SoftRank: optimizing non-smooth rank metrics

By Michael J. Taylor, John Guiver, Stephen E. Robertson, Tom Minka. Web Search and Data Mining, 2008

402 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Information Systems

View Paper: https://www.semanticscholar.org/paper/67984cf0d946854741cd96d64e030a0db797dc0c

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Information Systems

Citations in the corpus

Ranking Measures and Loss Functions in Learning to Rank (2009, background): “Note that recently people try to directly optimize ranking measures [17, 12, 14, 18].”
Learning diverse rankings with multi-armed bandits (2008, background): “Web search has become an essential component of the Internet infrastructure, and has hence attracted significant interest from the machine learning community (e.g. Herbrich et al., 2000; Burges et al., 2005; Radlinski & Joachims, 2005; Chu & Ghahramani, 2005; Metzler & Croft, 2005; Yue et al., 2007;”
BoltzRank: learning to maximize expected ranking gain (2009, methodology): “The initial increase in NDCG supports the argument of(Taylor et al.,2008) thatNDCGplaces too much emphasis on top-ranked documents, making learning di.cult.”
End-to-End Training of Object Class Detectors for Mean Average Precision (2016, background): “corporate NMS and train simultaneously for multiple classes. Thus, while [19] trains for AP over binary window classiﬁcation scores, ours trains directly for mAP over object detections. Taylor et al. [27] discuss a different formulation for gradient-descent optimisation of certain losses based on ra”
LETOR: A benchmark collection for research on learning to rank for information retrieval (2010, methodology): “2007), SoftRank (Taylor et al. 2008), SVM-MAP (Yue et al.”
Ordinal Depth Supervision for 3D Human Pose Estimation (2018, background): “Ordinal relations: There is a long history for learning from ordinal relations, outside the field of computer vision, with particular interest in the area of information retrieval, where many algorithms for learning-to-rank have been developed [7, 8, 46].”
Yahoo! Learning to Rank Challenge Overview (2010, methodology): “For instance, Taylor et al. (2008) reports in the table 1 of their paper that, on TREC data, SoftRank yields a massive 15.8% NDCG@10 improvement over mean squared error optimization; but on their internal web search data, both methods give similar results.”
Learning to Rank by Optimizing NDCG Measure (2009, background): “This mismatch has motivated the so called listwise approaches for information ranking, which treats each ranking list of documents for a query as a training instance [2, 12, 13, 14, 15, 16, 17].”
A Short Introduction to Learning to Rank (2011, methodology): “The listwise approach includes ListNet [18], ListMLE [19], AdaRank [20], SVM MAP [21], and Soft Rank [22].”
Fast Differentiable Sorting and Ranking (2020, methodology): “Taylor et al. (2008) use a random perturbation technique to compute expected ranks in O(n3) time, where n is the dimensionality of the vector to rank.”
TFMAP: optimizing MAP for top-n context-aware recommendation (2012, background): “SoftRank [31] was the first approach that proposed an explicit smoothed version of an evaluation measure, in which a rank distribution was employed, resulting in the expected values of document ranks that are smooth to the predicted relevance scores.”
dipIQ: Blind Image Quality Assessment by Learning-to-Rank Discriminable Image Pairs (2017, methodology): “Representative algorithms include SoftRank [79], SVM map [80], and RankGP [81].”
Ranking with ordered weighted pairwise classification (2009, methodology): “In the first category, the methods define smooth approximations of quality measures used in IR, in a regression framework (Cossock & Zhang, 2006) or using probability distributions on predicted ranks (Taylor et al., 2008).”