基于LAL的微博舆情演化趋势预测与实证研究

Prediction and Empirical Study on the Evolution Trend of Public Opinion in Microblog Based on LAL

[目的/意义]实现对微博舆情演化趋势的有效预测,为舆情预警和相关部门舆情治理提供有力依据。[方法/过程]定义了微博舆情信息量,并提出了一种微博舆情信息量组合预测模型,该模型分别利用Logistic模型学习舆情信息量变化规律,ARIMA模型学习舆情信息量线性关系,LSTM模型学习舆情信息量非线性关系。在此基础上,使用方差倒数法求得组合权重,最后将各单项预测结果进行加权求和得到组合预测结果。[结果/结论]实验表明,Logistic-ARIMA-LSTM(LAL)组合预测模型不仅能够捕捉舆情信息量的整体演化趋势,而且能够捕捉时间序列中的线性关系和非线性关系,有效降低预测误差。LAL模型可以应用于舆情预警、舆情反转识别和舆情推演等方面,为相关部门进行舆情治理提供参考。

[Purpose/significance]The paper aims to realize the effective prediction of the evolution trend of public opinion in mi⁃croblogs,and provide a strong basis for public opinion early warning and public opinion management of relevant departments.[Method/process]Information quantity of public opinion in microblog is defined,and a combined prediction model is proposed.The model uses Logistic model to learn the changing law,uses ARIMA model to learn the linear relationship,and uses LSTM model to learn the nonlinear relationship.On this basis,the combined weights are obtained by using the reciprocal variance method,and finally the combined predic⁃tion results are obtained by weighted summation of the individual prediction results.[Result/conclusion]The experiment shows that the Logistic-ARIMA-LSTM(LAL)combined forecasting model can not only capture the overall evolution trend of public opinion informa⁃tion,but also capture the linear relationship and nonlinear relationship in time sequence,thus effectively reducing the prediction errors.The LAL model can be applied to public opinion early warning,public opinion reversal identification and public opinion deduction,which provides reference for relevant departments’public opinion management.