ISSN 1000-1239 CN 11-1777/TP

• 论文 • 上一篇    下一篇


李 聪1 梁昌勇1 马 丽2   

  1. 1(合肥工业大学管理学院 合肥 230009) 2(西华师范大学商学院 四川南充 637002) (
  • 出版日期: 2008-09-15

A Collaborative Filtering Recommendation Algorithm Based on Domain Nearest Neighbor

Li Cong1, Liang Changyong1, and Ma Li2   

  1. 1( School of Management, Hefei University of Technology, Hefei 230009) 2( Business College, China West Normal University, Nanchong, Sichuan 637002)
  • Online: 2008-09-15

摘要: 协同过滤是目前电子商务推荐系统中广泛应用的最成功的推荐技术,但面临严峻的用户评分数据稀疏性和推荐实时性挑战. 针对上述问题,提出了基于领域最近邻的协同过滤推荐算法,以用户评分项并集作为用户相似性计算基础,将并集中的非目标用户区分为无推荐能力和有推荐能力两种类型;对于前一类用户不再计算用户相似性以改善推荐实时性,对于后一类用户则提出“领域最近邻”方法对并集中的未评分项进行评分预测,从而降低数据稀疏性和提高最近邻寻找准确性. 实验结果表明,该算法能有效提高推荐质量.

关键词: 协同过滤, 推荐算法, 领域最近邻, 用户相似性, 平均绝对误差

Abstract: Currently E-commerce recommender systems are being used as an important business tool by an increasing number of E-commerce websites to help their customers find products to purchase. Collaborative filtering is the most successful and widely used recommendation technology in E-commerce recommender systems. However, traditional collaborative filtering algorithm faces severe challenge of sparse user ratings and real-time recommendation. To solve the problems, a collaborative filtering recommendation algorithm based on domain nearest neighbor is proposed. The union of user rating items is used as the basis of similarity computing among users, and the non-target users are differentiated into two types that without recommending ability and with recommending ability. To the former users, user similarity will not be computed for improving real-time performance; to the latter users, “domain nearest neighbor” method is proposed and used to predict missing values in the union of user rating items when the users have common intersections of rating item classes with target user, and then the needed items space for missing values predicting can be reduced to the few common intersections. Thus the sparsity can be decreased and the accuracy of searching nearest neighbor can be improved. The experimental results show that the new algorithm can efficiently improve recommendation quality.

Key words: collaborative filtering, recommendation algorithm, domain nearest neighbor, user similarity, MAE