协同过滤推荐算法的改进研究

发布时间：2018-04-09 14:52

本文选题：协同过滤推荐算法　切入点：奇异值分解　出处：《辽宁科技大学》2016年硕士论文

【摘要】：随着互联网的普及和信息技术的飞速发展,互联网中用户的数量和信息量也急剧增长,人们想要从海量的信息中快速地获取自己感兴趣的信息已经变得越来越困难。如社交媒体、电子商务等以用户为中心的网站,信息量越来越庞大,产生了海量的与用户兴趣相关的数据,然而用户所关注的只是其中很少的一部分信息。个性化推荐技术强调从用户兴趣出发,针对不同的用户提供特定的个性化推荐服务。个性化推荐技术并不需要用户提供明确的需求,而是从用户的历史行为和数据出发,以此为依据,通过建立相关的模型挖掘出用户的兴趣和需求,从而从海量的信息中为用户筛选出其感兴趣的信息。因此,个性化推荐技术在用户需求不明确时,显得尤为重要。目前为止,许多推荐算法已经被提出,但是协同过滤推荐算法是这些推荐算法中应用最广泛且最为成功的算法之一。然而,虽然协同过滤推荐算法已经被成功地应用到许多推荐系统中,但是随着系统规模的不断扩大,系统中用户的数量和项目的数量不断增加,协同过滤推荐算法却面临着数据稀疏性问题、冷启动问题和可扩展性问题等多方面的严峻挑战。针对协同过滤推荐算法存在的这些问题,本文提出了一种组合协同过滤推荐算法。论文首先阐述了协同过滤推荐算法的基本思想,并阐述了算法中存在的问题。然后针对协同过滤推荐算法中存在的问题分别提出了基于奇异值分解和层次聚类的改进算法和改进的Slope One协同过滤推荐算法,以此来展开对协同过滤推荐算法的研究。为了验证改进算法的有效性,将两种改进算法分别在MovieLens数据集上进行对比实验。实验结果表明,这两种改进算法的推荐质量优于传统的协同过滤推荐算法。本文在最后提出将这两种改进算法相结合的组合协同过滤推荐算法。同样在MovieLens数据集上进行实验,实验结果验证了组合协同过滤推荐算法的有效性。
[Abstract]:With the popularization of the Internet and the rapid development of information technology, the number of users and the amount of information in the Internet are also increasing rapidly. It is becoming more and more difficult for people to quickly obtain the information they are interested in from the mass of information.Such as social media, electronic commerce and other user-centered websites, the amount of information is increasing, resulting in a huge amount of data related to user interest, but users only focus on a small part of the information.Personalized recommendation technology emphasizes on providing specific personalized recommendation services for different users based on users' interests.Personalized recommendation technology does not need users to provide clear requirements, but from the user's historical behavior and data, based on this, through the establishment of relevant models to mine the interests and needs of users.From the mass of information for the user to filter out the information of interest.Therefore, personalized recommendation technology is particularly important when user needs are not clear.So far, many recommendation algorithms have been proposed, but collaborative filtering recommendation algorithm is one of the most widely used and most successful recommendation algorithms.However, although collaborative filtering recommendation algorithms have been successfully applied to many recommendation systems, the number of users and the number of projects in the system are increasing with the expansion of the scale of the system.However, collaborative filtering recommendation algorithms face many challenges, such as data sparsity, cold start and scalability.Aiming at these problems of collaborative filtering recommendation algorithm, a combined collaborative filtering recommendation algorithm is proposed in this paper.In this paper, the basic idea of collaborative filtering recommendation algorithm is introduced, and the problems in the algorithm are discussed.Then, an improved algorithm based on singular value decomposition and hierarchical clustering and an improved Slope One collaborative filtering recommendation algorithm are proposed to solve the problems in collaborative filtering recommendation algorithm.In order to verify the effectiveness of the improved algorithm, the two improved algorithms are compared with each other on the MovieLens dataset.Experimental results show that the recommendation quality of the two improved algorithms is superior to that of the traditional collaborative filtering recommendation algorithm.In the end, this paper proposes a combined collaborative filtering recommendation algorithm which combines these two improved algorithms.The experimental results on MovieLens data set show the effectiveness of the combined collaborative filtering recommendation algorithm.
【学位授予单位】：辽宁科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】