微博热点隐话题发现及其时序特性研究

There are many constantly updated hot topics in microblog, which are caused by the specific events, and are a kind of explicit topics. Extracting essential attributes from the explicit topics, we can get some abstract upper concepts, called “implicit topics”. Digging the implicit topics and their rules in microblog is very meaningful in some fields, such as information retrieval, public opinion assessment, business marketing, and so on. However, little research has been done in this area. In such circumstances, this project researches the theory, method and technology of implicit topic detection and time sequence property analysis, by using the explicit topics, subtopics, the temporal relations between subtopics in knowledge graph, and combining the methods of upper concept extraction and series data mining. The focus of this project include four aspects: 1) Setting up the model of implicit topics extraction, extracting the implicit topics and the hierarchy. 2) Building derived implicit topic sequences, designing the algorithms of fast mining and updating sequential patterns. 3) Performing a statistical analysis of the heat time series of implicit topics, and proposing an algorithm for forecasting the heat trend. 4) Developing a system facing public opinion supervision and early warning in microblog, with the method of sequence property analysis of implicit topics.

微博中不断更新的热点话题涉及具体事件，是一种显式话题。抽取显式话题共同本质属性得到的上位概念具有抽象性，我们称之为“隐话题”。挖掘微博热点隐话题与其发展规律，对于信息检索、舆情研判、企业营销等具有积极意义，但目前国内外对这方面的研究还仅仅处于初级阶段。针对这种情况，本项目利用微博知识图谱提供的显式话题、子话题及子话题间的时序关系，结合上位概念抽取与序列数据挖掘方法，研究隐话题发现及其时序特性分析的理论、方法和技术。重点研究：1)搭建隐话题抽取模型，提取隐话题及其层次结构。2)构建衍生隐话题序列，设计增量式序列模式挖掘方法及快速更新算法。3)统计隐话题热度时变数据，分析隐话题热度时序特征，建立隐话题热度时序预测模型。4)结合隐话题时序特性分析方法，研发面向微博舆情监测预警的应用系统。

有效提取微博热点显式话题背后的隐含知识及其变换模式，对于舆情研判、信息检索、企业营销等具有重要的科学和应用价值。目前，这方面的研究正处于起步阶段，仍然存在许多有待研究的科学问题。本项目旨在研究隐话题发现及其时序特性分析的理论、方法和技术，首先构建层次隐话题抽取模型。进一步地，设计增量式序列模式树生成算法，以及隐话题热度时变特征分析方法和预测模型。最后将研究方法应用于实际，建立面向微博的网络舆情预警监测系统。