【作者】 孙静；

【导师】 李杰；

【作者基本信息】 河北工业大学 ， 管理科学与工程， 2011， 硕士

【摘要】 随着Internet在人们的日常生活中占据着越来越重要的位置,电子商务也随之迅速发展起来,为此推荐系统也得以广泛的在各种电子商务系统里使用。它是根据用户曾经的购买记录,找出记录中存在的潜在的用户喜好信息,然后在此基础上向用户推荐符合其兴趣的商品。目前,在各个领域中都能够看到推荐系统的影子,如淘宝、易趣网、HC360慧聪网、中华商务网等许多大型的电子商务网站,都或多或少的地使用了各种各样的推荐系统。协同过滤推荐技术,是目前个性化推荐技术中,学者研究最多、应用最广、推荐的个性化程度高的一种推荐技术。本文主要是对协同过滤推荐算法进行研究,分析了现有算法中存在的问题,特别对推荐算法中存在的稀疏性问题等推荐问题进行了剖析,结合模糊聚类技术提出一种基于模糊聚类的电子商务协同过滤推荐改进算法。论文主要内容包括以下三个方面:第一,分析了国内外电子商务推荐系统的研究现状和未来的研究方向;简单介绍了电子商务推荐系统的基于内容和基于关联规则的推荐算法;对协同过滤推荐算法进行了详细的阐述,介绍了传统的协同过滤推荐算法的基本步骤,分析了协同过滤推荐算法的优点,同时指出了其中存在的一些现实问题。第二,针对传统协同过滤算法的缺点,结合FCM聚类技术,提出了一种基于模糊聚类的协同过滤推荐算法。其基本思路是:首先将FCM聚类技术引进到用户-项目评分模型中,针对协同过滤推荐算法中存在的问题,对FCM进行改进。然后,用改进的FCM进行聚类,得到待测用户的邻居用户,根据邻居用户算得预测值,进行推荐。第三,使用MovieLens数据库,对本文提出的基于FCM改进的协同过滤推荐算法进行实验仿真与测试,将结果与基于K-means算法和传统的算法进行对比,通过实验显示改进后的算法比另外两种算法具有更好的推荐质量。更多还原

【Abstract】 With the internet occupy more and more important position in people’s daily life.Electronic commerce system also begin to expand rapidly along with it,Therefore the recommendation system can also apply in the electronic commerce system widespread.The recommendation system is the system that it can use the users’purchase history,according to their former customs recommend information or the commodity to the users which suit to them.At the present,the recommendation system has been utilized in every profession, such as Amazon、CDNOW、eBay、and Dang dang on-line bookstore etc,have used various recommend -ation system in the different degree.As the most successful personalized recommendation technology of recommendation system in present,collaborative filtering recommendation is obtaining more and more attention of researchers.This article basied on researching collaborative filtering recommendation algorithm analyzes the question which still existed in the present research,its emphasis is analyse the cold start question,the sparse question,the new project question,the new user question,and so on user multi-interests question existing in the system.Then,the article proposes an improvem -ent algorithm which based on fuzzy clustering to research collaborative foltering recommendation in electronic commerce. The article’s concrete work contains the following several aspects:First, The article has analyzed the domestic and foreign electronic commerce recommendat ion system’s present research situation and future development direction.It also has introduced the electronic commerce recommendation system simply which based on the content and based on the connection rule recommendation algorithm;It elaborated collaborative filtering recommendation algorithm in detail,introduced its basic steps,analyzed its merit,at the same time,pointed out some realistic questions which exists in algorithm. Second, the article specificed to the collaborative foltering recommendation algorithm’s shortcoming,and unified the FCM cluster technology,it came up with an algorithm which is collaborative filtering recommendation in electronic commerce based on fuzzy clustering.Its basic mentality is: First, the article bring FCM clustering technology into user-project grading model,according to the questions which exists in the collaborative filtering recommendation algorithm,made the improvement to FCM.Then, the article carried on the cluster with improvement FCM to the data set,that it can obtain the testing user’s neighbor users,according to the neighbor users calculated the testing user’s predicted value,at last,it gave testing user the recom-mendation.Third,using the MovieLens database,the artlcle carried on experiment simulation which and test to improvement algorithm proposed,compared with the based on k-means algorithm and traditional collaborative filtering recommendation algorithm,the experiment demonstrated that the algorithm has the high recommendation quality compared to the traditional algorithm.更多还原