基于自然语言处理的防洪重大事件新闻舆情分析——以长江5号洪水为例

Analysis on news and public opinions on major flood control events based on natural language processing——taking Yangtze River Flood No.5 as an example

目前对防洪重大事件新闻舆情研究较少,针对今日头条平台中关于长江5号洪水过境重庆期间,头条新闻的新闻报道和公众评论组成的新闻舆情进行研究。基于自然语言处理的中文分词、LDA主题模型等方法对由新闻报道和公众评论建立的语料进行探究,通过添加防洪先验知识改进LDA主题模型挖掘效果,从起始、集中、衰退3个阶段进行舆情发展思考,从新闻媒体和公众2个方面对洪水过境重庆的热度和主题2个维度组成的热点进行分析,提出防洪重大事件新闻舆情挖掘分析框架。分析结果表明:新闻舆情时间性强,热度、主题与时间关联度高,新闻舆情热点演化过程与洪水演进过程基本一致,可为政府部门把握舆情演化路径、提前做好舆情预警、掌握舆论主动权提供参考。

So far,there has been little research on the news and public opinions on major flood control events.This paper focuses on the news and public opinions from headlines and public comments on Yangtze River Flood No.5in Chongqing on Chinese news platformToutiao.The corpus established by news reports and public comments is analyzed based on natural language processing method such as Chinese word segmentation,LDA topic model etc.The mining performance of LDA topic model is improved by adding priori knowledge of flood control.This paper considers the development of public opinions from three stages(i.e.,initiation,concentration and recession)to analyze the hot spots in two dimensions(i.e.,popularity and topics)of the flood event in Chongqing from two aspects:news media and the public.The mining and analysis framework of news and public opinions on flood control events is then proposed.The results indicate that the news and public opinions has strong timeliness,showing high correlation between popularity,theme and time.The evolution process of hotspots of news and public opinions is basically the same as that of flood evolution,which can provide reference for government departments to grasp the evolution path of public opinions,develop early warnings of public opinions,and lead the initiative of public opinions.