基于商品关系改进的协同过滤推荐算法

发布时间：2018-01-28 13:20

本文关键词： 推荐系统协同过滤隐式商品关系显式商品关系关联规则商品类别矩阵分解　出处：《燕山大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着互联网中的信息量剧增,用户个性化需求日趋急切,推荐系统已经成为信息过滤的热门工具。协同过滤推荐算法是应用最广泛的推荐算法。为进一步提升推荐结果的准确度,大量基于用户关系的协同过滤算法被提出,例如基于社交网络用户关系的推荐算法等。然而,被众多的研究者忽略的商品关系对于提高推荐算法的准确度却有更好的效果。本文旨在将商品关系融合到传统的协同过滤推荐算法中以进一步提升推荐效果。通常商品关系可分为隐式关系和显式关系。首先,为解决利用相似度算法发掘隐式商品关系方法中存在的不足,例如对称性以及不能同时考虑多个商品间的关系等问题,本文利用改进的关联规则技术挖掘一对一及多对一的隐式商品关系,并以此关系作为正则项融合到矩阵分解模型中。同时,为进一步研究不同商品关系对提升推荐效果的影响,本文设计了四种不同的选取隐式商品关系的策略。其次,在真实的电子商务系统中,商品之间往往存在显式的关系:具有相似特征的商品会被分配到同一类别中,反之则被分配到不同类别中。同时考虑到两种不同的情形:一件商品可能只属于一个类别即一对一商品类别关系或是同时属于多个类别即一对多商品类别关系,本文提出了一个新颖的基于矩阵分解的商品推荐模型。其中,有别于传统的矩阵分解方法,我们加入类别信息来更准确地描述用户和商品潜在特征向量,并将显式商品关系用作正则项加以限制商品特征向量的学习过程。最后,为了验证提出算法的合理性及有效性,本文在四个真实世界的数据集上做了严密且充分的对比实验。
[Abstract]:With the rapid increase in the amount of information in the Internet, the personalized needs of users are becoming increasingly urgent. Recommendation system has become a popular tool for information filtering. Collaborative filtering recommendation algorithm is the most widely used recommendation algorithm. In order to further improve the accuracy of recommendation results. A large number of collaborative filtering algorithms based on user relationships have been proposed, such as recommendation algorithms based on user relationships in social networks. The commodity relationship neglected by many researchers has a better effect on improving the accuracy of recommendation algorithm. This paper aims to integrate commodity relationship into traditional collaborative filtering recommendation algorithm to further improve the recommendation effect. The constant commodity relation can be divided into implicit relation and explicit relation. In order to solve the problems of using similarity algorithm to explore the implicit commodity relations, such as symmetry and cannot consider the relationship between multiple commodities at the same time, and so on. In this paper, the improved association rule technique is used to mine the one-to-one and many-to-one implicit commodity relations, and the relations are fused into the matrix decomposition model as regular terms. In order to further study the impact of different commodity relationships on the promotion of recommendations, this paper designs four different strategies to select implicit commodity relationships. Secondly, in the real e-commerce system. There is often an explicit relationship between goods: goods with similar characteristics are assigned to the same category. On the contrary, they are assigned to different categories. Two different situations are considered: a commodity may belong to only one category, that is, a one-to-one commodity category relationship, or multiple categories, that is, a one-to-many commodity category relationship at the same time. In this paper, a novel commodity recommendation model based on matrix decomposition is proposed. Different from the traditional matrix decomposition method, we add class information to describe the potential feature vectors of users and commodities more accurately. The explicit commodity relationship is used as a regular term to limit the learning process of the commodity feature vector. Finally, in order to verify the rationality and effectiveness of the proposed algorithm. In this paper, four real world data sets are closely and fully contrasted with each other.
【学位授予单位】：燕山大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】