基于关联规则和神经网络分析的推荐系统的研究

发布时间：2018-05-25 16:13

本文选题：推荐系统 + 关联规则　；参考：《杭州电子科技大学》2017年硕士论文

【摘要】：随着电商的迅速发展,推荐系统已经成为各大电商网站的标配。推荐系统不但给使用者提供了个性化的推荐服务,同时增加了使用者对该网站的忠诚度。推荐系统已经成为电商网站之间竞争的重要手段,所以推荐系统逐渐成为大型电子商务公司的核心研究部分。推荐系统的最重要部分是推荐算法,关联规则与机器学习结合是当前最流行的推荐系统研究方式之一。本文的主要工作如下:1、详细分析了传统的推荐算法中的协同过滤的工作流程,分别举例论证说明基于用户的协同过滤与基于物品的协同过滤的优劣。同时分析了几种常见的推荐算法的适用场景,比如奇异值分解(SVD)分析的推荐、基于隐语义分析的推荐、基于图分析的推荐。2、分析研究了关联规则中的传统算法Apriori算法和FP-tree算法的优缺点,继承FP-tree算法只扫描两次数据集的前提上进行正、负项目的关联规则挖掘。并通过实验实例证明该算法不仅在时间耗时上有了显著的提高,同时可以自动删除互相矛盾的关联规则。该算法是对负关联规则在FP-tree算法的一种增强扩充。3、为了结合用户的偏好模型,通过机器学习中的BP网络进行融合,详细说明的用户偏好模型建立过程,同时通过BP神经网络进行学习得出的目标用户的偏好模型。利用标准的电影评分评测集Movieslens作为实验的数据集,分别从算法的有效性和鲁棒性这两个方面进行验证。实验效果证明BP神经网络结合用户偏好模型不仅可以提高算法的有效性,而且还可以保证了算法的稳定性。上述研究已在科研实践中取得良好效果,可泛化到各个领域中的推荐系统。同时为了将来更多的优秀的机器学习算法加入推荐系统的研究打下了理论的基础。
[Abstract]:With the rapid development of e-commerce, recommendation system has become the standard of e-commerce websites. The recommendation system not only provides users with personalized recommendation service, but also increases their loyalty to the website. Recommendation system has become an important means of competition among e-commerce websites, so recommendation system has gradually become the core research part of large e-commerce companies. Recommendation algorithm is the most important part of recommendation system. The combination of association rules and machine learning is one of the most popular research methods of recommendation system. The main work of this paper is as follows: 1. The workflow of collaborative filtering in traditional recommendation algorithm is analyzed in detail, and the advantages and disadvantages of user-based collaborative filtering and object-based collaborative filtering are illustrated by examples. At the same time, this paper analyzes the applicable scenarios of several common recommendation algorithms, such as the recommendation of singular value decomposition (SVD) analysis, the recommendation based on implicit semantic analysis, Based on the recommendation of graph analysis, the advantages and disadvantages of Apriori algorithm and FP-tree algorithm in association rules are analyzed and studied. The association rules mining of positive and negative items is carried out on the premise that FP-tree algorithm only scans the data sets twice. The experimental results show that the algorithm not only improves the time consuming, but also automatically removes the contradictory association rules. This algorithm is an extension of negative association rules in FP-tree algorithm. In order to combine user preference model with BP network in machine learning, the process of establishing user preference model is explained in detail. At the same time, the preference model of target user is obtained by BP neural network. The standard film score evaluation set (Movieslens) is used as the experimental data set to verify the effectiveness and robustness of the algorithm. Experimental results show that BP neural network combined with user preference model can not only improve the effectiveness of the algorithm, but also ensure the stability of the algorithm. These studies have achieved good results in scientific research practice and can be generalized to recommendation systems in various fields. At the same time, it lays a theoretical foundation for more excellent machine learning algorithms to join recommendation system in the future.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】