摘要
利用多个相关领域舆情之间的复杂关联性,处理特定领域新闻的不连续所导致的舆情缺失问题,使其能够用于分项价格指数预测。提出基于自学习图神经网络模型的舆情指数网络方法,将若干相互关联的具体领域的舆情指数作为网络节点,日度舆情量化值视为时间序列,通过学习其中隐性图结构特征和舆情数据时序特征,构建成动态舆情指数网络,实现对稀疏舆情值的填补。对食品价格指数和有色金属生产价格指数预测的实证研究表明,经过该方法补齐的舆情指数和相应价格统计指数相关性达到最高,且预测精度得到提升,验证了此方法的有效性。
The complex relevancy between public opinion in multiple related fields is exploited to deal with the lack of public opinion consistency caused by the discontinuity of news in specific fields,so that it can be used for price index prediction of sub items.This research proposes a public opinion index network method based on self-learning graph neural network model,which takes public opinion indexes in several related specific fields as network nodes,and daily public opinion quantitative values are regarded as time series.By learning the structural characteristics of hidden graphs and the temporal characteristics of public opinion data,a dynamic public opinion index network is constructed to fill the sparse public opinion values.The empirical research on the prediction of food price index and non-ferrous metal production price index shows that the correlation between the public opinion index supplemented by this method and the corresponding price statistics index reaches the highest,and the prediction accuracy is improved,which verifies the effectiveness of this method.
作者
曹雷
尚维
谢士尧
王向
CAO Lei;SHANG Wei;XIE Shiyao;WANG Xiang(Chinese Academy of Sciences,Beijing,China;State Grid Corporation of China,Beijing,China)
出处
《管理学报》
北大核心
2023年第3期411-421,431,共12页
Chinese Journal of Management
基金
国家电网有限公司总部管理科技资助项目(1400-202157207A-0-0-00)。
