Extreme F-measure Maximization using Sparse Probability Estimates

Kalina Jasinska, Krzysztof Dembczynski, Robert Busa-Fekete, Karlson Pfannschmidt, Timo Klerx, Eyke Hullermeier
Proceedings of The 33rd International Conference on Machine Learning, PMLR 48:1435-1444, 2016.

Abstract

We consider the problem of (macro) F-measure maximization in the context of extreme multi-label classification (XMLC), i.e., multi-label classification with extremely large label spaces. We investigate several approaches based on recent results on the maximization of complex performance measures in binary classification. According to these results, the F-measure can be maximized by properly thresholding conditional class probability estimates. We show that a naive adaptation of this approach can be very costly for XMLC and propose to solve the problem by classifiers that efficiently deliver sparse probability estimates (SPEs), that is, probability estimates restricted to the most probable labels. Empirical results provide evidence for the strong practical performance of this approach.

Cite this Paper


BibTeX
@InProceedings{pmlr-v48-jasinska16, title = {Extreme F-measure Maximization using Sparse Probability Estimates}, author = {Jasinska, Kalina and Dembczynski, Krzysztof and Busa-Fekete, Robert and Pfannschmidt, Karlson and Klerx, Timo and Hullermeier, Eyke}, booktitle = {Proceedings of The 33rd International Conference on Machine Learning}, pages = {1435--1444}, year = {2016}, editor = {Balcan, Maria Florina and Weinberger, Kilian Q.}, volume = {48}, series = {Proceedings of Machine Learning Research}, address = {New York, New York, USA}, month = {20--22 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v48/jasinska16.pdf}, url = { http://proceedings.mlr.press/v48/jasinska16.html }, abstract = {We consider the problem of (macro) F-measure maximization in the context of extreme multi-label classification (XMLC), i.e., multi-label classification with extremely large label spaces. We investigate several approaches based on recent results on the maximization of complex performance measures in binary classification. According to these results, the F-measure can be maximized by properly thresholding conditional class probability estimates. We show that a naive adaptation of this approach can be very costly for XMLC and propose to solve the problem by classifiers that efficiently deliver sparse probability estimates (SPEs), that is, probability estimates restricted to the most probable labels. Empirical results provide evidence for the strong practical performance of this approach.} }
Endnote
%0 Conference Paper %T Extreme F-measure Maximization using Sparse Probability Estimates %A Kalina Jasinska %A Krzysztof Dembczynski %A Robert Busa-Fekete %A Karlson Pfannschmidt %A Timo Klerx %A Eyke Hullermeier %B Proceedings of The 33rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2016 %E Maria Florina Balcan %E Kilian Q. Weinberger %F pmlr-v48-jasinska16 %I PMLR %P 1435--1444 %U http://proceedings.mlr.press/v48/jasinska16.html %V 48 %X We consider the problem of (macro) F-measure maximization in the context of extreme multi-label classification (XMLC), i.e., multi-label classification with extremely large label spaces. We investigate several approaches based on recent results on the maximization of complex performance measures in binary classification. According to these results, the F-measure can be maximized by properly thresholding conditional class probability estimates. We show that a naive adaptation of this approach can be very costly for XMLC and propose to solve the problem by classifiers that efficiently deliver sparse probability estimates (SPEs), that is, probability estimates restricted to the most probable labels. Empirical results provide evidence for the strong practical performance of this approach.
RIS
TY - CPAPER TI - Extreme F-measure Maximization using Sparse Probability Estimates AU - Kalina Jasinska AU - Krzysztof Dembczynski AU - Robert Busa-Fekete AU - Karlson Pfannschmidt AU - Timo Klerx AU - Eyke Hullermeier BT - Proceedings of The 33rd International Conference on Machine Learning DA - 2016/06/11 ED - Maria Florina Balcan ED - Kilian Q. Weinberger ID - pmlr-v48-jasinska16 PB - PMLR DP - Proceedings of Machine Learning Research VL - 48 SP - 1435 EP - 1444 L1 - http://proceedings.mlr.press/v48/jasinska16.pdf UR - http://proceedings.mlr.press/v48/jasinska16.html AB - We consider the problem of (macro) F-measure maximization in the context of extreme multi-label classification (XMLC), i.e., multi-label classification with extremely large label spaces. We investigate several approaches based on recent results on the maximization of complex performance measures in binary classification. According to these results, the F-measure can be maximized by properly thresholding conditional class probability estimates. We show that a naive adaptation of this approach can be very costly for XMLC and propose to solve the problem by classifiers that efficiently deliver sparse probability estimates (SPEs), that is, probability estimates restricted to the most probable labels. Empirical results provide evidence for the strong practical performance of this approach. ER -
APA
Jasinska, K., Dembczynski, K., Busa-Fekete, R., Pfannschmidt, K., Klerx, T. & Hullermeier, E.. (2016). Extreme F-measure Maximization using Sparse Probability Estimates. Proceedings of The 33rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 48:1435-1444 Available from http://proceedings.mlr.press/v48/jasinska16.html .

Related Material