对于无监督的排序学习算法来说，排序结果的评价指标是非常具有挑战性的问题.从一致性的角度，比较了4种比较典型的无监督排序学习方法，并在机器学习标准数据库中进行试验比较分析.结果显示，RPC这种非线性的无监督排序融合方法产生的排序结果有最小的Kendall距离和Spearman简捷距离，体现了RPC在无监督排序方法的优越性.
Unsupervised ranking has a big challenge that there is no standard metric to measure the ranking results. This paper tries to propose a comparison scheme for unsupervised ranking based on ranking consensus. Some metrics used in supervised ranking can be used here to measure unsupervised ranking consensus. However, these metrics are taken between ranking lists and attributes, rather than ranking lists and ranking labels. Experimental results on UCI datasets show that RPC, which is one kind of unsupervised ranking aggregation method, has the minimum Kendall Distance and Spearman Footrule Distance than the other representative unsupervised ranking methods.
[1] DWORKC,KUMAR R, NAOR M, SIVAKUMAR D.Rank aggregation methods for the web [J]. Proceedings of 10th International Conference on World Wide Web, 2001, 613-622
[2] LI H. A short introduction to learning to rank [J].IEICE Transactions on Information Systems, 2011, E94-D(10):1-9
[3] VOLKOVS M N, ZENEL R S. New learning methods for supervised and unsupervised preference aggregation [J]. Journal of Machine Learning Research, 2014, 15:1135-1176
[4] PEDRONETTE D C G, DA SILVA TORRES R, CALUMBY R T. Using contextual spaces for image re-ranking and rank aggregation[J]. Multimedia Tools and Applications, 2014, 69(3): 689-716
[5] LIN S. Rank aggregation methods [J]. Wiley Interdisciplinary Reviews: Computational Statistics, 2010, 2(5):555-570
[6] KUMAR R, VASSILVITSKII S. Generalized distances between rankings [J]. Proceedings of International Conference on World Wide Web, 2010, 571-580.
[7] KENDALLM. A new measure of rank correlation [J], Biometrika, 1938, 30:81-89
[8] CONTRERAS I. Emphasizing the rank positions in a distance-based aggregation procedure[J]. Decision Support Systems, 2011, 51(1): 240-245.
[9]BAEZA-YATES R, RIBEIRO-BETO B.Information retrieval [M]. Addison-Wesley, 1999.
[10]LI CG, MEI X, and HU B-G. Unsupervised ranking on multi-attribute objects based on principal curves. arXiv:1402.4542[cs.LG], 19 Feb. 2014.
[11]BISHOP C. Pattern recognition and machine learning [M]. New York: Springer, 2006.
