一种基于信任传播和奇异值分解的个性化推荐方法的研究

发布时间：2018-04-03 07:54

本文选题：推荐系统　切入点：社会化推荐　出处：《昆明理工大学》2017年硕士论文

【摘要】：随着大数据时代的到来,互联网中充斥着大量纷繁的数据,如何快速、高效的发现用户需要的信息已经成为人们关心的热点之一。推荐系统是基于信息过滤技术向用户推荐感兴趣信息的推荐工具,推荐系统的出现有效的缓解了目前存在的信息过载问题。然而传统推荐系统仍然存在以下挑战:第一,大多数的传统推荐算法假设用户之间是相互独立的,这种假设忽略了用户之间的社交关系对用户决策的影响,这与我们现实世界中的人际关系是不相符合的。第二,虽然现在已经有一些推荐算法为了考虑社交网络中用户之间的关系对用户决策的影响也开始把用户之间的社交关系作为推荐系统的影响因素。然而用户社交关系的稀疏性导致大多数的数据集中只挖掘到了极少数的用户之间的关系数据。本文针对上述推荐算法所面临的问题和挑战提出了一种推荐算法,该算法融合了信任传播和奇异值分解模型。首先,本文针对传统推荐算法忽略了用户之间的社交关系的问题,提出了一种信任传播算法,该算法通过具有直接关联关系的用户作为传输信任的纽带推导出没有直接关联关系的用户之间的信任度,从而填充信任关系矩阵,缓解了用户信任关系矩阵稀疏的难题。其次,由于用户评分矩阵的稀疏性导致了推荐系统的推荐质量降低,奇异值分解模型可以将数据映射到低维空间,然后计算低维空间中的item之间的相似度,对用户未评分的item进行评分预测,最后将预测评分高的item推荐给用户,提高了推荐系统的推荐质量。本文中同时结合信任传播规则和奇异值分解模型进行推荐将有利于提高推荐系统的推荐质量。最后,为了验证本文所提出的算法的有效性我们通过在两个公开网站Epinions.com和,Filmtrust.com网站上面爬取的数据集与传统推荐算法进行实验对比,实验结果表明本文提出的方法取得了较好的效果。
[Abstract]:With the arrival of big data era, the Internet is full of a lot of data, how to quickly and efficiently find the information users need has become one of the hot spots.Recommendation system is a recommendation tool based on information filtering technology to recommend interested information to users. The appearance of recommendation system effectively alleviates the existing problem of information overload.However, the following challenges still exist in traditional recommendation systems: first, most of the traditional recommendation algorithms assume that users are independent of each other, and this assumption ignores the impact of social relationships between users on users' decisions.This is incompatible with our real-world relationships.Second, although there are some recommendation algorithms to consider the relationship between users in social networks on the impact of user decision-making, also began to consider the social relationship between users as a recommendation system factors.However, the sparsity of user social relationships leads to the mining of only a few relational data between users in most datasets.In this paper, we propose a recommendation algorithm, which combines trust propagation and singular value decomposition model, in view of the problems and challenges faced by the above recommendation algorithms.Firstly, aiming at the problem that traditional recommendation algorithms ignore the social relationship between users, a trust propagation algorithm is proposed in this paper.The trust degree between users without direct relationship is derived by using the user with direct association relationship as the link of transmission trust, thus filling the trust relationship matrix, which alleviates the problem of sparse user trust relationship matrix.Secondly, due to the sparsity of the user score matrix, the recommendation quality of the recommendation system is reduced. The singular value decomposition model can map the data to the low-dimensional space, and then calculate the similarity between the item in the low-dimensional space.In the end, the item with high prediction score is recommended to the user, which improves the recommendation quality of the recommendation system.In this paper, combining trust propagation rules and singular value decomposition model for recommendation will be helpful to improve the recommendation quality of recommendation system.Finally, in order to verify the effectiveness of the proposed algorithm, we compare the data set crawled on two public websites, Epinions.com and Filmtrust.com, with the traditional recommendation algorithm.The experimental results show that the proposed method has achieved good results.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】