TAC 2008 Update Summarization Task of ICL

The update task of multi-document summarization aims at automatically generating the summaries of some event developing with time going. Based on our previous system of multi-document summarization, we summarize the document sets with or without history. In our previous system, the design of features is the important part. Here, in order to adapt to the updating task we introduce a new ‘filtering’ feature. With history, filtering features are calculated to exclude those sentences which are similar to the “history”. The principle of designing filtering features is to distinguish the current documents from the previous documents, and reflect the main idea of current documents. In this work, we combine two kinds of similarity metrics and two computing strategies to get the filtering feature value. The experimental results will show which combination is more effective to catch the developing process of an event.