基于AP聚类算法的推荐系统研究

发布时间：2018-05-22 15:18

本文选题：推荐系统 + 聚类　；参考：《河北大学》2017年硕士论文

【摘要】：进入21世纪以来,在互联网和电子商务网站飞速发展的背景下,电子商务网站中的信息量变得更为庞大和复杂,繁冗的数据给电子商务的发展带来巨大的挑战。为了解决这一难题,针对电子商务的推荐系统应运而生,电子商务推荐系统主要目的是帮助用户迅速的定位到自己喜欢的商品。在目前主流的各种推荐算法中,协同过滤算法是一种应用较广的推荐算法,但传统的协同过滤存在“稀疏性”、“冷启动”和“可扩展性”等问题。近年来关于推荐系统的研究中,一些学者提出将聚类技术引入到推荐系统中用以解决上述问题。基于聚类的推荐算法通过先对用户或者项目进行聚类划分,使得相似度较高的对象聚集到同一个类中,从而简化查找最近邻居的过程,大大减小了整体计算复杂度和时间消耗。另外由于聚类过程可以在线下完成,所以大大提升了推荐系统整体的实时性。本文提出了基于AP聚类的推荐算法,主要研究内容如下:(1)提出并设计了基于AP聚类的推荐算法。将AP聚类算法引入到推荐系统的用户分类过程中,仅需要将目标用户通过AP聚类方法进行分类,简化查找最近邻居和计算对象相似度的过程,降低了在整体计算中的复杂度和时间消耗。(2)传统的AP聚类不包括类别的合并过程,使得聚类的精度较差,尤其是对结构复杂的数据。本文提出了一种基于属性加权的度量方法,基于此对AP聚类算法进行了改进。(3)设计并实现了基于改进AP聚类的推荐算法。在公共数据集上进行了仿真实验,评价指标为平均绝对偏差(MAE)值,芮氏指标(RI)和纯度指标(Purity),实验结果表明了本文算法的有效性。
[Abstract]:Since the beginning of the 21st century, with the rapid development of the Internet and e-commerce websites, the amount of information in e-commerce websites has become larger and more complex, and the redundant data has brought great challenges to the development of e-commerce. In order to solve this problem, E-commerce recommendation system emerges as the times require. The main purpose of E-commerce recommendation system is to help users locate their favorite products quickly. Collaborative filtering is one of the most popular recommendation algorithms, but there are some problems in the traditional collaborative filtering, such as "sparsity", "cold start" and "expansibility". In recent years, in the research of recommendation system, some scholars have proposed to introduce clustering technology to the recommendation system to solve the above problems. The recommendation algorithm based on clustering makes objects with high similarity gather into the same class by clustering users or items, thus simplifying the process of finding nearest neighbors and greatly reducing the overall computational complexity and time consumption. In addition, the clustering process can be completed off-line, so it greatly improves the real-time performance of the recommendation system as a whole. This paper proposes a recommendation algorithm based on AP clustering. The main research contents are as follows: 1) A recommendation algorithm based on AP clustering is proposed and designed. The AP clustering algorithm is introduced into the user classification process of recommendation system. It is only necessary to classify the target users by AP clustering method to simplify the process of finding nearest neighbor and computing object similarity. It reduces the complexity and time consumption in the whole computation.) the traditional AP clustering does not include the merging process of the categories, which makes the accuracy of the clustering worse, especially for the data with complex structure. In this paper, an attribute weighted measurement method is proposed. Based on this, an improved AP clustering algorithm is designed and a recommendation algorithm based on improved AP clustering is implemented. Simulation experiments are carried out on common data sets. The evaluation indexes are mean absolute deviation (mae), Rui's index and purity index. The experimental results show the effectiveness of the proposed algorithm.
【学位授予单位】：河北大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】