基于双层采样主动学习的社交网络虚假用户检测方法

发布时间：2018-04-17 17:29

本文选题：社交网络 + 虚假用户　；参考：《自动化学报》2017年03期

【摘要】：社交网络的飞速发展给用户带来了便捷,但是社交网络开放性的特点使得其容易受到虚假用户的影响.虚假用户借用社交网络传播虚假信息达到自身的目的,这种行为严重影响着社交网络的安全性和稳定性.目前社交网络虚假用户的检测方法主要通过用户的行为、文本和网络关系等特征对用户进行分类,由于人工标注用户数据需要的代价较大,导致分类器能够使用的标签样本不足.为解决此问题,本文提出一种基于双层采样主动学习的社交网络虚假用户检测方法,该方法使用样本不确定性、代表性和多样性3个指标评估未标记样本的价值,并使用排序和聚类相结合的双层采样算法对未标记样本进行筛选,选出最有价值的样本给专家标注,用于对分类模型的训练.在Twitter、Apontador和Youtube数据集上的实验说明本文所提方法在标签样本数量不足的情况下,只使用少量有标签样本就可以达到与有监督学习接近的检测效果;并且,对比其他主动学习方法,本文方法具有更高的准确率和召回率,需要的标签样本数量更少.
[Abstract]:The rapid development of social networks brings convenience to users, but the openness of social networks makes them vulnerable to the influence of false users.False users use social networks to spread false information for their own purposes, which seriously affects the security and stability of social networks.At present, the detection methods of false users in social networks mainly classify users by their behavior, text and network relationship, because of the high cost of manually tagging user data.As a result, the classifier can use insufficient sample labels.In order to solve this problem, this paper proposes a double-layer sampling active learning based method for detecting false users in social networks. The method uses three indexes: sample uncertainty, representativeness and diversity to evaluate the value of unlabeled samples.The double-layer sampling algorithm combined with sorting and clustering is used to screen unlabeled samples, and the most valuable samples are selected for expert tagging, which is used to train the classification model.Experiments on the Twitter Apontador and Youtube datasets show that the method proposed in this paper can achieve a detection effect close to that of supervised learning by using only a small number of tagged samples when the number of tag samples is insufficient, and compared with other active learning methods.The proposed method has higher accuracy and recall rate and requires fewer tag samples.
【作者单位】：信息物理社会可信服务计算教育部重点实验室;重庆大学软件学院;悉尼科技大学工程与信息技术学院量子计算与智能系统研究中心;广州博冠信息科技有限公司;
【基金】：国家重点基础研究发展计划(973计划)(2013CB328903) 重庆市基础与前沿研究计划(cstc2015jcyj A40049) 国家自然科学基金(71102065) 国家科技支撑计划(2015BAF05B03) 中央高校基础研究基金(106112014CDJZR095502)资助~~
【分类号】：TP393.09;TP181

【相似文献】