Secretary Problem Study Based on Adaptive Benchmark
The Secretary Problem (SP) describes the decision task of the sequential observation and selection with the characteristics of dynamic decision process where decision makers have to decide to accept or reject the current alternative without knowledge of future available alternatives. Many studies demonstrated the cut-off rule to solve the SP where the maximal one of all observed items is set as the benchmark is the most important characteristics of the decision rule. However, the established rule is based on the hypothesis of human complete rationality, and recent studies show that decision-maker is apt to prematurely stop the search process. We propose the assumption of the mean-benchmark and sub-maximum-benchmark on the ground of the human bounded rationality, then testify and compare the effectiveness of the proposed benchmarks by means of simulation experiment and deductive solution. Results show that the improved rule is better than the established rule from the perspective of decision optimization degree which exhibits an U-shape curve change with the changing benchmark, and the best point is located at the mean-benchmark 20％l or so (l denotes the number of the observed items). On the other hand, the probability of the winning becomes declines linearly with the changing benchmark, and under the sub-maximum-benchmark the largest probability of the winning is 25％ when the cut-off threshold is 50％n (n denotes the number of all items).
Decision analysis secretary problem simulation experiment benchmark
Kan-liang WANG Qing-shun LIU Yuan WANG Hui-min LIU
The School of Management,Xian Jiaotong University,China
国际会议
北京
英文
2007-05-30（万方平台首次上网日期，不代表论文的发表时间）