Extractive text summarization aims to create a condensed version of one or more source documents by selecting the most informative sentences. Research in text summarization has therefore often focused on measures of the usefulness of sentences for a summary. We present an approach to sentence extraction that maps sentences to nodes of a hierarchical ontology. By considering ontology attributes we are able to improve the semantic representation of a sentence's information content. The classifier that maps sentences to the taxonomy is trained using search engines and is therefore very flexible and not bound to a specific domain. In our experiments, we train an SVM classifier to identify summary sentences using ontology-based sentence features. Our experimental results show that the ontology-based extraction of sentences outperforms baseline classifiers, leading to higher Rouge scores of summary extracts.
展开▼
