Abstract: For personalized recommendation，either content-based method or collaborative filtering based method，the similarity between two items or users is the basic operation.In mass public data set，experiments show those functions，which implement SVM in content-based method and introduce discriminability in collaborative filtering，can improve the accuracy of recommendations.Besides，the modified MapReduce flows also get more effective similarity computation.

Key words: personalized recommendation, distributed computation, Support Vector Machine（SVM）, discriminability

摘要： 相似度计算在个性化推荐系统中是基本运算，但无论是基于内容还是基于协同过滤的推荐，目前常用的向量相似度计算还存在可以改进之处。在海量公开的数据集上的实验表明，在基于内容的推荐中引入机器学习方法以及在协同过滤推荐中引入区分度来改善相似度计算，可以获取更高的准确率。对MapReduce的分布式计算流程的改进，使得相似度计算更为高效。

关键词: 个性化推荐, 分布式计算, 支持向量机, 区分度