Learning to Rank with Nonsmooth Cost Functions

By C. Burges, Robert J. Ragno, Quoc V. Le. 2006

913 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.

In Artificial Intelligence

Citations in the corpus

Gradient descent optimization of smoothed information retrieval metrics (2010, background): “The solution proposed in (Burges et al, 2007) is to perform gradient descent on a smoothed version of the objective function.”
LambdaMerge: merging the results of query reformulations (2011, methodology): “The scoring parameters θ and gating weights π of λ-Merge are trained to optimize NDCG using a method based on LambdaRank [4].”
Training Efficient Tree-Based Models for Document Ranking (2013, methodology): “In this context, our work uses LambdaMART [3], which is the combination of LambdaRank [11] and MART [12]—a class of boosting algorithms that performs gradient descent using regression trees.”
L2R²: Leveraging Ranking for Abductive Reasoning (2020, methodology): “s of pairs of hypotheses whose labels are different, where ranking is reduced to a classification on hypotheses pairs. Here, the pairwise loss functions of Ranking SVM [7], RankNet [2] and LambdaRank [3] are used. Hinge loss used in Ranking SVM and logistic (cross entropy) loss used in RankNet both ”
IRGAN: A Minimax Game for Unifying Generative and Discriminative Information Retrieval Models (2017, methodology): “Traditional learning to rank methods like LambdaMART are not particularly eective in this type of semi-supervised seing, which may be due to its reliance on the ∆NDCG scoring for each document pair [5].”
Yahoo! Learning to Rank Challenge Overview (2010, background): “Finally in LambdaRank (Burges et al., 2007), the weight of each preference pair is the NDCG difference resulting from swapping that pair.”
More Accurate Question Answering on Freebase (2015, methodology): “7 We have experimented with state-of-the-art techniques for the learning-to-rank approach from IR [14] [16], including: RankSVM [14], RankBoost [12], LambdaRank [6] and AdaRank [23].”
Real-time Personalization using Embeddings for Search Ranking at Airbnb (2018, methodology): “Next, we formulate the problem as pairwise regression with search labels as utilities and use dataD to train a Gradient Boosting Decision Trees (GBDT) model, using package 4 that was modified to support Lambda Rank.”
Learning to Rank for Information Retrieval and Natural Language Processing (2011, neutral): “IR SVM [13] GBRank [115] Frank [97] LambdaMART [102] LambdaRank [12] Listwise SVM MAP [111] AdaRank [108] ListNet [14] SoftRank [95]”
Topology-Preserving Class-Incremental Learning (2020, background): “However, it is diﬃcult and ineﬃcient to optimize the nonsmooth global ranking [2], while the local ranking can not well preserve the global relations.”
Recommender Systems (2016, neutral): “The main challenge in ranking-based methods is that the underlying objective functions are often non-smooth [490], which can be hard to optimize with off-the-shelf gradient-descent techniques.”
Scalable Learning of Non-Decomposable Objectives (2016, background): “The work of [19] achieves both scalability and generality, but does not cover objectives which place a constraint on the model, such as recall at a ﬁxed precision, precision at a ﬁxed recall, or accuracy at a quantile.”
Beyond User Self-Reported Likert Scale Ratings: A Comparison Model for Automatic Dialog Evaluation (2020, methodology): “Similar to LambdaRank (Burges et al., 2006), it is possible to calculate the exact gradient of O(N2) possible dialog pairs with O(N) forwards and backpropagations.”
Enhancing Single-Document Summarization by Combining RankNet and Third-Party Sources (2007, methodology): “For details, see (Burges et al., 2006; Burges et al., 2005).”