Summarization Based on Event-cluster

Abstract

Event-based summarization extracts and organizes summary sentences in terms of the events which stand for complete meaning of sentences. However, the basic event-based extracting method does not take the similarity of events into account, which leads to data sparseness. As a way to solve the problem, we explored a new method, what we call the shallow semantic pattern, which extracts a semantic representation of crucial information in the text. By employing shallow semantic pattern in event-based summarization, not only can we group up the similar events according to the acceptation of word, but also the similarity based on frequent application is detected. We chose four assessment methods in ROUGE to evaluate our system, and used the text sets in DUC 2005 as the inputs of our system to get the summaries. In order to do the comparison, the results of the experiments done on the other four systems are listed, and the outcome shows that our method achieves an encouraging level.