Campos Aleatórios Condicionais Aplicados à Detecção de Estrutura Retórica em Resumos de Textos Acadêmicos em Português

• Alexandre C. Andreani UEM
• Valéria D. Feltrim UEM

Abstract

This paper presents CRF-AZPort, a rhetorical structure predictor for abstracts written in Portuguese. Unlike the AZPort classifier, which predicts the category of each sentence independently, the proposed predictor uses Conditional Random Fields for predicting a sequence of rhetorical categories for a given abstract. CRF-AZPort uses three new attributes in addition to the original AZPort attributes. Results show that rhetorical classification can be modeled as a sequence labeling problem and that considering rhetorical structure contributes to the classification.