Learning Discourse Relations from News Reports: An Event-driven Approach

Nowadays, technologies allows us to store large volumes of information in different formats. It represents a challenge due to the lack of semantic in retrieval and extraction process of information efficiently. A possible strategy is to transform unstructured information into structured data. In recent years, ontologies have been widely used as an alternative to represent structured data from texts. This paper presents a new approach based on linguistic markers for ontology learning and population by considering cognitive aspects in order identify discourse relations between events from news reports. The main idea is to find concepts (event type), discourse relations (ontological relations) between events and class instances (real events). Our approach shows promising results for learning discourse relations in terms of F-measure.