基于最佳信任路径的协同过滤推荐算法的研究与设计

发布时间：2018-01-24 08:35

本文关键词： 最佳路径路径信任兴趣相似度协同过滤预测评分　出处：《广东技术师范学院》2017年硕士论文　论文类型：学位论文

【摘要】：不断发展的信息技术在带给人们丰富网络资源的同时也使人们陷入信息过载的困境,如何帮助用户在海量数据中快速找到有价值的相关信息,是推荐技术研究的核心问题。协同过滤推荐算法是在众多推荐技术中使用最为广泛,但随着在线社会网络的不断发展和社会网络日益复杂,用户间信任关系在一定程度上影响着推荐结果。传统的协同过滤推荐方法主要存在以下问题:(1)数据稀疏性。用户的评分数据稀疏,使计算的相似度不准确,导致评分预测不准确,用户无法得到合适的推荐信息。(2)易受攻击性。开放的推荐系统允许用户自由发布评分或评论,可能有些用户提供虚假信息,导致推荐结果产生严重偏差,无法向用户提供满意推荐。(3)没有考虑信任关系。传统算法只是考虑了用户的评分数据,未考虑用户间可能存在的信任关系以及这种信任关系对推荐系统的价值。针对协同过滤推荐算法中存在的上述问题与挑战,本文提出了一种改进的协同过滤算法方案,主要研究工作包括:(1)针对协同过滤推荐系统的数据稀疏问题,采用计算用户间兴趣相似度作为判断用户间相似度的一个依据,区别于传统算法只考虑用户对项目的评分。(2)针对已有算法的易受攻击问题,在推荐过程中,综合考虑用户间信任关系与兴趣相似度以计算用户间综合相似度,从而缓解因用户虚假评分导致的推荐结果不准确问题。(3)提出一种基于最佳信任路径的协同过滤推荐算法。用最佳信任路径代替原算法的多路径取平均值方法,在充分考虑信任路径中其他用户威望值的基础上,选择多条信任路径中的最佳信任路径,改善了原算法只考虑最终用户威望值进而缺乏客观性的问题。实验结果表明,与基于用户的协同过滤推荐算法和融合信任的协同过滤推荐算法相比,本文算法具有以下优势:(1)推荐准确度更高;(2)运行效率更高,本文算法运行时间是融合信任推荐算法运行时间的四分之一,当信任路径增加时,本文算法的优势更加明显。
[Abstract]:The continuous development of information technology not only brings people rich network resources, but also makes people into the plight of information overload, how to help users quickly find valuable relevant information in the massive data. Collaborative filtering recommendation algorithm is the most widely used recommendation technology, but with the continuous development of online social network and the increasing complexity of social network. Trust relationship between users affects the recommendation results to some extent. The traditional collaborative filtering recommendation methods mainly have the following problems: 1) data sparsity. Because the similarity of calculation is not accurate, the score prediction is not accurate, and the user can not get the appropriate recommendation information. The open recommendation system allows the user to publish the rating or comment freely. It is possible that some users provide false information, which leads to serious deviation of recommendation results, which can not provide satisfactory recommendation to users. Trust relationship is not considered. The traditional algorithm only takes into account the users' rating data. This paper does not take into account the possible trust relationship between users and the value of this trust relationship to the recommendation system. Aiming at the above problems and challenges in collaborative filtering recommendation algorithm. In this paper, an improved collaborative filtering algorithm is proposed. The main research work includes: 1) data sparsity in collaborative filtering recommendation system. The interest similarity between users is calculated as a basis for judging the similarity between users, which is different from the traditional algorithm, which only considers the user's score on the items.) it is aimed at the vulnerability of the existing algorithms. In the process of recommendation, the trust relationship and interest similarity between users are considered comprehensively to calculate the comprehensive similarity between users. This alleviates the problem of inaccurate recommendation results caused by users' false ratings. This paper presents a collaborative filtering recommendation algorithm based on the best trust path, and uses the best trust path instead of the original algorithm to average the multi-path. On the basis of fully considering the other user prestige values in the trust path, we choose the best trust path in multiple trust paths. The original algorithm only considers the value of end-user prestige and lacks objectivity. The experimental results show that compared with the user-based collaborative filtering recommendation algorithm and the fusion trust collaborative filtering recommendation algorithm. This algorithm has the following advantages: 1) recommendation accuracy is higher; The running time of the proposed algorithm is 1/4 of that of the fusion trust recommendation algorithm. When the trust path is increased, the advantages of the proposed algorithm are more obvious.
【学位授予单位】：广东技术师范学院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】