社会网络中的微博用户推荐算法研究
发布时间：2018-10-17 07:19
【摘要】：随着微博平台中用户的爆炸式增长，其用户创造的信息也随之呈指数级增长。从而导致过量的数据使得用户无法有效地获取自己想要的信息，即信息的使用率反而降低，信息过载的问题则日益加剧。目前的搜索引擎等技术只能满足人们部分的需求，没有个性化的考虑，仍无法有效地解决这个问题。用户推荐作为一种信息过滤手段，是解决这个问题非常有潜力的方法。因而如何发展高效的，可扩展的，非常精确的用户推荐算法是一个巨大的挑战。 本文根据目前流行的微博平台的特性提出了两种用户推荐算法，，一种是基于领域偏好度的名人推荐算法，另一种是基于社区信息传播力的用户推荐算法。基于领域偏好度的名人推荐算法将用户推荐问题转化为一个基于链接预测的分类问题，它基于名人用户所属的领域来围绕目标用户和被推荐名人用户提取一系列的特征并以此构建一个n维的特征向量，再利用分类器过滤有限的名人集合而得到该用户的名人推荐集合。基于社区信息传播力的用户推荐算法则是基于社区划分的思想，即将兴趣相似的用户聚到一个社区，通过分析该社区的消息流动情况，来挖掘社区中对消息传播具有控制能力的消息中间人，同时结合目标用户自身的特点从消息中间人中选取合适的用户推荐给他。另一方面，为了解决当前海量数据处理的问题，本文针对两种推荐算法还提出基于Map-Reduce的并行化实现方法。 通过在微博平台数据集上的实现与测试，验证了两种推荐算法的可行性及有效性。根据推荐算法的一般评估方法，本文提出的两种推荐算法与其它常用的推荐算法相比，效果均有所提高。同时基于Map-Reduce的并行化实现，算法性能明显高于其单机环境。
[Abstract]:With the explosive growth of users in Weibo platform, the information created by its users has also increased exponentially. As a result, excessive data makes users unable to obtain the information they want effectively, that is, the utilization rate of information is reduced, and the problem of information overload is aggravated day by day. The current search engine and other technologies can only meet the needs of some people, without personalized consideration, still can not effectively solve this problem. As a kind of information filtering method, user recommendation is a potential method to solve this problem. Therefore, how to develop efficient, extensible, very accurate user recommendation algorithm is a huge challenge. According to the characteristics of Weibo platform, this paper puts forward two kinds of user recommendation algorithms, one is celebrity recommendation algorithm based on domain preference, the other is user recommendation algorithm based on community information transmission ability. The celebrity recommendation algorithm based on domain preference degree transforms the user recommendation problem into a classification problem based on link prediction. It extracts a series of features around the target user and the recommended celebrity user based on the domain to which the celebrity user belongs and constructs an n-dimensional feature vector. Then the classifier is used to filter the limited celebrity set to get the user's celebrity recommendation set. The user recommendation algorithm based on the ability of community information dissemination is based on the idea of community division, which brings users with similar interests to a community, and analyzes the information flow in that community. In order to mine the message middleman who has the ability to control the message propagation in the community, at the same time, combine the target user's own characteristic, select the appropriate user from the message intermediary to recommend to him. On the other hand, in order to solve the problem of mass data processing, this paper proposes a parallel implementation method based on Map-Reduce for two recommended algorithms. Through the implementation and test on Weibo platform data set, the feasibility and effectiveness of the two recommended algorithms are verified. According to the general evaluation method of the recommendation algorithm, the effect of the two recommendation algorithms proposed in this paper is improved compared with other commonly used recommendation algorithms. At the same time, the parallel implementation based on Map-Reduce shows that the performance of the algorithm is obviously better than that of its single computer environment.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.3;TP393.092
本文编号：2275940
[Abstract]:With the explosive growth of users in Weibo platform, the information created by its users has also increased exponentially. As a result, excessive data makes users unable to obtain the information they want effectively, that is, the utilization rate of information is reduced, and the problem of information overload is aggravated day by day. The current search engine and other technologies can only meet the needs of some people, without personalized consideration, still can not effectively solve this problem. As a kind of information filtering method, user recommendation is a potential method to solve this problem. Therefore, how to develop efficient, extensible, very accurate user recommendation algorithm is a huge challenge. According to the characteristics of Weibo platform, this paper puts forward two kinds of user recommendation algorithms, one is celebrity recommendation algorithm based on domain preference, the other is user recommendation algorithm based on community information transmission ability. The celebrity recommendation algorithm based on domain preference degree transforms the user recommendation problem into a classification problem based on link prediction. It extracts a series of features around the target user and the recommended celebrity user based on the domain to which the celebrity user belongs and constructs an n-dimensional feature vector. Then the classifier is used to filter the limited celebrity set to get the user's celebrity recommendation set. The user recommendation algorithm based on the ability of community information dissemination is based on the idea of community division, which brings users with similar interests to a community, and analyzes the information flow in that community. In order to mine the message middleman who has the ability to control the message propagation in the community, at the same time, combine the target user's own characteristic, select the appropriate user from the message intermediary to recommend to him. On the other hand, in order to solve the problem of mass data processing, this paper proposes a parallel implementation method based on Map-Reduce for two recommended algorithms. Through the implementation and test on Weibo platform data set, the feasibility and effectiveness of the two recommended algorithms are verified. According to the general evaluation method of the recommendation algorithm, the effect of the two recommendation algorithms proposed in this paper is improved compared with other commonly used recommendation algorithms. At the same time, the parallel implementation based on Map-Reduce shows that the performance of the algorithm is obviously better than that of its single computer environment.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.3;TP393.092
【参考文献】
相关期刊论文 前6条
1 郭新涛,梁敏,阮备军,朱扬勇;挖掘Web日志降低信息搜寻的时间费用[J];计算机研究与发展;2004年10期
2 张娥,郑斐峰,冯耕中;Web日志数据挖掘的数据预处理方法研究[J];计算机应用研究;2004年02期
3 黄晓斌;;社会书签与网络信息推荐服务[J];情报理论与实践;2006年01期
4 田莹颖;;基于社会化标签系统的个性化信息推荐探讨[J];图书情报工作;2010年01期
5 董倩;王克俭;韩宪忠;苑迎春;;基于贝叶斯分类的网上书店潜在用户挖掘[J];微型机与应用;2011年01期
6 胡昌平;胡吉明;邓胜利;;基于社会化群体作用的信息聚合服务[J];中国图书馆学报;2010年03期
本文编号：2275940
本文链接：https://www.wllwen.com/kejilunwen/sousuoyinqinglunwen/2275940.html
