【作者】 李辉；

【导师】 卜佳俊； 王灿；

【作者基本信息】 浙江大学 ， 计算机应用技术， 2012， 硕士

【摘要】 在互联网信息爆炸的时代,新闻资讯越来越便捷,更新的速度也越来越快。当重大事件发生时,大量的相关新闻报道以及追踪报道便会涌现。对用户而言,如何能找到感兴趣的事件并能迅速了解整件事所包含的信息,成为亟待解决的一个难题。事件一般是随着时间而变化的,其中包含了数个事件发展脉络,挖掘出事件的发展脉络有助于用户快速了解整件事的来龙去脉。本文就重大事件中的事件发展脉络挖掘和事件信息快速展示展开研究,提出了基于时间线的事件组织与摘要技术。围绕关键技术进行了如下工作：首先研究了一种基于时间线的事件组织算法。由用户提供的关键词检索出用户感兴趣的内容,通过聚类算法寻找出重大的新闻事件,并根据事件中新闻在时间线上的的分布挖掘出事件的发展脉络。新闻的时效性保证了当事件有重大发展时,往往会出现密集的跟踪报道,于是新闻在时间轴上的聚集分布便代表事件的重大发展阶段。通过时间线的先后顺序将这些发展阶段进行组织,形成该事件的发展脉络。然后提出了一种基于时间动态发展的事件摘要算法。该算法考虑到了事件随时间线发展的特性,每个发展阶段必须单独生成摘要。提出了历史内容过滤模型,并综合无向句图模型和句子重要度模型为每个句子打分。由于事件中每个发展阶段的重要程度不一样,根据重要度为每个阶段选取相应数量的内容。将每个阶段的摘要按发生时间顺序拼接组成整个事件的摘要。设计并实现了基于时间线的事件组织与摘要原型系统,采用路透社的英文数据集进行实验。实验结果表明这两个算法取得了良好的效果。更多还原

【Abstract】 In the Internet era of information explosion, news information is convenient, and update speed is faster and faster. When an important event occurs, a large number of news reports and follow-up reports will emerge. To the user, how to find events of interest and can quickly understand the information contained in the event has become urgent to solve. Events generally change over time, which contains a number of stages of development, and find out the development venation of the event to help users quickly understand the history of the matter. In this thesis, do a research of finding out the development venation of important events and displaying event information. And propose a timeline-based events organization and summarization technology. The works are the following:Firstly, I propose a timeline-based event organization algorithm. Use keywords provided by the user interested in retrieving the news, through the clustering algorithm to find out the important news events, and find out the development venation of the event through the distribution of news reports based on the timeline. The timeliness of news to ensures that when there is a significant development, often intensive follow-up reports will appear, then news reports on the timeline of events aggregated distribution would represent a significant stage of development. By order of the timeline to organize these stages of development, and build the development venation of the event.Secondly, propose a timeline based event summarization algorithm. The algorithm takes into account the development of events over time-line characteristics of each stage of development must generate a separate summary. Proposed a model of historical content filter, and comprehensive undirected sentence graph model and important sentence model to score each sentence. Since the importance of each stage of development is not the same, according to an important degree select the appropriate number of news for each stage. A summary of each stage in order by time construct the event summary. Design and implement the prototype system of the timeline based event organization and summarization. Using Reuters’s data set in English to do experiments. Experimental results show that both algorithms achieve good results.更多还原