中国安全科学学报 ›› 2020, Vol. 30 ›› Issue (12): 37-42.doi: 10.16265/j.cnki.issn 1003-3033.2020.12.006

• 安全社会科学与安全管理 • 上一篇    下一篇

基于ARIMA与LSTM的新冠肺炎网络关注度趋势研究

景楠1 副教授, 胡怡1, 韩喜双**2 研究员   

  1. 1 上海大学 悉尼工商学院,上海 201800;
    2 哈尔滨工业大学 深圳研究生院,广东 深圳 518055
  • 收稿日期:2020-09-17 修回日期:2020-11-18 出版日期:2020-12-28 发布日期:2021-07-15
  • 通讯作者: **韩喜双(1969—),男,黑龙江哈尔滨人,博士,研究员,主要从事城市灾害与应急管理方面的研究。E-mail: xshan@hit.edu.cn。
  • 作者简介:景 楠 (1978—),男,满族,上海人,博士,副教授,从事决策科学、数据挖掘、用户行为分析。E-mail: jingnan@shu.edu.cn。

Trend of COVID-19 network attention based on ARIMA and LSTM

JING Nan1, HU Yi1, HAN Xishuang2   

  1. 1 SHU-UTS SILC Business School, Shanghai University, Shanghai 201800, China;
    2 Shenzhen Graduate School, Harbin Institute of Technology, Shenzhen Guangdong 518055, China
  • Received:2020-09-17 Revised:2020-11-18 Online:2020-12-28 Published:2021-07-15

摘要: 为有效监控和管理新型冠状病毒肺炎(COVID-19)引起的网络舆情,基于自回归移动平均(ARIMA)模型和长短期记忆(LSTM)神经网络预测和分析舆情数据,利用百度指数收集全国及武汉市网民对COVID-19的关注度数值,形成时间序列数据,并构建舆情模型;对舆情模型进行参数估计、模型诊断和模型评价。结果表明:此疫情的网络舆情前驱期为4天,爆发期为7天,波动期为14天, 消退期为32天,到达峰值的时间为13天;该模型可较好地模拟COVID-19网络舆情关注度的变化趋势,且局部地区的数据拟合模型预测效果优于全国数据拟合模型。

关键词: 自回归移动平均(ARIMA)模型, 长短期记忆(LSTM), 新型冠状病毒肺炎(COVID-19), 网络舆情, 时间序列

Abstract: In order to effectively monitor and manage online public opinion caused by COVID-19, data of public opinion were predicted and analyzed based on ARIMA model and LSTM neural network. Then, attention value of COVID-19 from network users in Wuhan and the whole country was collected by using Baidu index. Time series data were developed, and prediction models were established. Finally, parameter estimation, model diagnosis, and model evaluation were carried out for each prediction model. The results show that prodromal period, outbreak period, fluctuation period and fading period of internet public opinion are 4 days, 7 days, 14 days and 32 days respectively, and the time it takes to reach a peak is 13 days. The model can well simulate change trend of COVID-19 network public opinion attention, and prediction results of local data fitting model is better than that of national one.

Key words: auto regressive integrated moving average (ARIMA), long-short term memory (LSTM), corona virus disease 2019 (COVID-19), network public opinion, time series

中图分类号: