Listwise approach to learning to rank: theory and algorithm Export

@inproceedings{citeulike:3940833, abstract = {This paper aims to conduct a study on the listwise approach to learning to rank. The listwise approach learns a ranking function by taking individual lists as instances and minimizing a loss function defined on the predicted list and the ground-truth list. Existing work on the approach mainly focused on the development of new algorithms; methods such as RankCosine and ListNet have been proposed and good performances by them have been observed. Unfortunately, the underlying theory was not sufficiently studied so far. To amend the problem, this paper proposes conducting theoretical analysis of learning to rank algorithms through investigations on the properties of the loss functions, including consistency, soundness, continuity, differentiability, convexity, and efficiency. A sufficient condition on consistency for ranking is given, which seems to be the first such result obtained in related research. The paper then conducts analysis on three loss functions: likelihood loss, cosine loss, and cross entropy loss. The latter two were used in RankCosine and ListNet. The use of the likelihood loss leads to the development of a new listwise method called ListMLE, whose loss function offers better properties, and also leads to better experimental results.}, address = {New York, NY, USA}, author = {Xia, Fen and Liu, Tie Y. and Wang, Jue and Zhang, Wensheng and Li, Hang}, booktitle = {ICML '08: Proceedings of the 25th international conference on Machine learning}, citeulike-article-id = {3940833}, citeulike-linkout-0 = {http://portal.acm.org/citation.cfm?id=1390156.1390306}, citeulike-linkout-1 = {http://dx.doi.org/10.1145/1390156.1390306}, doi = {10.1145/1390156.1390306}, isbn = {978-1-60558-205-4}, keywords = {listwise, ranking}, location = {Helsinki, Finland}, pages = {1192--1199}, posted-at = {2009-06-07 12:38:40}, priority = {2}, publisher = {ACM}, title = {Listwise approach to learning to rank: theory and algorithm}, url = {http://dx.doi.org/10.1145/1390156.1390306}, year = {2008} }

TY - CONF ID - citeulike:3940833 L3 - citeulike-article-id:3940833 N2 - This paper aims to conduct a study on the listwise approach to learning to rank. The listwise approach learns a ranking function by taking individual lists as instances and minimizing a loss function defined on the predicted list and the ground-truth list. Existing work on the approach mainly focused on the development of new algorithms; methods such as RankCosine and ListNet have been proposed and good performances by them have been observed. Unfortunately, the underlying theory was not sufficiently studied so far. To amend the problem, this paper proposes conducting theoretical analysis of learning to rank algorithms through investigations on the properties of the loss functions, including consistency, soundness, continuity, differentiability, convexity, and efficiency. A sufficient condition on consistency for ranking is given, which seems to be the first such result obtained in related research. The paper then conducts analysis on three loss functions: likelihood loss, cosine loss, and cross entropy loss. The latter two were used in RankCosine and ListNet. The use of the likelihood loss leads to the development of a new listwise method called ListMLE, whose loss function offers better properties, and also leads to better experimental results. CY - Helsinki, Finland SN - 978-1-60558-205-4 AD - New York, NY, USA EP - 1199 TI - Listwise approach to learning to rank: theory and algorithm PB - ACM T2 - ICML '08: Proceedings of the 25th international conference on Machine learning SP - 1192 KW - listwise KW - ranking AU - Xia, Fen AU - Liu, Tie AU - Wang, Jue AU - Zhang, Wensheng AU - Li, Hang PY - 2008/// UR - http://dx.doi.org/10.1145/1390156.1390306 ER -