基于用户协同过滤推荐算法的研究

发布时间：2018-03-01 21:25

本文关键词： 协同过滤稀疏性扩展性冷启动奇异值分解聚类　出处：《天津大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着“互联网+”的蓬勃发展和大数据时代的来临,网上购物已经成为人们日常生活的重要组成部分。如何快速有效的解决数据的搜索、分类、推荐,成为摆在人们面前亟需解决的问题。推荐系统能够根据用户的个人偏好,向用户提供个性化的服务推荐,从而使得广大用户能够迅速且准确地找到自己满意的商品或服务。尽管推荐系统近来取得了长足进步,但随着互联网用户数量的急剧增长,以及移动互联网的快速发展,推荐系统存在的缺陷和不足,更加制约着推荐系统的健康发展。本文针对基于用户协同过滤推荐系统所面临的稀疏性、扩展性、冷启动问题,通过将奇异值分解与聚类融合,并针对基于用户协同过滤推荐系统的特性,改进了相似性度量公式与Top-N推荐权值赋值系数,提出一种融合奇异值分解和聚类的协同过滤推荐算法—SCW(SVD Clustering Weight)算法。SCW算法主要融合了奇异值分解降维、聚类和Top-N推荐三个原理。具体研究内容可概括为以下几个方面:首先,本文利用奇异值分解从Movielens数据集中构建一个主题空间,然后在该空间中计算相似度。其次,对降维后的评分矩阵中的用户进行聚类,聚类分析将数据划分成有意义的簇。最后,完成聚类后,按照用户聚类的每个簇都是行为模式相似的用户,遍历目标用户所在簇的其他用户的物品集,去除目标用户物品集生成推荐列表。然后利用改进的预测评分公式计算推荐列表物品的预测评分,排序推荐。本文利用Moivelens公开数据集进行实验验证,并将本文的提出算法与传统的基于皮尔森相关度协同过滤算法和主成分分析算法进行对比。实验结果表明,本文提出的算法能够有效完成个性化推荐,提高了推荐准确率并降低了预测评分值的绝对值误差,在一定程度上解决了基于用户协同过滤推荐系统中存在的稀疏性、扩展性与冷启动问题。
[Abstract]:With the vigorous development of "Internet" and the advent of big data era, online shopping has become an important part of people's daily life. How to quickly and effectively solve the data search, classification, recommendation, Recommendation system can provide personalized service recommendation to users according to their personal preferences. As a result, the vast number of users can quickly and accurately find the goods or services they are satisfied with. Although the recommendation system has made great progress recently, with the rapid growth of the number of Internet users and the rapid development of the mobile Internet, The shortcomings and shortcomings of recommendation system restrict the healthy development of recommendation system. This paper aims at the problems of sparseness, expansibility and cold start of recommendation system based on user collaborative filtering, by combining singular value decomposition and clustering. According to the characteristics of user collaborative filtering recommendation system, the similarity measurement formula and the assignment coefficient of Top-N recommendation weight are improved. This paper proposes a collaborative filtering recommendation algorithm, SCWN SVD Clustering weight, which combines singular value decomposition and clustering. SCW algorithm mainly integrates three principles of singular value decomposition, clustering and Top-N recommendation. The specific research contents can be summarized as follows: firstly, In this paper, we use singular value decomposition to construct a topic space from the Movielens dataset, and then calculate the similarity in this space. Secondly, the users in the reduced dimension scoring matrix are clustered, and the data is divided into meaningful clusters by clustering analysis. After clustering, according to the user clustering each cluster is a user with similar behavior patterns, traversing the items set of other users of the target user cluster. The recommended list is generated by removing the target user item set, and then the prediction score of the recommended list item is calculated by the improved prediction scoring formula, and the recommendation is sorted. In this paper, Moivelens is used to expose the data set for experimental verification. The proposed algorithm is compared with the traditional Pearson correlation collaborative filtering algorithm and the principal component analysis algorithm. The experimental results show that the proposed algorithm can effectively complete personalized recommendation. The accuracy of recommendation is improved and the absolute error of prediction score is reduced. To some extent, the problems of sparseness, expansibility and cold start of recommendation system based on user collaborative filtering are solved.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】