基于网络结构与内容分布的新媒体事件聚类研究

马昊 马晓悦

摘 要：[目的/意义]现有新媒体事件的聚类研究聚焦于事件的单一维度属性，并未考虑事件传播的网络结构特征和文本分布特征。[方法/过程]本研究基于信息熵的相关概念，提出基于网络结构熵与内容分布熵的事件聚类模型。模型在表征事件网络结构特征、内容分布特征的基础上完成跨内容事件相似度对比，并使用图表示学习算法与k-means聚类算法对事件进行分析与聚类。本文选取113例微博事件作为实验对象，并使用事件基本属性（点赞、评论、转发等）作为聚类对照实验组。[结论/发现]实验结果分析表明，本研究提出的模型能够捕捉到新媒体事件更深层次的传播、分布特征，能够对现有相似度计算指标进行完善与补充。[创新/价值]本研究不仅能够从多维度层次提取事件的传播特征，即事件网络结构特征和内容分布特征，还能够为舆情预测、管控提供支持，通过熵维度的信息变化监测不同事件之间的传播共性，辅助后续舆情事件的预测与监管。

关键词：网络结构;内容分布;新媒体事件;微博传播;网络结构熵;信息分布熵;事件聚类;图表示学习

DOI：10.3969/j.issn.1008-0821.2021.02.004

〔中图分类号〕G206 〔文献标识码〕A 〔文章编号〕1008-0821（2021）02-0030-12

Abstract：[Purpose/Significance]The existing clustering research of new media events focuses on the single-dimensional attributes of events，and does not consider the network structure characteristics and text distribution characteristics of event propagation.[Method/Process]This research was inspired by the concept of information entropy，and proposed an event clustering model based on network structure entropy and content distribution entropy.The model completed cross-content event similarity comparison on the basis of characterizing event network structure characteristics and content distribution characteristics，then Network Representation Learning algorithm and k-means clustering algorithm cluster the events.This paper selected 113 microblog events as the experimental objects，and used the basic attributes of the events（likes，comments，reposts，etc.）as the cluster control experimental group.[Results/Conclusion]The analysis of the experimental results showed that the model proposed in this study could capture the deeper communication and distribution characteristics of new media events.At the same time，it could improve and supplement existing similarity calculation indicators.[Originality/Value]This research can not only extract the propagation characteristics of the event from multi-dimensional levels，that is，the characteristics of the event network structure and the distribution of event content.Also it can provide support for public opinion prediction and control.The model can also monitor the communication commonality between different events through the entropy dimension of information changes to assist subsequent reflection on public opinion events.

Key words：network structure entropy;information distribution entropy;event clustering;network representation learning;network structure;content distribution;new media events;microblog;communication

信息技術和自媒体行业的飞速发展使得互联网行业中用户创造内容的数量呈现指数级增长[1]。在海量用户生产数据的背景下，信息的自动聚类与分类成为研究焦点[2-3]。具体到新媒体环境中，由用户生成信息所构成的新媒体事件聚类与分类是新媒体舆情管理与检测的一项重点研究[4-5]。……