协同过滤推荐系统中稀疏性数据的算法研究

发布时间：2019-02-20 08:12

【摘要】：伴随网络的高速发展和计算机技术的日新月异,电子商务中出现了信息过载的现象。为了尽快地处理类似情况,推荐系统随之诞生,此中以协同过滤算法为基础的系统更是得到了具大的肯定。但还是存在一些不易解决的问题,其中影响最大的就是数据过于稀疏的问题。本文主要针对协同过滤推荐系统中稀疏性数据的现象,提出优化传统算法的方案。首先,深入研究传统协同过滤推荐算法在稀疏性数据的场景下,相似度计算存在的优缺点,对比并且对其修正,得到基于改进计算方法的协同过滤算法。然后,根据稀疏数据的特征对用户—项目矩阵利用改进PCA减维。将通过建立可扩展聚类方法,对原始数据利用优化的K-Means算法进行有效聚类,得到针对稀疏性数据的基于降维和聚类的协同过滤算法,并将在电影推荐场景中使用优化后的算法。主要研究内容包括:(1)深入研究传统计算方法在评分数据异常稀疏的场景下,无法准确的表达用户间相似程度的问题,并在此基础上提出相应的改进方案。针对传统的相似度计算方法普遍没有考虑用户之间的共同评分项数量的问题,以及考虑到每个用户的评价标准是不一致的问题,对相似度计算进行改进。得到优化的协同算法,能更加全面考虑传统计算方法存在的缺陷。(2)研究处理稀疏数据的有效方案。分析改进主成分分析PCA,并将其用来对过于稀疏的评分矩阵进行减维处理,有效较低数据的稀疏度。研究可扩展性的改进聚类方法,对初始数据集利用优化的K-Means算法对其进行有效聚集,将具有高相似性的用户划分到相同的类中,能更好的构造用户邻居集,从而减少稀疏数据对算法的影响,同时改善传统协同过滤算法在选取邻居时计算过于复杂度的问题。最后利用电影数据集movielens对本文的优化算法进行验证,仿真表明在电影推荐场景下,确实能改善稀疏性问题。
[Abstract]:With the rapid development of network and the rapid change of computer technology, the phenomenon of information overload appears in electronic commerce. In order to deal with the similar situation as soon as possible, the recommendation system was born, and the system based on collaborative filtering algorithm was recognized. However, there are still some problems that are difficult to solve, among which the most important problem is that the data is too sparse. Aiming at the phenomenon of sparse data in collaborative filtering recommendation system, this paper proposes a scheme to optimize the traditional algorithm. Firstly, the advantages and disadvantages of the traditional collaborative filtering recommendation algorithm in sparse data scene are studied, compared and modified, and a collaborative filtering algorithm based on the improved algorithm is obtained. Then, the improved PCA is used to reduce the dimension of the user-item matrix according to the characteristics of sparse data. An extensible clustering method is established to effectively cluster the raw data using the optimized K-Means algorithm, and a collaborative filtering algorithm based on reduced and clustering for sparse data is obtained. The optimized algorithm will be used in the movie recommendation scene. The main research contents are as follows: (1) the traditional computing method can not accurately express the similarity between users in the scene where the scoring data are very sparse, and on the basis of this, the corresponding improvement scheme is put forward. In order to solve the problem that the traditional similarity calculation method does not consider the number of common score items among users, and considering that the evaluation criteria of each user are inconsistent, the similarity calculation is improved. The optimized cooperative algorithm can take into account the shortcomings of traditional computing methods more comprehensively. (2) an effective scheme to deal with sparse data is studied. The improved principal component analysis (PCA,) is analyzed and used to reduce the dimension of the score matrix which is too sparse to reduce the data sparsity effectively. The improved clustering method of extensibility is studied. The initial data set is effectively aggregated by the optimized K-Means algorithm, and the users with high similarity can be divided into the same classes, which can better construct the user neighbor set. In order to reduce the influence of sparse data on the algorithm and improve the complexity of traditional collaborative filtering algorithm in selecting neighbors. Finally, the optimization algorithm of this paper is validated by movielens, and the simulation results show that the sparse problem can be improved in the movie recommendation scenario.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】