基于关联规则的微博话题动态检测与演化分析

发布时间：2018-04-23 09:34

本文选题：话题检测 + 演化分析　；参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：新浪微博目前是国内最大的微博服务平台,微博流中包含着数量众多的,各个领域的新闻事件。目前新浪上有各种各样的带标签的话题事件,已有很多的研究针对于微博上的话题检测,但很少有专门针对特定话题进行研究的。一些热点事件,起源于微博,在社会上引起了巨大的轰动,作为新闻媒体或者事件相关的公关团队,更是极为关注该事件的走向,因此针对目标话题的检测与演化分析具有很好的市场需求。话题分为多种类别,包括突发事件,热点事件等等,通过对现有的微博话题检测方法进行分析,发现目前基于微博上的话题检测很少是有目标性的检测,并且要么检测类别单一,要么无法区分所检测事件类别。本文在总结了前人工作的基础上,主要针对上述问题,从以下几个方面进行研究:第一,我们介绍了微博内容的特点,结合微博的话题标签及关联规则挖掘方法,研究了如何针对特定目标话题进行话题检测与跟踪演化分析。第二,考虑到话题可以分为多种类别(突发事件、热点事件、消逝的事件等),我们借鉴了关联规则在商场顾客购物习惯上的应用方法,分析了微博话题标签在微博中的作用,从微博的话题标签入手,修改并提出了关于关联规则的4种演化模式,即新规则、热点事件规则、变化中的规则、消逝的规则等,从而达到同时检测多种类别话题,并明确其各自所属类别的目的,以便为后续的话题演化分析提供支持。第三,针对目标话题的演化进行跟踪。将我们话题检测中所用到方法直接运用于演化分析中,利用关联规则的4种演化模式对目标话题进行演化分析。采用一种方法同时完成话题的检测与演化分析的任务,一定程度上降低了话题检测与演化分析链接的复杂度。第四,将现有的应用到微博话题检测上的方法应用到我们的场景中,并与我们的方法进行对比,对各种方法的结果分析进行了探讨。本文实验结果证明了以下几点:第一,我们的方法可以有效的对特定目标话题进行检测与演化分析。第二,论文中所采用的方法可以同时检测不同类别的话题,并明确其所属类别。第三,本论文中的方法可以同时用于微博的话题检测与演化分析。
[Abstract]:Sina Weibo is currently the largest service platform in the country, Weibo flow contains a large number of news events in various fields. At present, there are a variety of tagged topic events on Sina. There have been many studies on topic detection on Weibo, but few have focused specifically on specific topics. Some hot events, originated from Weibo, have caused a great stir in society. As news media or public relations teams related to the incident, they are particularly concerned about the trend of the incident. Therefore, the detection and evolution analysis of target topics has a good market demand. Topics are divided into many categories, including emergencies, hot events and so on. By analyzing the existing methods of topic detection of Weibo, we find that topic detection based on Weibo is rarely targeted detection. And either detect a single category, or can not distinguish the category of events detected. On the basis of summarizing the previous work, this paper mainly studies the above problems from the following aspects: first, we introduce the characteristics of Weibo's content, combined with the topic label and association rules mining method of Weibo, This paper studies how to analyze the topic detection and tracking evolution for specific target topics. Second, considering that the topic can be divided into many categories (unexpected events, hot events, evanescent events, etc.), we draw lessons from the application of association rules in shopping habits of shopping malls, and analyze the role of Weibo topic labels in Weibo. Starting with Weibo's topic label, this paper modifies and puts forward four evolution modes of association rules, that is, new rules, hot event rules, changing rules, vanishing rules, and so on, so as to detect many kinds of topics simultaneously. The purpose of their respective categories is defined in order to provide support for the subsequent analysis of topic evolution. Third, track the evolution of the target topic. The methods used in topic detection are directly applied to evolutionary analysis, and four evolutionary models of association rules are used to analyze the evolution of target topics. The task of topic detection and evolution analysis is accomplished simultaneously by a method, which reduces the complexity of the link between topic detection and evolution analysis to a certain extent. Fourth, the existing methods applied to Weibo topic detection are applied to our scene, and compared with our methods, the results of various methods are discussed. The experimental results show that: first, our method can effectively detect and analyze the specific target topics. Secondly, the methods used in this paper can detect different categories of topics simultaneously and identify their categories. Thirdly, the method in this paper can be used for Weibo's topic detection and evolution analysis.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.1;TP393.092

【参考文献】