基于信任度排序的社交网络异常账户检测模型的研究

发布时间：2018-04-20 05:35

本文选题：社交网络 + 账户检测　；参考：《上海交通大学》2014年硕士论文

【摘要】：社交网络是web2.0时代兴起的一种网络服务，它将线下的社交活动拓展到线上，允许用户注册账户并在网络上进行交互。社交网络提倡良好的线上社交行为，但是依旧存在通过社交网络账户发布垃圾信息的情况。由于社交网络的开放性与即时性，这些垃圾信息能够迅速而广泛的传播，由垃圾信息传播而引发的负面事件也呈现出越发严重的趋势。因此，针对专门用于发布垃圾信息的异常账户进行识别与限制，对减少社交网络中的垃圾信息具有重要作用。本文的贡献在于将信任度的概念引入社交网络中，提出一种计算模型对社交网络账户信任度进行评估，从而根据评估结果对账户进行排序。同时，对社交网络账户间关系进行深入挖掘，对评估与排序结果进行修正。这种排序不仅可以用于检测社交网络中的异常账户，也可以作为用户判断其他账户是否可信的依据。文章主要成果如下： 1）提出基于账户特征与行为特征的社交网络账户信任度计算模型。论文在账户特征、行为特征方面提出多个能够用以区分异常账户的特征，引入粗糙集理论的属性约简方法进行特征提取，并提出一个基于数量分布的特征相似度评估方法，最后得到账户信任度的计算模型。 2）对社交网络账户间关系与账户间交互行为进行深入挖掘，，提出AccountRank算法对账户信任度进行修正，从而得到更加准确的结果。在社交网络中，被关注程度越高的账户越值得信任，与值得信任的账户交互越多的账户越值得信任。基于这个现象，本文参考著名的PageRank算法，根据社交网络中账户间关系与账户交互行为的特点进行修改后得到AccountRank算法，对所得信任度进行修正。 3）以新浪微博为实验对象，获取了大量真实的数据进行实验，以验证模型的有效性。实验结果显示，计算得到的账户信任度能够用于账户的信任排序，为用户判断账户的可信任程度提供有力依据。设定合理的阈值后，能够对异常账户进行自动检测。同时，利用账户间关系对上述结果进行修正后，相关指标都得到提升。
[Abstract]:Social network is a kind of network service rising in the era of web2.0. It extends offline social activities to online, allowing users to register their accounts and interact on the network. Social networks promote good online social behaviour, but spam is still posted through social network accounts. Due to the openness and immediacy of social networks, these spam information can spread rapidly and widely, and the negative events caused by the dissemination of spam information also show an increasingly serious trend. Therefore, it is very important to identify and restrict the abnormal account which is used to release spam information in social network. The contribution of this paper is to introduce the concept of trust into social networks, and propose a computational model to evaluate the trust of social network accounts, and then sort the accounts according to the evaluation results. At the same time, the relationship between social network accounts is deeply excavated, and the results of evaluation and ranking are revised. This sort can be used not only to detect abnormal accounts in social networks, but also to judge the credibility of other accounts. The main results of this paper are as follows: 1) A social network account trust calculation model based on account feature and behavior feature is proposed. In this paper, a number of features that can be used to distinguish abnormal accounts are proposed in terms of account features and behavioral features, and a feature similarity evaluation method based on quantitative distribution is proposed by introducing the attribute reduction method based on rough set theory for feature extraction. Finally, the calculation model of account trust is obtained. 2) the relationship between social network accounts and the interaction between accounts are deeply mined, and the AccountRank algorithm is proposed to modify the trust degree of the account, so as to get more accurate results. In social networks, accounts with a higher degree of attention are more trustworthy, and accounts that interact with trusted accounts are more trustworthy. Based on this phenomenon, this paper refers to the famous PageRank algorithm, according to the relationship between accounts and the characteristics of account interaction in social networks to modify the characteristics of the AccountRank algorithm, to modify the resulting trust. 3) taking Weibo of Sina as the experimental object, a large number of real data were obtained to verify the validity of the model. The experimental results show that the calculated trust degree can be used to sort the trust of the account and provide a powerful basis for the user to judge the degree of trust of the account. After setting a reasonable threshold, the abnormal account can be automatically detected. At the same time, using the relationship between accounts to revise the above results, the related indicators are improved.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.08

【参考文献】