Abstract: Internet is an important source of information and a lot of news websites have become the main places for generating and propagating information. Most information on Internet is irregular, noisy and dynamic. To organize and aggregate the information effectively, this paper proposes a method combining a static model and a dynamic model for network event aggregation. The static model clusters textual documents based on the similarity of content, and the dynamic one is to describe the dynamic process of network events based on Hidden Markov Models（HMMs） and determines whether the current news follow the law of event development. The proposed method describes network events based on both static content and dynamics instead of only the former used in a lot of previous studies, which benefits the recognition of network events. The experiment is conducted on a real-world dataset, and the experimental results demonstrate the effectiveness of the proposed method in network event aggregation.

Key words: event aggregation, dynamic development law, dynamic time series model, static modeling

摘要： 网络世界是人们获取信息的重要来源，各大新闻门户网站成为信息产生和传播的主要聚集地。网络信息具有不规范、噪声大、动态性等特点。为了有效地组织网络信息，提出一种结合静态和动态建模的方法来对网络新闻事件进行聚合。静态建模基于新闻发帖内容的相似程度进行事件的初步聚合；动态建模是指基于隐马尔可夫模型来刻画事件的动态发展过程，并判断当前新闻发帖是否符合事件的发展规律。该方法将事件聚合从仅考虑内容空间中的相似度推广至包含内容及事件动态规律两个方面的空间来进行相似度度量，对内容相似的不同事件能够进行更好地区分。通过实际数据上的实验测试结果表明，该方法能够有效提升事件聚合的准确性。

关键词: 事件聚合, 动态发展规律, 动态时序模型, 静态建模