【摘要】 提出并行Slope One算法PSL,并展示了它的Map-Reduce计算模型实现。PSL有效地克服了(1)原Slope One算法空间复杂度过高,单机无法存储其中间文件的缺点;(2)计算代价高,通常只能使用在小规模数据上的不足。并且此算法:(1)能够进行增量计算,实现准实时的用户推荐,(2)可以运行在成本低廉的集群上,不仅具有学术意义,更具商业价值。采用了Net-flix Prize真实数据进行实验,结果表明,PSL可以应用在真实的大规模数据之上,并且具有良好的容错性和伸缩性。更多还原

【Abstract】 PSL,a parallel Slope One algorithm,is proposed and presented our implementation on Map-Reduce computing model.PSL effectively overcomes(1) High space complexity of the original Slope One algorithm,where intermediate files can’t be stored on a single machine;(2) High computing overhead,usually only used in small-scale dataset.Meanwhile,this algorithm(1) can do incremental computing,achieve a quasi-real-time user recommendation,(2) can run on low-end clusters,which means it’s not only academic significant,but also commercial valuable.We use Netflix Prize,a real-world dataset,for our experiments.Results show that,PSL can be applied to real-world,large-scale data applications,and achieves a good fault-tolerance and scalability.更多还原