多维视角下的网络舆情预测研究——以公共政策网络舆情为例

At present, central government agencies put a high value on the management of network public opinions. When the government is contained in the event, large-scale network public opinions will commonly form in a short time. Related departments cannot provide timely and effectively coping strategies by solely relying on monitoring techniques, and thus falling into a passive position. Therefore, through focusing on the key scientific problem that how to effectively predict the risk of network public opinions, this project aims to put forward the control of public opinions and innovatively establish a forecasting system at multiple angles of view by fusing frontier theories and methods from various fields. ① Based on the complex network theory, deduce the general form of the degree distribution function with the steady state of public opinion subject network, in order to achieve the effective prediction of subject structure of public pinions. ② Propose a new information popularity of network public opinions prediction model by combining neural network algorithms and evolutionary algorithms; ③ Create a new population sentiment-evolution dynamic model by applying the Mass-Spring-Damping system, and use this model to predict the sentiment evolution of network public opinions through employing optimization algorithms; ④ Build the case base of network public opinions caused by the release of public policies and extract key elements of public opinions by using data mining and deep learning methods, in order to carry out empirical research. The purpose of this project is to provide timely and diversified theoretic and data support for network public opinion management, in order to prevent and resolve potential social contradictions from the origin.

当前，中央各部门愈加重视网络舆情治理。当涉及政府主体的事件在网络中发酵，往往迅速形成大规模网络舆情，仅依靠监测技术，相关部门难以及时做出有效应对，导致陷入被动地位。本项目针对如何有效预测舆情风险这一关键科学问题，将舆情防控关口前移，通过融合多领域前沿理论与方法，创新性的构建多维视角下网络舆情预测体系。①基于复杂网络理论，推导舆情主体网络稳态度分布函数，以实现舆情主体结构的有效预测；②融合神经网络算法和进化算法，提出舆情信息热度预测模型；③引入质量-弹簧-阻尼系统理论，构建一种新的舆情群体情绪演化动力学模型，并结合优化算法，实现对舆情情绪演化的预测；④公共政策网络舆情预测分析，构建公共政策网络舆情案例库，基于数据挖掘和机器学习方法对舆情关键要素进行提取，形成数据集，展开实证研究。本课题目标是为舆情治理提供及时、多元化的理论和数据支撑依据，以期从源头上预防和化解由于网络舆情可能引发的社会矛盾。

党和国家对公共政策网络舆情服务决策的需求与日俱增，尤其是如何对各种可能出现的舆情情景进行充分预研和预判以及如何从即时处置型向事前和事中预警型转变,进而实现舆情防控关口前移，把握网络舆情控制的主动权。本项目基于舆论动力学模型、复杂网络、自然语言处理等方法，构建了一系列网络舆情预测模型与面向多领域的实证模型，并基于真实公共政策相关舆情数据对模型效用以及所提出的理论假设进行验证。具体研究内容和解决问题如下: 构建了公共政策网络舆情案例库，构建了面向不相关离散意见的网络舆情预测模型，构建了基于阻尼震荡模型的网络舆情情绪预测模型，构建了面向网络谣言的网络舆情治理效果预测模型，构建了基于富人俱乐部检测指标体系的关键词优化算法，面向新冠肺炎疫情、医疗安全、高铁交通等不同重点领域展开了一系列实证研究，等。在本项目的资助下，完成论文15篇，其中SCI/SSCI期刊论文11篇，国家自然基金委认定的A类期刊3篇。培养硕士研究生5名。总体来说，本项目研究成果研究结果不仅丰富和完善了现有网络舆情研究的理论与方法体系，也为政府相关部门治理网络舆情问题提供定性与定量相结合的决策支持。