基于数据摘要的流式子模优化算法研究

Research of streaming submodular optimization algorithm based on data summarization

针对如何从数据中提取出公平摘要的问题,文中采用流式子模最大化方法来解决该问题,并对其算法进行改进,提出一种公平约束下的流式子模最大化算法。该算法根据摘要的个数以及数据属性范围设置上下界构成公平约束,能够确保提取出代表性子集涵盖原始数据集的所有属性范围。仿真结果表明,该文算法与其他流式子模最大化算法相比,不仅时间复杂度减少了8.6%以上,而且在不同数据集下都能保证数据摘要结果的公平性与多样性。

Aiming at the problem of how to extract fair summarization from data,this paper adopts the streaming submodular maximization method to solve it, improves the algorithm, and proposes a streaming submodular maximization algorithm with fairness constraint. By setting the upper and lower bounds according to the number of summarization and the range of data attributes,the algorithm can guarantee that the representative subset covers all the range of attributes of the original data set. Through simulation experiments,it is concluded thatcompared with other streaming submodular maximization algorithms,the proposed algorithm not only reduces the time complexity by more than 8.6%,but also ensures the fairness and diversity of data summarization results under different data sets.