基于信息画像的突发事故灾难舆情传播效果的预测模型研究

A Prediction Model for the Effectiveness of Public Opinion Dissemination about Accident Disasters Based on Information Portrait

【目的/意义】对突发事故灾难舆情信息进行精准画像,实现高传播信息的早期分类与识别,并实施精准化的引导对策。【方法/过程】以长沙自建房倒塌事件的微博数据为例,首先使用熵权法对信息传播效果进行评价,其次采用K-Modes聚类对高传播信息构建信息画像,最后基于XGBoost算法构建分类预测模型,并比较不同模型的预测效果。【结果/结论】根据信息画像可将突发事故灾难舆情信息划分为“高传播-官方救援报道类信息”“高传播-官方事故处置类信息”“高传播-大V情感表达类信息”“高传播-官方事故损失类信息”和“低传播信息”五类。同时,XGBoost算法相比其他机器学习分类算法预测性能最好,准确率可达93.94%。【创新/局限】提出一种基于画像的网络舆情信息传播效果的预测方法,以实现对突发事故灾难舆情信息的精准预测;未来会增加多个舆情事件作为数据集并结合深度学习算法,进一步提升模型预测效果。

【Purpose/significance】To accurately portray public opinion information on accident disasters,to realize the early classification and identification of highly disseminated information,and to make precise guidance measures.【Method/process】Taking the microblogging data of the self-built house collapse in Changsha as an example,we firstly use the entropy weight method to evaluate the information dissemination effect,secondly,use K-Modes clustering to construct an information portrait of the highly disseminated information and finally build a classification prediction model based on the XGBoost algorithm and compare the prediction effect of different models.【Result/conclusion】Based on the information portrait,we can classify public opinion information on accident disasters into five categories:"highly disseminated-official accident rescue information","highly disseminated-official accident penalty information","highly disseminated-self-media emotional information","highly disseminated-official accident loss information"and"lowly disseminated information."Meanwhile,the XGBoost algorithm has the best prediction performance compared with other algorithms,with an accuracy rate of 93.94%.【Innovation/limitation】We propose a method for predicting the effect of online public opinion information dissemination based on portraits to realize the problem of accurate prediction of public opinion information on accident disasters;we will add multiple public opinion events as datasets and combine them with deep learning algorithms to further improve the model effect.