Efficient Crowdsourced Pareto-Optimal Queries Over Partial Orders With Quality Guarantee
Efficient Crowdsourced Pareto-Optimal Queries Over Partial Orders With Quality Guarantee
复制标题
具有质量保证的部分订单的高效众包帕累托最优查询
DOI：
10.1109/tetc.2020.3017198
复制
发表时间：
2022-01
影响因子：
5.9
通讯作者：
Xuetao Wei
中科院分区：
文献类型：
--
作者：
Bo Yin;Xuetao Wei
众包市场的发展利用了人类智能的力量来解决计算上具有挑战性的问题。由于越来越复杂的比较标准（例如偏序），帕累托最优查询在微妙和全面的决策支持中变得越来越流行。然而，之前的工作都没有集中于以最小的货币成本和具有置信水平的质量保证对部分订单进行高效的众包帕累托最优查询。在本文中，我们提出了一个具有成本效益的框架，以最小的货币成本和具有置信水平的质量保证来找到帕累托最优对象。我们首先提出了一个基于Student的<inline-formula><tex-math notation="LaTeX">$t$</tex-math><alternatives><mml:math><mml:mi>t</mml:mi></mml:math><inline-graphic的动态状态判断模型xlink:href="yin-ieq1-3017198.gif"/></alternatives></inline-formula>-distribution，给出判断的置信区间，以保证两两比较的质量，同时最大限度地减少每次两两比较所需的众包人员数量。然后，我们提出了一种过滤验证方案，充分利用传递性来避免不必要的众包比较，从而显着减少众包成对比较的数量。我们大量的实验结果表明，动态状态判断模型在保持准确性的情况下需要少量众包进行成对比较，而过滤验证方案可以平均减少 40% 的成对比较次数。
The development of crowdsourcing marketplaces has leveraged the power of human intelligence into tackling computationally challenging problems. Pareto-optimal queries become more and more popular in subtle and comprehensive decision support due to the increasingly complex comparison criteria, e.g., partial orders. However, none of previous work focused on efficient crowdsourced Pareto-optimal queries over partial orders with minimum monetary cost and quality guarantee with a confidence level. In this article, we propose a cost-efficient framework to find Pareto-optimal objects with minimum monetary cost and quality guarantee with a confidence level. We first propose a dynamic-status judgment model based on Student’s <inline-formula><tex-math notation="LaTeX">$t$</tex-math><alternatives><mml:math><mml:mi>t</mml:mi></mml:math><inline-graphic xlink:href="yin-ieq1-3017198.gif"/></alternatives></inline-formula>-distribution, which gives a confidence interval of the judgment to ensure the quality of pairwise comparisons while minimizing the number of crowdsourcers required for each pairwise comparison. We then propose a filtering-verification scheme that takes full advantage of transitivity to avoid unnecessary crowdsourcing comparisons, which significantly reduces the number of crowdsourcing pairwise comparisons. The results of our extensive experiments demonstrate that the dynamic-status judgment model requires a small number of crowdsourcers for a pairwise comparison while maintaining the accuracy, and the filtering-verification scheme can reduce the number of pairwise comparisons by 40 percent in average.