中文摘要：

现有自底向上的角色工程方法挖掘任务繁重，挖掘规模庞大，且难以体现挖掘结果的可用性。为提取有效、完整的可用角色集，结合枚举角色挖掘的研究及用户之间的相似度属性，将角色挖掘问题转换为聚类M题，提出一种基于划分与压缩方法的改进角色挖掘。使用围绕中心点划分方法分解用户聚类，采用聚类压缩算法压缩划分用户集，利用矩阵的稠密性分析法合并重构有效矩阵，并在构造和真实数据集上进行测试与分析。实验结果表明，该方法能够降低求解问题的复杂性及挖掘规模，并能提取出有效、完整的可用角色集；与枚举法挖掘相比，压缩率控制在40％以内、支持度阈值取约0．6时，改进挖掘的效果比较理想。

英文摘要：

Mining roles for large scale datasets were heavy tasks in existing approaches to bottom-up role engineering,and re- sults of role mining were unusable for actual system deployment. In order to derive available roles, this paper proposed an im- proved role mining based on methods of partitioning and compression,which transformed the role mining into clustering mining problem. The method decomposed the clustering of users by using the partitioning around medoids algorithm, and compressed users in each partition by adopting the compression algorithm. It utilized the method of analysing density to reconstruct the new matrix, It was tested and evaluated on several datasets, both synthetic and real. Experimental results show that it reduces the complexity and the scale, and it can discover the set of roles that is effective and complete. Also, the method performs well if it appropriately choose the compression ratio and the threshold.