改进相似性度量方法的协同过滤推荐算法

COLLABORATIVE FILTERING RECOMMENDATION ALGORITHM ON IMPROVED SIMILARITY MEASURE METHOD

协同过滤推荐技术是电子商务推荐系统中应用最成功的个性化推荐技术。但随着电子商务规模的扩大,用户数目和商品数目呈指数级的增长,传统的推荐技术其性能越来越差。因此提出一种新的相似性度量方法,自动生成权重因子,以动态组合项目属性相似度和评分相似度,形成合理的项目相似度,产生项目最近邻居,实现用户评分推荐。实验结果表明,所提的算法在一定程度上提高了推荐的稳定性和精确度,同时解决冷启动问题。

Collaborative filtering recommendation technology is the most successful personalised recommendation technology ever applied to e-commerce recommendation systems.As the scale of e-commerce expands,the magnitudes of users and commodities grow rapidly,which persistently worsens the performance of traditional recommendation technology.Therefore a new similarity measure method is put forward to automatically generate weighting factors,dynamically combine attribute similarity and score similarity,create a reasonable item similarity to find out the nearest neighbouring item,and finally realise user rating recommendation.Experimental results prove the algorithm improves recommendation steadiness and precision to a certain extent and solves the cold start problem.