基于广度优先遍历的不信任网络扩展推荐算法

发布时间：2018-03-08 02:18

本文选题：不信任　切入点：信任　出处：《吉林大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着网络技术的不断发展人们已经从一个信息十分匮乏的时代过渡到信息量巨大的信息海洋世界里,面对如此庞大的信息数据,人们想要从中快速的获取自己想要的信息是十分困难的事情。由此推荐系统应运而生,能够提供给人们个性化的网络服务,使得人们在浏览网页的时候能快速精准的定位自己需要的信息。推荐系统最早应用在电子商务领域,在用户购买物品的时候能够得到一个良好的推荐,提升了客户体验度。但由于推荐系统的数据稀疏性和冷启动等问题的困扰使得一些用户并不能得到很好的推荐。由此一种基于社交网络的推荐系统诞生了。这种算法将社交网络中的信任关系考虑在推荐系统当中,具体的结合方式有多种。比如可以用信任关系筛选信任用户形成信任用户-项目评分矩阵,计算相似度筛选最邻近;而有一些是把相似度与信任值相结合形成一个新的综合值来筛选最邻近等。无论以上哪种方式都对推荐系统的效率和准确率有了明显的改善和提高,同时也使得长尾效应的影响变得更加小了。但是我们也知道,社交网络是一个十分复杂的社会关系网,其中不仅仅有信任关系,还有许多不信任关系等因素,因此在推荐算法中将不信任因素考虑进去也是理所应当的事情。虽然现有阶段已有一些国外的学者讨论了这方面的问题,但都不够全面,没有给出一个完整的不信任模型,并且没有很好的解决不信任关系网络的数据稀疏性问题。为了解决上述问题,使得能有一个更好的个性化推荐算法服务于用户,更好的解决数据稀疏性问题和考虑方面过于单一化问题。本文的主要工作如下:提出了一个全新的基于广度优先遍历的不信任关系传递算法。本文提出了不信任的传递算法,用以拓展社交网络中不信任关系网络。为了实现以上提出的基于广度优先遍历的不信任关系传递算法,本文设计了不信任关系传递模型,使得不信任关系得以量化。本文将基于广度优先遍历的不信任关系传递算法与基于信任的协同过滤算法相结合,产生了一个全新的基于广度优先遍历的不信任网络扩展推荐算法。该算法显著的解决了社交网络中不信任矩阵稀疏性的问题,同时也兼顾缓解了推荐系统稀疏性的问题,在总体上提高了推荐系统的准确性和效率。以Epinions为数据集进行实验验证,并与传统的协同过滤推荐算法进行分析对比,证实了新算法在推荐效果上要优于传统的协同过滤推荐算法。
[Abstract]:With the continuous development of network technology, people have been transitioning from an era of very scarce information to a world of information ocean with huge amount of information, faced with such huge information data. It is very difficult for people to get the information they want quickly from it. So that people can quickly and accurately locate the information they need when they browse the web. Recommendation system was first applied in the field of electronic commerce, and it can get a good recommendation when users buy items. But because of the problems of data sparsity and cold startup of recommendation system, some users are not able to get good recommendation. Thus, a recommendation system based on social network was born. The algorithm considers the trust relationship in the social network as a recommendation system. For example, trust relationship can be used to filter trust users to form trust user-item scoring matrix, and to calculate similarity filtering nearest to each other. Some of them combine similarity and trust to form a new comprehensive value to select the nearest neighbor. Either way, the efficiency and accuracy of the recommendation system have been improved and improved. But we also know that social networks are a very complex social network, which has not only trust relationships, but also many factors such as distrust. Therefore, it is natural to take the distrust factor into account in the recommendation algorithm. Although some foreign scholars have discussed this problem at the present stage, it is not comprehensive enough to give a complete model of distrust. In order to solve the above problem, there is a better personalized recommendation algorithm to serve users. The main work of this paper is as follows: a new algorithm based on breadth-first traversal for the transfer of distrust relationship is proposed. In order to realize the above proposed algorithm based on breadth-first traversal, this paper designs a model of distrust relation transfer. In this paper, we combine the breadth-first traversal of the distrust transfer algorithm with the trust based collaborative filtering algorithm. A new extended recommendation algorithm based on breadth-first traversing is presented. The algorithm solves the problem of the sparsity of distrust matrix in social networks and also alleviates the sparse problem of recommendation system. On the whole, the accuracy and efficiency of the recommendation system are improved. The experimental results are verified by using Epinions as the data set, and compared with the traditional collaborative filtering recommendation algorithm. It is proved that the new algorithm is superior to the traditional collaborative filtering recommendation algorithm in recommendation effect.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【相似文献】