For research tasks 1 and 2: Mean, median, and standard deviation (over 50 samplings of discovery/validation data) of the difference (both unscaled and scaled) between the value of the evaluation criterion on the validation data and the corresponding value on the discovery data.

Additionally, the effect size (mean divided by standard deviation) is reported. ARIdiscov denotes the best ARI on the discovery data and ARIvalid the ARI resulting from the corresponding method combination on the validation data. The quantities #hubsdiscov, #hubsvalid (number of hubs) are defined analogously.

此外，本研究同时报告了效应量（effect size，即均值除以标准差）的计算结果。ARIdiscov指代发现数据集上的最优调整兰德指数（Adjusted Rand Index, ARI），ARIvalid则指代对应方法组合在验证数据集上得到的调整兰德指数。#hubsdiscov与#hubsvalid（即枢纽节点数量）的定义方式与此类似。