Computer Engineering and Applications ›› 2012, Vol. 48 ›› Issue (21): 21-25.

Previous Articles     Next Articles

Research on personal hybrid recommendation overcoming data sparse problem

JIANG Wei1, PANG Xiuli2,1   

  1. 1.School of Management, Harbin Institute of Technology, Harbin 150001, China
    2.School of Economic and Business Management, Heilongjiang University, Harbin 150080, China
  • Online:2012-07-21 Published:2014-05-19

面向数据稀疏问题的个性化组合推荐研究

姜  维1,庞秀丽2,1   

  1. 1.哈尔滨工业大学 管理学院,哈尔滨 150001
    2.黑龙江大学 经济与工商管理学院,哈尔滨 150080

Abstract: Collaborative filtering is one of the typical personal recommendations, but some difficulties exist, such as the data sparse problem, cold starting problem, scaling expanding problem. Two respects of work are done: on one hand, an improved similarity measure is presented to overcome the data sparse problem, on the other hand, the SVD based special user erasing method is presented to overcome the noise-sample problem. The experiments show that the improved method increases 4.30% in terms of MAE, and the combined model, which adopts the weighted average method, further increases 1.26%.

Key words: personal recommendation, collaborative filtering, data sparse problem, combined recommendation

摘要: 协同过滤技术是推荐系统中应用最为广泛的算法,其面临着数据稀疏性问题、冷启动、规模可扩展性等问题。工作体现在两点:一是在基于项的协同过滤模型中,改进了项间的相似度计算方法,相比调整余弦方法仅考虑一个要素,包含了三个要素:两项的具有共同用户的评分、共同评分用户数量、非共同评分用户数量;二是组合基于用户、基于项和基于奇异值分解的协同过滤推荐,通过多模型组合提高推荐性能。实验结果表明在基于项过滤中MAE指标上提高了4.30%。进一步,加权的组合多种模型方法比基于项方法提高了1.26%。

关键词: 个性化推荐, 协同过滤, 数据稀疏问题, 组合推荐