Yahoo! Learning to Rank Challenge Overview

By Olivier Chapelle, Yi Chang. 2010

671 citations recorded by the source corpus.

How researchers see this paper

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

Citations in the corpus

Machine Learning that Matters (2012, background): “Learning to Rank Challenge (Chapelle & Chang, 2011) in that they do not focus on any single problem domain, nor a particular technical capability.”
Pre-training Tasks for Embedding-based Large-scale Retrieval (2020, methodology): “One example is BM-25 (Robertson et al., 2009), which remains to be the most commonly-used (Nguyen et al., 2016; Yang et al., 2017; 2019a) and hard to beat (Chapelle & Chang, 2011; Lee et al., 2019) algorithm.”
An up-to-date comparison of state-of-the-art classification algorithms (2017, methodology): “Learning to Rank Challenge”, Chapelle and Chang (2011) pointed out that decision trees (especially GBDT) were the most popular class of functions among the top competitors.”
Cascade Ranking for Operational E-commerce Search (2017, background): “While most research in learning to rank for search only focuses on the effectiveness issue (e.g., [6, 10, 16]), some valuable prior research of learning to efficiently rank (e.g., [21, 23]) has been proposed to address the trade-off between effectiveness and efficiency.”
"On the (Non-)existence of Convex, Calibrated Surrogate Losses for Ranking" (2012, background): “This characterization and the ﬁrst two points give the result.”
Efficient Cost-Aware Cascade Ranking in Multi-Stage Retrieval (2017, methodology): “Two di � erent test collections were used for the collection is the C14 Webscope Yahoo Learning To Rank dataset [9].”
Relative confidence sampling for efficient on-line ranker evaluation (2014, unknown): “YLR Set 1 19,944 473,134 519 [4] YLR Set 2 1,266 34,815 596 [4]”
Online Structured Prediction via Coactive Learning (2012, methodology): “We evaluated our Preference Perceptron algorithm on the Yahoo! learning to rank dataset (Chapelle & Chang, 2011).”
Learning Scoring Functions with Order-Preserving Losses and Standardized Supervision (2011, methodology): “The ERR (Chapelle et al., 2009) was recently used as the oﬃcial evaluation metric in a learning to rank challenge (Chapelle & Chang, 2011).”
MergeDTS (2018, methodology): “g uniform randomness as in MergeRUCB, to choose the arms to be played. We experiment with MergeDTS in the scenario of online ranker evaluation on the widely used Microsoft and Yahoo! learning to rank [7, 11, 32] datasets, and also provide theoretical performance guarantees for it. We show that Merge”
Shallow pooling for sparse labels (2021, background): “Unfortunately, attempts to extend these methods to content-based ranking have had a mixed record unless substantial volumes of data training are available (Svore & Burges, 2009; Chapelle and Chang, 2011).”
DART: Dropouts meet Multiple Additive Regression Trees (2015, methodology): “For example, in the Yahoo! learning to rank challenge, the winners employed boosted trees (Chapelle and Chang, 2011) based on the LambdaMart method (Burges, 2010).”
VHT: Vertical hoeffding tree (2016, methodology): “The algorithm is in spirit similar to the VHT, as it uses vertical parallelism and runs on top of distributed SPEs.”
A cross-benchmark comparison of 87 learning to rank methods (2015, background): “Learning to Rank Challenge datasets (Chapelle & Chang, 2011), the Yandex Internet Mathematics 2009 contest, 2 the LETOR datasets (Qin, Liu, Xu, & Li, 2010), and the MSLR (Microsoft Learning to Rank) datasets.”
The Greedy Miser: Learning under Test-time Budgets (2012, methodology): “We conduct experiments on two benchmark tasks from very different domains: the Yahoo Learning to Rank Challenge data set (Chapelle & Chang, 2011) and the scene recognition data set from Lazebnik et al. (2006).”
LIMSI Submission for WMT’14 QE Task (2014, background): “tasks (Chapelle and Chang, 2011), in which only a few dense and continuous features are available, possibly because of their ability to take into account complex interactions between features and to automatically partition the continuous features value into a discrete set of intervals that achieves ”