A. Reyes, Azucena Montes, Juan G. González, David E. Pinto


This paper presents a classification of semantic roles using syntactic, semantic and contextual features. The aim of our work is to identify types of semantic roles involving events and their actors; therefore, we fulfill a feature analysis in order to select the best feature subset which improves the fulfillment of the task. In addition, we compare four classification algorithms: Support Vector Machine (SVM), k-nearest neighbor (k-NN), Bayes classifier and decision tree classifier C4.5. This comparison was made in order to analyze the performance of these algorithms with all features against relevant features for each semantic role category. In our experimentation, we obtain that feature selection improved the performance of algorithms in our classification task, since with relevant features we obtained the best performance of 84.6% with decision tree classifier C4.5. The results for the labeling task can be used for knowledge representation or ontology learning.


Semantic roles classification, knowledge acquisition, natural language processing, machine learning.

Full Text: PDF (Spanish)