Optimisation methods for ranking functions with multiple parameters Export

@inproceedings{citeulike:4131580, abstract = {Optimising the parameters of ranking functions with respect to standard IR rank-dependent cost functions has eluded satisfactory analytical treatment. We build on recent advances in alternative differentiable pairwise cost functions, and show that these techniques can be successfully applied to tuning the parameters of an existing family of IR scoring functions (BM25), in the sense that we cannot do better using sensible search heuristics that directly optimize the rank-based cost function NDCG. We also demonstrate how the size of training set affects the number of parameters we can hope to tune this way.}, address = {New York, NY, USA}, author = {Taylor, Michael and Zaragoza, Hugo and Craswell, Nick and Robertson, Stephen and Burges, Chris}, booktitle = {CIKM '06: Proceedings of the 15th ACM international conference on Information and knowledge management}, citeulike-article-id = {4131580}, citeulike-linkout-0 = {http://portal.acm.org/citation.cfm?id=1183614.1183698}, citeulike-linkout-1 = {http://dx.doi.org/10.1145/1183614.1183698}, doi = {10.1145/1183614.1183698}, isbn = {1-59593-433-2}, keywords = {bm25, learn2rank, msr, ranknet}, location = {Arlington, Virginia, USA}, pages = {585--593}, posted-at = {2009-03-04 10:35:42}, priority = {2}, publisher = {ACM}, title = {Optimisation methods for ranking functions with multiple parameters}, url = {http://dx.doi.org/10.1145/1183614.1183698}, year = {2006} }

TY - CONF ID - citeulike:4131580 L3 - citeulike-article-id:4131580 N2 - Optimising the parameters of ranking functions with respect to standard IR rank-dependent cost functions has eluded satisfactory analytical treatment. We build on recent advances in alternative differentiable pairwise cost functions, and show that these techniques can be successfully applied to tuning the parameters of an existing family of IR scoring functions (BM25), in the sense that we cannot do better using sensible search heuristics that directly optimize the rank-based cost function NDCG. We also demonstrate how the size of training set affects the number of parameters we can hope to tune this way. CY - Arlington, Virginia, USA SN - 1-59593-433-2 AD - New York, NY, USA EP - 593 TI - Optimisation methods for ranking functions with multiple parameters PB - ACM T2 - CIKM '06: Proceedings of the 15th ACM international conference on Information and knowledge management SP - 585 KW - bm25 KW - learn2rank KW - msr KW - ranknet AU - Taylor, Michael AU - Zaragoza, Hugo AU - Craswell, Nick AU - Robertson, Stephen AU - Burges, Chris PY - 2006/// UR - http://dx.doi.org/10.1145/1183614.1183698 ER -