基于内容和协同过滤的混合算法在推荐系统中的应用研究

发布时间：2018-03-20 03:14

本文选题：推荐系统　切入点：协同过滤算法　出处：《东华大学》2016年硕士论文　论文类型：学位论文

【摘要】：在电子商务为越来越多用户提供便利的同时,它的结构也越发复杂。用户拥有了更大的选择空间,然而海量信息的同时呈现,也让用户无法快速准确地寻找到自己想要的商品。推荐系统就是为解决这样的困境产生的,它能为用户个性化定制,产生用户希望看到的结果。如今,几乎所有的电子商务网站都应用了推荐系统,例如淘宝、ePay等。为了能产生更高的推荐质量,技术人员已经提出了多种推荐技术,包括基于内容的推荐,协同过滤推荐,关联挖掘推荐等。然而随着研究的深入以及用户要求的提升,推荐算法渐渐暴露了诸多缺点,例如协同过滤推荐的冷启动和数据稀疏性问题、推荐质量和实时性的平衡等。本文主要对基于内容和协同过滤算法进行研究,具体内容如下:(1)对推荐算法的发展情况及关键技术进行综述;(2)针对协同过滤算法运行时间长、推荐速度慢的缺陷,提出了将k-means聚类应用于协同过滤算法,通过离线预处理用户—项目评分矩阵对用户进行聚类,并生成虚拟用户空间,替代全部用户的集合,缩小了邻居用户的搜寻范围,可以提高在线推荐的实时性,解决了传统协同过滤算法的速度瓶颈问题;(3)在改进后协同过滤算法的基础上,提出了基于内容和改进后协同过滤的混合算法,将用户评分和项目特征相结合,建立用户特征评分矩阵,替代传统的用户—项目评分矩阵,对用户集进行k-means聚类,并进行推荐。改进后的算法不仅能解决传统协同过滤算法的数据稀疏性问题;同时针对新项目,也能根据项目特征与用户特征评分矩阵的匹配,来预测可能对新项目感兴趣的用户,并生成推送列表,有效解决“冷启动”中的新项目问题。实验证明,本文改进的算法在解决数据稀疏、冷启动和在线推荐的速度瓶颈问题方面,有显著的作用,能保证较好的推荐质量。
[Abstract]:While e-commerce provides convenience for more and more users, its structure is becoming more and more complicated. It also makes it impossible for users to find the product they want quickly and accurately. Recommendation systems are created to solve this dilemma. It can be customized for the user and produce the results that the user wants to see. Today, Almost all e-commerce websites use recommendation systems, such as Taobao ePay. In order to generate higher recommendation quality, technicians have proposed a variety of recommendation technologies, including content-based recommendations, collaborative filtering recommendations, etc. However, with the deepening of research and the improvement of user requirements, the recommendation algorithm gradually exposes many shortcomings, such as cold start of collaborative filtering recommendation and data sparsity. This paper mainly studies the content-based and collaborative filtering algorithms, which are as follows: (1) A review of the development and key technologies of the recommendation algorithms; (2) the long running time of the collaborative filtering algorithms. In this paper, we apply k-means clustering to collaborative filtering algorithm, and cluster users by off-line preprocessing user-item scoring matrix, and generate virtual user space to replace the set of all users. It reduces the search range of neighbor users, improves the real-time performance of online recommendation, and solves the speed bottleneck problem of traditional collaborative filtering algorithm based on the improved collaborative filtering algorithm. A hybrid algorithm based on content and improved collaborative filtering is proposed, which combines user score with item feature, establishes user feature scoring matrix, replaces the traditional user-item scoring matrix, and carries out k-means clustering on user set. The improved algorithm can not only solve the data sparsity problem of the traditional collaborative filtering algorithm, but also match the scoring matrix according to the item features and user characteristics. To predict users who might be interested in new projects, and generate push lists to effectively solve the problem of new projects in "cold start". Experiments show that the improved algorithm in this paper is to solve the problem of sparse data. The speed bottleneck of cold start and online recommendation plays a significant role in ensuring better recommendation quality.
【学位授予单位】：东华大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】