网络舆情事件链演化分析

为防范舆情风险,分析网络舆情的细粒度演化,提出一种去冗余的衍生事件内容关联演化分析框架。通过文本挖掘技术从海量文本流中提取主要的衍生事件,将舆情内容压缩到人工可判读的数量级;利用词移距计算相邻两个时间片上的衍生事件相似度,构建反映演化关系的衍生事件链图。以“上海特斯拉自燃”事件为例进行事件链演化分析,得到微博网络舆情事件发展不同阶段涉事主体在各个衍生事件中的话题转移关系,最后鲁棒性分析的结果验证了该分析方法具有降低微博短文本冗余信息的能力,提高了事件演化关联识别分析的准确性。该研究方法为舆情事件的事后复盘、同类舆情事件的预判和介入、衍生事件的科学研判提供了决策支持。

Evolution of online public opinion based on chain of sub-events

The evolution analysis of the public opinion in emergency is the foundation for the risk prevention and control. An analysis framework for the evolution of the public opinion was proposes based on the event chain. Firstly, text mining technology was used to extract the main sub-events from the massive text stream, thereby reducing the public opinion content to the order of magnitude that can be manual interpretation and discrimination. Secondly, the word mover''s distance was used to calculate the similarity of sub-events on two adjacent time slices, so that the event chain diagram could be constructed. The evolution analysis was carried out with the case of “Tesla spontaneous combustion in Shanghai”. The relationship of topic shifting was built between sub-events at different stages of the evolution. Finally, it was verified by the robustness analysis that the method can reduce the problem of redundant information in short texts of microblogs and improve the accuracy of sub-event correlation. The research results provide decision support for the post-event review, the prediction and intervention of similar public opinion events, and the scientific evaluation of sub-events.