摘要
[研究目的]在海量用户生成内容中及时探测和剖析网络暴力事件的衍生舆情能够为舆情事件链的演化分析、同类舆情的研判介入、衍生事件的监测预警提供理论支持。[研究方法]使用BERTopic模型对短文本内容主题建模并采用聚类的方式展示主题的潜在层次结构。根据词向量余弦相似度设计主题衍生度的计量算法,同时融合词共现网络在文档-词语层面信息捕捉的优势以及桑基图直观演示舆情演化过程的特点,衡量主题间的影响力与衍生关系。[研究结论]在开源数据集下多组主题模型的对照实验中,BERTopic模型在短文本建模以及下游任务的平均得分提高2.13%。在网络暴力热点事件的应用实例中,多维细粒度分析与交互式可视化方法可达到直观展示暴力事件的主题聚类、词义关联与演化态势的效果,实现网络暴力事件衍生舆情的探测与分析。
[Research purpose]Timely detection and analysis of the derived public opinion of cyber violence incidents in the mass user-generated content can provide theoretical support for the evolution analysis of the chain of public opinion events,the intervention of similar public opinion events,and the monitoring judgment and early warning of derivative events.[Research method]The BERTopic model is used to model short text content topics and the underlying hierarchy of topics is shown in a clustering manner.Combining the advantages of word co-occurrence network in capturing information at the document-word level and the characteristics of Sankei chart to visually demonstrate the evolution process of public opinion,as well as designing the measurement algorithm of topic derivation degree according to the cosine similarity of word vector,while combining the advantages of word co-occurrence network in capturing information at the document-word level and the characteristics of Sankei chart to visually demonstrate the evolution process of public opinion,the impact and derivative relationships between topics are measured.[Research conclusion]In the control experiments of multiple sets of theme models under the open source dataset,the BERTopic model increased by 2.13%in the short text model and downstream task scores.In the application examples of the hotspot cyber violence,the methods of multi-dimensional fine-grained analysis and interactive visual exhibition could directly present the results of theme cluster,word meaning association and evolutionary situation,and accurately detect the derivative public opinion of cyber violence incidents.
作者
胡凯茜
李欣
王龙腾
Hu Kaixi;Li Xin;Wang Longteng(School of Information and Cyber Security,People's Public Security University of China,Beijing 102623)
出处
《情报杂志》
北大核心
2024年第7期146-153,共8页
Journal of Intelligence
基金
中国人民公安大学网络空间安全执法技术双一流创新研究专项(编号:2023SYL07)研究成果。
关键词
网络舆情
网络暴力
衍生舆情
舆情监测
短文本
主题建模
BERTopic模型
network public opinion
cyber violence
derived public opinion
public opinion monitoring
short text
topic modeling
BERTopic model
