基于Hadoop的电子商务个性化推荐研究与实现

发布时间：2018-05-24 03:49

本文选题：云计算 + 电子商务　；参考：《兰州交通大学》2017年硕士论文

【摘要】：随着互联网的迅猛发展,人类社会进入了云计算时代,云计算时代个性化和差异化已经成为企业的核心竞争力,电子商务的兴起实现了从“以产品为中心”向“以客户为中心”的消费模式的转变,极大地提高了用户和企业之间的交易效率。然而电子商务在为用户提供快捷便利的服务时,用户和商品数量正以指数级别递增,由此引发信息超载和信息迷航,为使用户在互联网中能更便捷地获取所需信息,电子商务个性化推荐系统由此产生,它根据用户的兴趣偏好及购物习惯向其推荐感兴趣的商品。电子商务网站使用个性化推荐系统分析用户的兴趣偏好,帮助用户从庞大的商品目录中挑选真正适合自己需要的商品,为每个用户提供个性化服务,在很多电子商务个性化推荐技术中,作为当今运用最广泛、最成功的推荐策略,协同过滤推荐算法对电子商务个性化推荐系统性能的优劣至关重要。但是传统的协同过滤推荐算法存在稀疏性高、可扩展性和实时性差等问题,导致电子商务个性化推荐系统无法为用户提供高效的推荐服务。因此,针对协同过滤推荐算法中存在的稀疏性、可扩展性和实时性等问题,本文设计了一种基于Hadoop的电子商务个性化推荐算法,利用Hadoop云计算平台所拥有的强大计算及存储能力对基于项目的协同过滤推荐算法予以改进。首先根据用户的评分使用Canopy算法进行聚类预处理,将项目集合中的全部项目迅速划分到不同的Canopy中并生成相应的聚类中心。接着按照Canopy聚类中心集合使用FCM算法对项目集合中的每个项目迭代求出其相对于各个聚类的隶属度和聚类中心,并根据最大隶属度原则确定每个项目所属的聚类,从而得到最终的聚类集合。然后在Canopy-FCM聚类模型上根据目标项目与所有聚类中心之间的相似度,选择相似度最高的前若干个聚类中的项目构建目标项目的推荐候选项目空间。最后在推荐候选项目空间上使用IBCF算法在线完成推荐。在MovieLens数据集上与一些传统的协同过滤算法进行的分析和对比的实验结果表明,该算法在分布式集群上具有较好的可扩展性和推荐效率,并且推荐精度也有所提高,可以较好地改进电子商务个性化推荐中存在的稀疏性、可扩展性和实时性问题。
[Abstract]:With the rapid development of the Internet, the human society has entered the cloud computing era, the cloud computing era of individuation and differentiation has become the core competitiveness of enterprises, The rise of electronic commerce has realized the transformation from "product-centered" to "customer-centered" consumption mode, and greatly improved the transaction efficiency between users and enterprises. However, when electronic commerce provides users with fast and convenient services, the number of users and commodities is increasing exponentially, which leads to information overload and information confusion, so that users can obtain the information they need more easily on the Internet. E-commerce personalized recommendation system comes into being, according to the user's interest preference and shopping habits to recommend the interested goods to them. E-commerce websites use personalized recommendation system to analyze the interests of users, help users choose from the huge catalog of products that are really suitable for their own needs, and provide personalized services for each user. In many e-commerce personalized recommendation technology, as the most widely used and most successful recommendation strategy, collaborative filtering recommendation algorithm is very important to the performance of e-commerce personalized recommendation system. However, the traditional collaborative filtering recommendation algorithm has many problems, such as high sparsity, poor scalability and real-time performance, which leads to the lack of efficient recommendation service for users in E-commerce personalized recommendation system. Therefore, aiming at the problems of sparsity, extensibility and real-time in collaborative filtering recommendation algorithm, a personalized recommendation algorithm for e-commerce based on Hadoop is designed in this paper. Based on the powerful computing and storage capability of Hadoop cloud computing platform, the project based collaborative filtering recommendation algorithm is improved. Firstly, the Canopy algorithm is used to preprocess the clustering according to the user's score. All the items in the item set are quickly divided into different Canopy and the corresponding clustering centers are generated. Then, according to the Canopy clustering center set, each item in the item set is iterated by FCM algorithm to find out its membership degree and cluster center relative to each cluster, and according to the principle of maximum membership degree, each item belongs to the cluster. Finally, the final clustering set is obtained. Then, according to the similarity between the target items and all the clustering centers on the Canopy-FCM clustering model, the items in the first few clusters with the highest similarity are selected to construct the recommended candidate space of the target items. Finally, the IBCF algorithm is used to complete the recommendation online in the recommended candidate space. The results of analysis and comparison with some traditional collaborative filtering algorithms on MovieLens dataset show that the proposed algorithm has better scalability and recommendation efficiency in distributed clusters, and the recommendation accuracy is also improved. It can improve the sparse, extensibility and real-time problems in e-commerce personalized recommendation.
【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】