新闻热点话题发现及演化分析研究与应用

【作者】 陈龙；

【导师】 徐建；

【作者基本信息】 南京理工大学 ， 软件工程（专业学位）， 2017， 硕士

【摘要】 热点话题是因网络报道而引起人们广泛关注的话题,热点话题发现与演化研究有利于社会大众知晓当前舆论焦点和政府进行良性舆论引导,能够防止有心之徒利用网络的便捷性、不可控性牟取不正当利益,制造社会矛盾。本文主要就新闻热点话题发现及对热点话题演化偏移过程进行研究,主要包括以下几个方面:1、引入了 LDA主题模型,对新闻报道采用基于TF-IDF的词-权值模型和基于语义理解的LDA模型两种文本向量建模方式。在此基础上,针对传统单核心话题描述模型对多核话题描述欠缺的问题,提出了一种多核心话题描述模型,能够识别同一话题下不同的关注核心,并给出了模型构造方法:采用划分聚类与层次聚类结合的方法对新闻报道进行精确聚类。实验表明,多种文本向量建模相结合的方式以及多核心话题描述模型能够提高新闻话题的聚类效果。2、根据热点话题特征分析的结果,将新闻的热度量化为媒体报道热度和网民关注热度,并采用基于两者的复合关注度描述热点话题的热度;同时引入"话题指数",采用基于时间窗口的分段话题聚类方法对热点话题生命周期演化过程进行分析,提出了一种基于多核心话题描述模型的话题演化偏移分析方法,将演化过程看成话题内核心事件的转移过程。实验表明该方法能很好的发现热点话题的演化偏移过程。3、基于上述研究成果,设计并实现了新闻热点话题发现及演化分析子系统,该子系统是移动新闻监测和分析平台的一个重要功能模块,集成了新闻报道预处理、热点话题发现、热点话题演化分析等功能,能够实时发现当前热点话题并展示给用户。更多还原

【Abstract】 Hot topics are those that are widely reported by network social medias and attract much attention from people.The research on the hot top i c discovery and evolution is beneficial to the public awareness of the current public opinion and the guidance of benign public opinion by Government.It can also prevent people from making social contradictions and seeking illegitimate interests through network.This paper mainly focuses on the discovery of news hot topics and the evolution process of hot topic,which mainly includes the following aspects:1.We use the word-weight model of TF-IDF and the LDA model of semantic understanding for news report based on LDA theme model.On the basis,we proposed a multi-core topic description model that can identify different core of attention under the same topic for the lack of traditional single-core topic description model.The model combine the division of clustering and hierarchical clustering method for precise clustering.Experiments show that the model can improve the clustering effect of news topic.2.According to the result of the analysis of hot topic characteristics,the news heat is quantified as the media coverage and the user’s attention about the heat,We use the combination of the above two factors into the description of the heat.At the same time,the"topic index" is introduced,and the evolution process of hot topic life cycle is analyzed by segmented topic clustering method based on time window.Based on multi-core topic description model,a method of evolution analysis of topic evolution is proposed.The evolution process is regarded as the transfer process of the core event in the topic.Experiments show this method can well find out the evolution process of hot topics.3.Based on the above research results,we design and implement a hot news topic discovery and evolution analysis subsystem,which is an important function module of mobile news monitoring and analysis platform.The system integrates news gathering,news preprocessing,hot topic discovery and hot topic evolution analysis,which can discover hot topics and present them to users in real time.更多还原