基于传播树映射与兴趣降噪的微博转发预测模型  

Integrating Propagation Tree Mapping and Interest Denoising for Microblog Reposting Prediction

作　　者：徐建民 冯帆 张雄涛 Xu Jianmin;Feng Fan;Zhang Xiongtao(Hebei University,Baoding 071000;Nanjing University,Nanjing 210000)

机构地区：[1]河北大学网络空间安全与计算机学院,保定071000 [2]南京大学信息管理学院,南京210000 [3]河北大学管理学院,保定071000

出　　处：《情报杂志》2025年第4期117-126,共10页Journal of Intelligence

基　　金：国家社会科学基金项目“智媒时代‘信息茧房’效应度量及应对策略研究”(编号:23BTQ092)研究成果。

摘　　要：[研究目的]微博转发预测是网络舆情分析的关键环节,旨在通过学习准确的微博表示和用户兴趣表示提升微博转发预测性能。[研究方法]针对现有研究在表示学习方面的不足,提出一种基于传播树映射与兴趣降噪的微博转发预测模型。该模型采用传播树映射方法和滤波器降噪方法对微博表示和用户兴趣表示进行深入研究。首先,将微博映射为传播树,设计双重图池化机制对微博传播树进行表示学习;其次,基于用户历史转发微博集,设计一种时间感知滤波器对用户兴趣进行降噪学习;最后,根据待预测微博表示与用户兴趣表示之间的相似度,预测微博的转发概率。实验采用新浪微博数据对模型的合理性和有效性进行验证。[研究结果/结论]实验结果表明,相较于主流预测方法,所提模型在多项指标上均体现出约10%的性能提升。研究发现:采用传播树综合内容特征和传播特征有利于更充分地对微博进行表示,且降低用户兴趣中噪声信息有利于更准确地刻画用户兴趣。[Research purpose]Microblog reposting prediction is a critical component of network public opinion analysis,which aims to enhance the performance of microblog reposting prediction by learning accurate representations of microblogs and users'interests.[Research method]Addressing the shortcomings of current studies in representation learning,this paper proposes a microblog reposting prediction model that leverages propagation tree mapping and interest denoising.The model employs a propagation tree mapping technique and a filtering denoising approach to study the representation of microblogs and users'interests.Firstly,each microblog is mapped as a propagation tree,and the dual graph pooling mechanism is designed to learn its representation.Secondly,based on the user's historical reposting microblog,a time-aware filter is designed to reduce the noise within users'interests.Finally,the model predicts the reposting probability of each microblog by assessing the similarity between the microblog representation and the user interest representation.The model's rationality and validity are confirmed through the use of Sina microblog data.[Research result/conclusion]The experimental results show that,compared with the mainstream prediction methods,the proposed model has a performance improvement of about 10%across multiple metrics.The findings suggest that the use of propagation trees to synthesize content and propagation features significantly enriches the representation of microblogs,while reducing noise in users'interests facilitates a more precise characterization of users'interests.

关 键 词：微博转发预测 传播树映射 兴趣降噪 双重图池化机制 时间感知滤波器 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] G202[自动化与计算机技术—控制科学与工程]