基于时空语义的微博突发事件检测与短期预测研究

Microblogs have dramatically changed the mechanism of information propagation. It has been an inevitable issue for governments and enterprises to face the challenges that microblogs bring to public safety management. Since 2010, a lot of emergent events were firstly reported in microblogs, and this trend is even becoming more and more prominent. As microblogs have shown much influence in public safety and sentiment management, governments and enterprises have to employ effective ways to monitor and predict emergent events in microblogs; otherwise when emergent events in microblogs happen, they will be in a passive situation and even the society's stability will be affected. Based on the properties of microblog information and its diffusion, this project is focused on some key issues in detecting emergent events from microblogs. Firstly, we study the spatiotemporal information extraction from microblog and aim at constructing a spatiotemporal model for microblog events so as to describe the thematic and evolutional characters of microblog events effectively. Secondly, we introduce a measurement mechanism and a filtering approach to improve the quality of microblog information. Next, we explore the spatiotemporal-model-driven method to detect microblog events. Moreover, sentiment analysis in microblogs is researched and a new approach based on microblog forwarding and reviews is studied. Finally, we build a prototype to conduct short-term prediction over some selected microblog events, in order to test the applicability of our research results. This project can discover useful decision evidence for governments and enterprises to realize microblog events monitoring and emergency treatment. It can also provide some scientific support to maintain the society's stability and enhance China's ability in dealing with public emergencies.

微博颠覆了传统的信息扩散模式，如何适应微博带来的公共安全挑战是政府部门及企业必须深思的问题。近两年，微博在突发事件中展现了越来越强的首报能力和影响力，如果缺乏对微博突发事件的监控以及短期预测手段，一旦发生微博突发事件，将使政府部门及企业处于被动地位，甚至危及社会稳定。本课题针对微博信息及传播的特点，重点研究微博时空信息抽取和微博突发事件的时空模型，以刻画微博突发事件的主题特征和演化特征；引入微博质量评价和信息过滤技术，以提高微博数据质量；探索时空模型驱动的微博突发事件检测方法，并从转发和评论的角度研究微博情感分析，最终构建微博突发事件的主题、事件演化和情感演化的完整过程；在此基础上建立微博事件短期预测模型与原型系统针对特定事件进行应用研究。本研究可以为政府部门和企业进行微博突发事件监控与应急处置提供决策依据，为维护社会安定团结，增强我国突发事件应急管理水平提供科学支撑。

微博颠覆了传统的信息扩散模式，如何适应微博带来的公共安全挑战是政府部门及企业必须深思的问题。近年来，微博在突发事件中展现了越来越强的首报能力和影响力，如果缺乏对微博突发事件的监控以及短期预测手段，一旦发生微博突发事件，将使政府部门及企业处于被动地位，甚至危及社会稳定。本项目系统研究了微博突发事件检测问题，提出了一种基于时空信息的细粒度微博突发事件检测方法。该方法可以有效地表达微博突发事件的新闻学语义，并包括时间、地点物等。此外，项目从特征选择的角度研究了微博突发事件检测中的特征选择方法，提出了一种基于新闻学的微博突发事件特征选择方法，有效地提高了微博事件检测中的特征选择的效率。针对微博情感分析问题，我们提出了一种基于新词扩充和特征选择的微博观点句识别方法和基于“情感表达式”的微博情感倾向抽取方法。在实际微博数据集上的实验表明所提方法的性能优于已有方法。最后，本项目构建了一个微博突发事件检测与情感演化分析原型系统，并采集了真实的新浪微博数据集进行了实验验证。本项目的研究为政府部门和企业进行微博突发事件监控与应急处置提供了决策依据，为维护社会安定团结，增强我国突发事件应急管理水平提供了新的思路。. 本项目共出版专著1本，在国内外期刊和国际会议上发表了29篇论文，其中期刊论文19篇（SSCI/SCI期刊4篇，EI外文期刊8篇，国家自然科学基金委管理学部A类重点期刊《情报学报》2篇），会议10篇（其中CCF A类会议2篇，C类会议7篇），EI收录17篇。本项目的研究成果既具有理论创新，又有实际系统开发，项目成果具有广阔的应用前景和可以预期的经济效益。