基于时序共词网络的社交平台话题检测与演化研究

Topic Detection and Evolution in Social Media Platforms Based on a Temporal Co-word Network

社交平台是网民传达观点和情感的重要途径,分析社交平台话题分布及演化过程能够揭示舆情热点及传播发展过程,对引导公众舆论具有重要的参考作用。本研究利用网络社团演化的方法检测社交平台话题并分析其演化过程。首先,对用户发布的文本内容进行时间切片,构建时序共词网络并提取各时间切片的主干网络,利用Leiden算法检测社团来表示话题。其次,提出基于社团正向和反向转移概率及社团规模的话题演化事件检测方法,识别话题演化中的持续、增长、收缩、合并、分裂、新生以及消亡等事件。以新浪微博平台新冠肺炎疫情相关微博为例,在话题检测中发现,主干网络相较于原始网络能够检测到更多话题,话题内容区分粒度更细。在话题演化分析中,发现了公众情绪由消极转积极、防控和医疗工作专业化、国际疫情蔓延态势及疫情对经济的影响逐步扩大等演化路径。

Social media platforms are essential for netizens to express opinions and sentiments.Analyzing topic distribution and their evolution on social media platforms can reveal hot topics and their changes to provide important references for influencing public opinion.This study employs network community evolution discovery to detect topics and analyze their evolution on social media platforms.First,user-generated textual contents are divided into several slices to construct a temporal co-word network,and the backbones of co-word network in each time slice are extracted.Then,network communities are discovered through the Leiden algorithm to represent topics.To detect topic evolution,a method of detecting topic evolution events is proposed on the basis of the forward and backward transfer probabilities and community size.Therefore,events,such as continuing,growing,shrinking,merging,splitting,forming,and dissolving are identified.Considering the microblogs about COVID-19 in Sina Weibo as an example,a larger number of topics,with finer granularity are uncovered in backbones than in the original co-word networks.Topic evolution paths are also found,such as changes in users’sentiment from negative to positive,professionalization of pandemic prevention and medical work,the global spread of the pandemic,and the growing economic impact of the pandemic.