Abstract: For random sampling is not representative, it proposes a robust parallel improvement of algorithms when using CURE algorithm to handle non-uniform mass data. It uses the Binary-Positive algorithm to get the effective properties of the data, uses valid data for hierarchical clustering with MapReduce, which is a distributed parallel framework. It achieves the correct rate and efficiency of a trade-off. The tests show that the improved CURE algorithm has a higher efficiency in the implementation and has a good clustering result.

Key words: clustering, Clustering Using Representative（CURE）, Binary-Positive, MapReduce, parallel

摘要： 当CURE算法在处理不均匀的海量数据时，针对随机抽样不具有代表性的问题，提出了一种健壮的并行化改进算法。该算法使用Binary-Positive算法得到原始数据的有效属性，并利用MapReduce并行框架对有效数据进行层次聚类，从而实现了正确率与效率的一种权衡。实验分析表明，改进后的CURE算法具有更高的执行效率，且聚类效果良好。

关键词: 聚类, 利用代表点聚类（CURE）, Binary-Positive, MapReduce, 并行