基于微博信息扩散的链接预测研究

发布时间：2019-05-20 09:55

【摘要】：在微博媒体中每天都活跃着数以亿计的用户，这些用户每天都能够产生海量的信息，面临如此巨大的信息流，很多用户都无法快速高效的找到自己感兴趣的信息。在微博媒体中，用户主要通过关注其他用户来获取信息，因此，为了获取定制的自己感兴趣的信息流，用户会通过判断其他用户发布的信息是否是自己感兴趣的来决定是否关注此人。微博媒体上的链接预测问题，从根本上说是通过为用户推荐好友的方式来给用户提供他感兴趣的定制信息服务。本文希望通过研究微博网络中信息扩散特征结合网络结构特征来进行链接预测，提出了基于微博信息扩散的节点相似性链接预测算法、基于信息扩散的分类链接预测算法以及基于信息扩散的随机游走链接预测算法。基于微博信息扩散的节点相似性链接预测算法是传统社交网络基于节点相似性推荐方法的改进方案，通过微博媒体网络中信息扩散特征结合共同关注、共同粉丝等网络结构特征属性来计算节点相似性，选择相似性高的节点推荐给目标用户。基于信息扩散的分类器链接预测算法希望通过对用户信息扩散特征以及结构特征的分析进行分类，把分好类的推荐用户作为集合推荐给目标用户。基于信息扩散的随机游走算法通过计算概率图模型中目标节点随机游走到任意节点的概率，根据概率值大小进行排序得到推荐列表。从实验结果可以看出，，有监督机器学习的分类算法效果最好，随后依次是节点相似性算法和随机游走算法，同时，基于信息扩散特征的链接预测算法都能获得优于传统方法的效果，证明了信息扩散特征在社交网络链接预测中的重要意义。
[Abstract]:There are hundreds of millions of users active in Weibo media every day. These users can generate a lot of information every day. Facing such a huge information flow, many users can not find the information they are interested in quickly and efficiently. In Weibo media, users get information mainly by focusing on other users, so in order to obtain custom information flow of interest to themselves, Users decide whether to pay attention to other users by judging whether they are interested in publishing information. The link prediction problem on Weibo media is fundamentally by recommending friends to users to provide users with customized information services of interest to them. In this paper, we hope to study the characteristics of information diffusion in Weibo network combined with the characteristics of network structure to predict the link, and propose a node similarity link prediction algorithm based on Weibo information diffusion. Classification link prediction algorithm based on information diffusion and random walk link prediction algorithm based on information diffusion. The node similarity link prediction algorithm based on Weibo information diffusion is an improved scheme based on node similarity recommendation method in traditional social networks, which combines the information diffusion characteristics in Weibo media network with common concern. The network structure characteristic attributes such as common fans are used to calculate the node similarity, and the nodes with high similarity are selected and recommended to the target users. The link prediction algorithm of classifiers based on information diffusion hopes to classify the information diffusion characteristics and structural features of users and recommend the recommended users as a set to the target users. The random walk algorithm based on information diffusion calculates the probability that the target node walks randomly to any node in the probability graph model, and sorts it according to the probability value to get the recommended list. From the experimental results, it can be seen that the classification algorithm with supervised machine learning is the best, followed by node similarity algorithm and random walk algorithm. At the same time, The link prediction algorithm based on information diffusion feature can obtain better effect than the traditional method, which proves the significance of information diffusion feature in social network link prediction.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.092

【共引文献】