硕士研究生学位论文.docx

硕士研究生学位论文题目：基于BURST的社交网络事件挖掘姓名：树柏涵学号系：信息科学技术学院专业：计算机科学与技术研究方向：搜索引擎与网络信息挖掘指导教师：李晓明教授二〇一二年六月基于BURST的社交网络事件挖掘树柏涵(计算机科学与技术)指导教师：李晓明教授摘要本文以微博和Twitter为对象研究社交网络上的事件挖掘。本文的研究内容主要为两个部分：微博上的实时热门事件挖掘。我们将博文以小时划分，针对每小时的博文挖掘该小时内的热门事件。鉴于微博庞大的数据量，我们通过Burst检测的方法进行事件挖掘。本文以Kleinberg[3]的Burst挖掘方法为基础，结合微博特点提出改进的Burst挖掘算法。本文将挖掘出的Burst作为特征表示博文，对博文进行聚类。根据博文类的大小、类内相似度、HASHTAG等特征，我们对低质量博文类进行过滤，并对保留的博文类提取代表Burst集合。本文不直接展示代表Burst集合，为提高代表Burst集合的可理解性，我们对其进行扩展，并将扩展失败的博文类过滤。Burst集合的扩展不仅提高了自身的可理解性，也解决了噪音Burst所导致的与事件无关的博文类的问题。针对给定的查询，挖掘出与该查询相关的事件性Burst。鉴于社交网络上的噪音Burst，本文利用社交网络上多种的活动挖掘与事件相关的Burst。我们研究如何将多个活动结合得到全局的Burst趋势。我们提出了状态序列的平滑性问题，并给出了一种新的函数对序列平滑性进行建模。针对事件性Burst挖掘问题，本文提出了新的评测指标。以Twitter为数据集进行实验，我们的方法取得较好的效果。关键词：社交网络，微博，Burst挖掘，事件挖掘Burst Based Event Mining in Social NetworkBaihan Shu (Computer Science)Directed by Professor Xiaoming LiAbstractThis articleuses micro blog and Twitter as objects to study event mining in social network. Our study is presented in two parts:Real-time event mining in micro blog. We divide micro blog messages by hour and mine events for each hour. In view of the huge amount of data in micro blog, we utilize burst detection to mine events. Based on the Kleinberg’s[3] burst detection algorithm, we propose a new algorithm, which combines some features of micro blog. We use text-clustering instead of clustering burst, employing burst to represent micro blog messages. We filter inferior quality clusters, and for the remains we extract represented bursts. We expand the represented bursts to improve the intelligibility. Except for improving the intelligibility, the expanding process solves the noise burst problem.Identifying event-related bursts via social media activities for a given query. With respect to the noise bursts in social network, we propose to correlate multiple types of social media activities to find out a global bursty pattern. We model the smoothness of topic trends and define a novel function to quantitatively measure smoothness of state sequence. W