作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程

• •    

基于贝叶斯衍生分类器的社交网络用户影响力评价模型

  • 发布日期:2023-10-12

Social network user influence evaluation model based on Bayesian Derived Classifier

  • Published:2023-10-12

摘要: 为了防止社交网络中的负面信息快速传播,就需要通过评价社交网络中用户的影响力,来找出影响力大的社交网络节点。针对传统算法在社交网络领域中交叉特性缺失的问题,该文结合高斯贝叶斯衍生分类器提出了一种网络用户影响力评价模型。该模型首先结合用户活跃度、用户联系度、用户覆盖度等维度,建立了社交网络用户影响力刻画指标,同时考虑了社交网络用户之间的关系特征和用户自身的行为特征,降低僵尸粉和垃圾社交网络对网络评价结果的影响,通过建立连续属性朴素贝叶斯分类器的方法,提出了基于一种高斯贝叶斯衍生分类器的模型求解方法。使用新浪微博中152059423条媒体报纸用户评论作为实验数据,深入分析了影响该评价模型的关键因素,利用仿真软件完成了和HRank等传统模型对比实验,验证了模型的可行性。实验结果表明,该模型体现了社交网络用户的交叉特性,提升了模型的实用性。相比于其他传统算法,该模型在算法分类误差更趋于稳定,分类结果的误差率更低,适应性更好。

Abstract: To prevent the rapid spread of negative information in social networks, it is necessary to identify influential social network nodes by evaluating the influence of users in social networks. In response to the problem of missing cross characteristics in traditional algorithms in the field of social networks, this paper proposes a network user influence evaluation model combined with Gaussian Bayesian-derived classifiers. The model first combines user activity, contact, coverage, and other dimensions to establish social network user influence characterisation indicators. At the same time, it considers the relationship characteristics between social network users and the user's behaviour characteristics to reduce the impact of zombie fans and garbage social networks on network evaluation results. A model-solving method based on a Gaussian Bayesian-derived classifier was proposed by establishing a continuous attribute naive Bayes classifier method. The key factors affecting the evaluation model were analysed in depth using 152059423 media and newspaper user comments from Sina Weibo as experimental data. A comparative experiment with traditional models such as HRank was completed using simulation software to verify the feasibility of the model. The experimental results indicate that the model reflects the cross characteristics of social network users and improves the model's practicality. Compared to traditional algorithms, this model tends to have more stable classification errors, lower error rates in classification results, and better adaptability.