个性化推荐系统中的推荐算法研究

发布时间：2018-01-27 12:41

本文关键词： 推荐系统 K-means算法加权关联规则协同过滤算法　出处：《北京交通大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着电子商务的发展,电商网站的规模在不断增大,网上商品的种类越来越多。推荐系统的出现可以帮助用户在面对如此繁多的商品信息时做出选择。但目前推荐算法依然在发展中,其精确性还不能满足人们的需求,所以有很多关于此方面的研究。如何根据用户的特定喜好进行更准确、更便捷的推荐服务成为了越来越多专家学者致力研究的内容。关联规则和协同过滤是两种重要的推荐算法。但是由于事物数据库中项目分布的不均匀性以及协同过滤算法中的数据稀疏性造成的冷启动问题,这些算法的精确性还有待提高。本文针对这些问题,对关联规则和协同过滤算法提出了改进,并将改进后的算法应用在个性化web商城推荐系统中。论文的研究工作得到了国家自然科学基金项目(No.61172072、61271308)、北京市自然科学基金项目(No.4112045)和高等学校博士学科点专项科研基金(No.20100009110002)的支持。本论文的主要研究工作包括:(1)根据Apriori算法及Partition算法的问题,改进了关联规则。针对事物数据库中具有强规律性的项目,设计了一种权重公式给予其相应加权,以此来提高事务数据库中具有稳定时间间隔规律项的支持度;(2)为了解决协同过滤算法中的数据稀疏性问题,本文根据后台数据库中的数据记录,获取其中的隐式信息,并结合MapReduce技术将其应用在个性化推荐系统中来优化稀疏矩阵;(3)为了寻找用户的邻居集合,本文提出了一种基于二进制循环指数的多重K-means协同过滤算法来发掘目标用户的最近邻居集合并为其进行商品推荐服务。将提出的算法应用在web商城推荐系统中,在个性化的推荐系统中优化根据用户个人喜好的推荐以及根据最近邻居集合预测目标用户感兴趣项的推荐模式;(4)将改进的加权关联规则和优化的协同过滤算法进行有机的结合,本文设计出了一个推荐系统原型,并通过实验对此系统的性能效果进行了分析。实验结果表明:改进后的关联规在精确度上得到了提高;采用MapReduce技术挖掘用户隐式信息可以弥补web商城单一使用用户的显示信息造成的缺陷;改进的协同过滤算法在发掘最近邻居的效率及精确度上得到了提高;融合上述两种算法的推荐系统的性能得到了优化。
[Abstract]:With the development of electronic commerce, the scale of e-commerce website is increasing. There are more and more kinds of goods on the Internet. The appearance of recommendation system can help users to make choices in the face of so many commodity information. However, the recommendation algorithm is still in the process of development. Its accuracy can not meet the needs of people, so there is a lot of research on this aspect. How to carry out more accurate according to the specific preferences of users. More and more experts and scholars devote themselves to the research of more and more convenient recommendation services. Association rules and collaborative filtering are two important recommendation algorithms. However, because of the uneven distribution of items in the transaction database and collaboration. Cold start problem caused by data sparsity in filtering algorithm. The accuracy of these algorithms still needs to be improved. Aiming at these problems, this paper proposes some improvements on association rules and collaborative filtering algorithms. And the improved algorithm is applied to the personalized web mall recommendation system. The research work of this paper has been obtained by the National Natural Science Foundation Project No. 61172072 / 61271308). Beijing Natural Science Foundation Project No. 4112045) and the Special Research Foundation for doctoral subject points in institutions of higher Learning No. 20100009110002). The main research work of this thesis includes the problem of Apriori algorithm and Partition algorithm. The association rules are improved. For the items with strong regularity in the transaction database, a weight formula is designed to give the corresponding weights, so as to improve the support degree of the items with stable time interval in the transaction database. In order to solve the problem of data sparsity in the collaborative filtering algorithm, the implicit information is obtained according to the data records in the background database. The sparse matrix is optimized by using MapReduce technology in personalized recommendation system. In order to find a set of neighbors of the user. In this paper, a binary cyclic exponent based multifold K-means collaborative filtering algorithm is proposed to discover the nearest neighbor set of the target user and to recommend goods for the target user. The proposed algorithm is applied to web. In the Mall recommendation system. In the personalized recommendation system, we optimize the recommendation mode based on the user's personal preferences and the nearest neighbor set to predict the items of interest to the target user. 4) combining the improved weighted association rules with the optimized collaborative filtering algorithm, a prototype of the recommendation system is designed in this paper. The performance of the system is analyzed by experiments. The experimental results show that the precision of the improved correlation gauge is improved. Using MapReduce technology to mine the implicit information of users can make up for the defects caused by the single use of user's display information in web Mall. The improved collaborative filtering algorithm improves the efficiency and accuracy of finding the nearest neighbor. The performance of the recommendation system which combines the above two algorithms is optimized.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】