社交网络中基于关系强度的用户群体发现研究

发布时间：2018-06-08 14:46

本文选题：社交网络 + 用户社会关系　；参考：《东华大学》2015年硕士论文

【摘要】：随着互联网的飞速发展，各式各样的社交网络不停地涌现出来。作为一种新颖便捷的交友模式，社交网络吸引了大量的用户。越来越多的用户通过社交网络收集的各类资源信息来发表看法、交友等，国外知名社交网站Facebook每月活跃人数已经达到11亿人，国内社交网络代表新浪微博的用户数已经突破五亿。面对日益增长的庞大数据，无论用户还是社交网络的服务商都迫切需要解决一个问题：即如何寻找出与自己兴趣爱好或者看法一致的人进行交流互动。用户群体发现研究正是基于此目的而产生的，其目标是通过对社交网络中的用户关系图进行挖掘，从中发现具有相似兴趣的用户群体，进而支持广告投放、市场营销、好友推荐等实际应用。传统的用户群体发现方法是基于社交网络中用户之间的原始关系图，将用户视为图中的顶点，用户间的关系作为图的边，通过对图进行聚类分析从而获得用户的群体聚簇。这些传统的方法未考虑到用户关系的稀疏性，以及用户关系在社交网络与现实网络中的差异。本文在发现用户群体的过程中，一方面既考虑了用户在各个主题上相似信息的总体分布，另一方面也考虑了主题热门程度的差异对用户关系的影响。结合以上两个方面，本文给出了用户关系强度的计算模型，通过该计算模型针对社交网络的特点扩充了用户关系，最后使用聚类分析实现用户的群体发现。本文的具体工作内容主要包括： 1)首先介绍了相关技术，包括社交网络的相关理论基础、用户关系强度的计算方法，，以及MapReduce编程模型与局部敏感哈希的基本思想。 2)接着阐述了一种通过构建用户特征同现向量，计算用户关系强度的方法。该方法结合了多样性指数以及权重频率，从两个相互独立的角度，共同计算了用户间的关系强度。 3)面对社交网络的数据量挑战，将上述的计算过程通过MapReduce编程模型得以实现，并在关系强度的计算结果基础上，利用局部敏感哈希和MapReduce的特性实现了新的用户关系图上的用户群体发现。 4)使用社交网站Last.fm所开放的端口获取的数据进行实验，并对模型的相关参数进行了估算。实验结果从性能分析和可靠性分析上，证明了用户关系强度计算及群体发现的可行性与实用性。
[Abstract]:With the rapid development of the Internet, a variety of social networks are emerging. As a new and convenient way to make friends, social networks attract a large number of users. More and more users are expressing their opinions and making friends through various resources collected by social networks. The number of people active on Facebook, a well-known foreign social network, has reached 1.1 billion a month. The number of users representing Sina Weibo on domestic social networks has exceeded 500 million. In the face of the growing volume of data, both users and social network service providers urgently need to solve a problem: how to find out how to interact with people who share their interests or views. The research of user group discovery is based on this purpose. Its goal is to find user groups with similar interests through mining user relationship diagrams in social networks, and then support advertising and marketing. The traditional method of user group discovery is based on the original graph of users in social network. The user is regarded as the vertex of the graph, and the relationship between users is regarded as the edge of the graph. The cluster of users is obtained by cluster analysis of graph. These traditional methods do not take into account the sparsity of user relationships and the differences between user relationships in social networks and real networks. In the process of discovering user groups, on the one hand, we consider the general distribution of users' similar information on each topic, on the other hand, we also consider the influence of the difference of topic popularity on user relationship. Combined with the above two aspects, this paper presents a computing model of user relationship strength, which extends the user relationship according to the characteristics of social network. Finally, cluster analysis is used to realize user group discovery. The main contents of this paper are as follows: 1) this paper first introduces the relevant technologies, including the relevant theoretical basis of social networks, the calculation method of user relationship intensity, And the basic idea of MapReduce programming model and local sensitive hashing. 2) then a method to calculate the strength of user relationship by constructing user feature co-occurrence vector is presented. This method combines diversity index and weight frequency, calculates the relationship strength between users from two independent angles. 3) facing the challenge of social network data, the above calculation process can be realized by MapReduce programming model. On the basis of the calculation results of the relationship strength, the new user group discovery on the user relationship diagram is realized by using the characteristics of local sensitive hashing and MapReduce. 4) the data obtained by the open port of Last.fm is used to carry out experiments. The related parameters of the model are estimated. The experimental results prove the feasibility and practicability of user relationship strength calculation and group discovery from performance analysis and reliability analysis.
【学位授予单位】：东华大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP391.1

【参考文献】