基于大规模网络社群分割的国际分众舆情分析——以2022年G20峰会为例

Opinion Analysis of Focus Groups Based on Community Detection in Massive Networks:The Case of 2022 G20 Summit

[研究目的]国际舆论研究既需要覆盖范围广泛,也需要针对细分群体进行分众化分析,而已有的以文本为中心的舆情分析方法难以满足上述需求。因此本文提出一种以网络为中心的国际分众舆情分析方法。[研究方法]首先采用个性化网页排名算法对社交媒体用户随机游走采样,得到未被社交机器人污染的大规模用户网络,然后基于社交媒体舆论嵌入在各种同质网络中这一特征,用经典稀疏主成分分析对用户网络进行社群聚类,将具有共同纽带或身份的个体划分入同一社群,最后用深度学习分别针对各个社群进行舆情分析。[研究结论]通过对2022年G20峰会的国际舆情分析,发现各社群的舆论表达与其身份高度相关,验证了本文提出方法的有效性。除了在对抗社交机器人干扰、用户群体细分方面做出改进性工作,本文还为国际舆情演化的实证研究提供了方法论支撑。

[Research purpose]The study of public opinion requires aggregate-level measurement and the analysis of focus groups.The text-centric approach fails to meet these requirements.Therefore,we propose a network-centric approach for public opinion analysis of social media focus groups.[Research method]We first sample targeted accounts from a large social graph with the personalized page-rank.Then,we utilize the vintage sparse principal component analysis to cluster millions of accounts based on the social network structure,where like-minded individuals interact with each other and reinforce opinions.Opinions of each focus group are mined via a deep learning approach.[Research conclusion]We empirically analyze international public opinion related to the 2022 G20 summit.The results show that the focus groups express opinions closely connected to their identities,validating the proposed approach.In addition to suppressing the interference of social bots in opinion analysis and segmentation of the user group,this study further provides methodological contributions to empirical research on the evolution of international public opinion.