基于增量图聚类的动态多文档摘要算法

Update summarization using incremental graph clustering

目前的动态文摘方法几乎都基于文档批处理机制,无法适应实际应用中表现为不稳定数据流的文档数据,因此无法满足实时更新摘要的需求。针对上述问题,提出了一种基于K近邻句子图模型的动态文本摘要方法。根据K近邻规则构建一个双层句子图模型,用基于密度划分的增量图聚类方法对句子进行子主题划分,最后结合时间因素提高句子新颖度来抽取动态文摘。该方法能基于文档数据流增量式地抽取动态文摘,实现文摘内容的实时更新。在TAC2008和TAC2009的update summarization数据集上的实验结果显示了该方法在动态文摘抽取上的有效性。

Most of the current update summarization methods are based on the document batching mechanism. However, the document datas in practice are usually unstable streams. So these methods can＇ t always meet the need of summarizing docu- ments online. To solve the problem, this paper presented a new update summarization method based on the K-nearest neighbours idea. Firstly, this method adopted the K-nearest neighbour rule to build and update a two-level sentence graph model incrementally. Then it used the incremental graph clustering to divide the text subtopics. Finally it enhanced the sentence＇ s nov- elty by combining the time factor to extract summarizations. The proposed method can handle document data streams incrementally. It extracts and updates the summarizations in real-time. Experiments on TAC2008 and TAC2009 data sets indicate the effectiveness of this method.