基于信任的推荐方法及应用研究

发布时间：2018-08-12 12:26

【摘要】：在过去近十年的时间里,许多研究人员为解决传统推荐系统所面临的问题以及提高推荐质量,将信任加入到推荐当中,研究如何利用信任改进推荐算法。这类研究被称为基于信任的推荐算法(系统)研究,并被证实能够很好的解决传统推荐系统所面临的问题。基于信任的推荐是一种社会化推荐,因为这类推荐方法利用来自社会化网络的信任信息。本文从研究如何利用信任实现推荐的问题出发,针对具体的应用背景,研究应该采用怎样的信息作为信任信息,研究如何将抽象的信任概念具体化和定量化(信任模型、信任计算方法),研究如何将定量的信任加入到推荐过程当中。为了利用信任提高推荐质量,本文针对一般性的电子商务推荐问题提出了一种新的利用信任网络特性的基于信任的推荐方法。信任网络是由社区中的节点和节点间因信任关系连接的边所组成的网络。对信任网络的结构特性研究以及利用这些特性建立信任模型的探索还很少。首先,本文提出一种新的利用隐性信任信息构建信任网络的方法。然后,分析所建立的信任网络的结构特性,说明这个网络是一个动态的小世界网络。最后,根据这些特性给出信任计算方法,提出了一个新的基于信任的推荐方法,并用实验证明该方法在准确度上的表现优于经典的协同过滤方法。在微博这样一个基于用户关系的平台上,用户间的交互是非常重要的。微博的用户推荐帮助用户找到他们可能感兴趣的用户微博。木文将上述这种基于信任的推荐方法应用于微博用户推荐问题当中,提出了基于信任的微博用户推荐方法。首先,提出一种利用微博中合适的信息构建用户评分的方法,并提出用户信任的计算方法(信任模型)。分析微博中可获得的信息,选择那些能够反映用户对用户信任或用户对用户感兴趣和认同的信息。利用这些信息构建用户对用户的评分以及用户对用户的信任。然后,在此基础上,给出了两种利用信任的微博用户推荐方法。一种是对FoF用户推荐方法的直接改进,一种是在协同过滤的思想上利用信任进行用户推荐。最后,通过数值实验,对提出的基于信任的微博用户推荐方法的有效性和推荐质量进行了评价。
[Abstract]:In the past ten years, in order to solve the problems faced by traditional recommendation systems and improve the quality of recommendation, many researchers have added trust to recommendation, and studied how to use trust to improve recommendation algorithms. This kind of research is called trust based recommendation algorithm (system), and has been proved to be a good solution to the problems faced by traditional recommendation systems. Trust-based recommendation is a social recommendation because it utilizes trust information from social networks. Based on the research of how to use trust to realize recommendation, this paper studies what information should be used as trust information and how to concretize and quantify the abstract trust concept (trust model). How to add quantitative trust to the recommendation process. In order to improve the quality of recommendation by using trust, this paper proposes a new recommendation method based on trust to solve the general problem of E-commerce recommendation. A trust network is a network composed of nodes in the community and the edges connected by a trust relationship. There are few researches on the structural characteristics of trust networks and the establishment of trust models using these characteristics. First of all, this paper proposes a new method to construct trust network using implicit trust information. Then, the structural characteristics of the established trust network are analyzed, and it is shown that the network is a dynamic small-world network. Finally, according to these characteristics, a new trust based recommendation method is proposed, and the experimental results show that the proposed method is superior to the classical collaborative filtering method in accuracy. In Weibo, a user-based platform, user interaction is very important. Weibo users recommend helping users find users who may be interested in Weibo. In this paper, the trust-based recommendation method is applied to the Weibo user recommendation problem, and a trust-based Weibo user recommendation method is proposed. First of all, a new method is proposed to construct a user's score using the appropriate information in Weibo, and a trust model is proposed to calculate the user's trust. This paper analyzes the information available in Weibo, and selects the information that can reflect the user's trust in the user or the user's interest in and approval of the user. This information is used to construct the user's rating and user's trust in the user. Then, two Weibo user recommendation methods based on trust are presented. One is the direct improvement of the FoF user recommendation method, the other is the use of trust in collaborative filtering. Finally, the effectiveness and quality of the proposed Weibo user recommendation method based on trust are evaluated by numerical experiments.
【学位授予单位】：大连理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.3;TP393.092

【参考文献】