基于用户兴趣的MC协同过滤推荐方法研究

发布时间：2018-04-06 05:16

本文选题：移动电子商务　切入点：协同过滤推荐　出处：《哈尔滨理工大学》2015年硕士论文

【摘要】：伴随着无线网络的大面积覆盖、电子商务的迅猛发展以及智能手机的快速普及,使移动电子商务(M-Commerce,MC)在全球范围内迅速成长,其个性化推荐系统成为提高商家竞争力、满足用户个性化需求的工具。但移动电子商务的特殊性使传统的推荐系统难以简单地移植并满足“数字宇宙”时代的特殊需求。移动电子商务是一种情境依赖性较高的移动商务服务,当用户在商品个性化推荐时用户特征、项目特征和用户的当前情境特征都会使用户兴趣产生一定程度的影响。为了满足移动电子商务的个性化需求,提高推荐的质量,本文在考虑移动电子商务用户兴趣影响因素的基础上,构建了以用户特征维度、项目特征维度和情境特征维度为变量的三维度用户兴趣模型,并根据协同过滤推荐的过程,设计了基于用户兴趣的MC协同过滤推荐方法。该方法首先运用加权的Slope One方法对移动电子商务用户兴趣模型评分数据进行填充,以解决在协同过滤推荐过程中评分数据的稀疏性问题。其次,通过基于萤火虫改进的K-means聚类方法,对移动电子商务用户进行聚类,提高了聚类中心选取的准确性,减少目标用户最近邻居的搜索空间。最后,通过引入多维相似性理论,改进了协同过滤相似性计算方法,使用户之间的相似性度量和推荐更加全面、更加精准,为移动电子商务用户提供更个性化和更准确的推荐结果。
[Abstract]:With the wide coverage of wireless network, the rapid development of e-commerce and the rapid popularity of smart phones, mobile e-commerce M-Commerce MCS) has grown rapidly in the world, and its personalized recommendation system has become to improve the competitiveness of businesses.A tool to meet the individual needs of a user.However, the particularity of mobile electronic commerce makes it difficult for the traditional recommendation system to simply transplant and meet the special needs of the "digital universe" era.Mobile E-commerce is a highly context-dependent mobile commerce service. User characteristics, project features and users' current situational features will have a certain degree of impact on user interest when users recommend products in a personalized manner.In order to meet the personalized demand of mobile electronic commerce and improve the quality of recommendation, this paper constructs a user characteristic dimension on the basis of considering the influencing factors of user interest in mobile e-commerce.According to the process of collaborative filtering recommendation, a MC collaborative filtering recommendation method based on user interest is designed.Firstly, the weighted Slope One method is used to populate the scoring data of mobile e-commerce users' interest model to solve the problem of sparse rating data in collaborative filtering recommendation process.Secondly, the improved K-means clustering method based on firefly is used to cluster the mobile e-commerce users, which improves the accuracy of the clustering center selection and reduces the search space of the nearest neighbors of the target users.Finally, by introducing multi-dimensional similarity theory, the method of collaborative filtering similarity calculation is improved, which makes the similarity measurement and recommendation between users more comprehensive and accurate.Provide more personalized and accurate recommendation results for mobile e-commerce users.
【学位授予单位】：哈尔滨理工大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：F724.6;F274;F724.6

【参考文献】