Sentence Ordering with Event-Enriched Semantics and Two-Layered Clustering for Multi-Document News Summarization

We propose an event-enriched model to alleviate the semantic deficiency problem in the IR-style text processing and apply it to sentence ordering for multi-document news summarization. The ordering algorithm is built on event and entity coherence, both locally and globally. To accommodate the eventenriched model, a novel LSA-integrated two-layered clustering approach is adopted. The experimental result shows clear advantage of our model over event-agonistic models.