Ranking with ordered weighted pairwise classification

By Nicolas Usunier, David Buffoni, Patrick Gallinari. 2009

184 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Information Systems, Management Science and Operations Research, Evaluation & Benchmarks

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 Information Systems

In Management Science and Operations Research

In Evaluation & Benchmarks

Citations in the corpus

A Neural Network for Factoid Question Answering over Paragraphs (2014, methodology): “This leaves us with 21,041 history questions and 22,956 literature questions.”
Learning Visual Features from Large Weakly Supervised Data (2015, methodology): “In preliminary experiments, we also considered a pairwise ranking loss [50, 53].”
Zero-Shot Learning—A Comprehensive Evaluation of the Good, the Bad and the Ugly (2017, methodology): “ALE [30] uses the weighted approximate ranking objective [65] for zero-shot learning in the following way:”
Zero-Shot Learning — The Good, the Bad and the Ugly (2017, methodology): “ALE [3] uses weighted approximate ranking objective [36]: ∑”
Learning Joint Embedding with Multimodal Cues for Cross-Modal Video-Text Retrieval (2018, background): “2 and modify the loss function following the idea of weighted ranking [31] to weight the loss based on the relative ranking of positive labels.”
Label-Embedding for Image Classification (2015, neutral): “[59] propose to use the following ranking loss for S : where β r ( x n ,y n ) := r ( x n ,y n ) j =1 α j .”
Rank-GeoFM: A Ranking based Geographical Factorization Method for Point of Interest Recommendation (2015, methodology): “Following OWPC [20, 21], we adopt a smooth weighting scheme to convert Incomp ( y uℓ , ε ) into a loss.”
Learning to rank recommendations with the k-order statistic loss (2013, methodology): “For example the ListNet [10], SVMmap [11] or OWPC (ordered weighted pairwise classification) [6] objectives involve computing the rank of positive examples which is too slow in an SGD step when there are hundreds of thousands or millions of items.”
Training highly multiclass classifiers (2014, background): “(12) As in Usunier et al. (2009), Weston et al. (2011) suggest using a weight function w(y+) that is an increasing function of the number of violating classes.”
Top-k Multiclass SVM (2015, methodology): “In the future, one could study if the top-k hinge loss (3) can be generalized to the family of ranking losses [27].”
Towards good practice in large-scale learning for image classification (2012, background): “While [35] proposes an upper-bound on the loss, Weston et al.”
Good Practice in Large-Scale Learning for Image Classification (2014, background): “In this paper, we examine whether these ranking approaches scale well to large datasets and if they improve the performance.”
Label Partitioning For Sublinear Ranking (2013, methodology): “In order to reflect this we introduce a term weighting examples by the inverse of the rank of the relevant label using the original label scorer, following (Usunier et al., 2009).”
Loss Functions for Top-k Error: Analysis and Insights (2015, methodology): “[44], who described a general family of convex loss functions for ranking and classification.”
Top Rank Supervised Binary Coding for Visual Search (2015, methodology): “(4) was previously used for information retrieval [25] and image annotation [30].”
Polysemous Codes (2016, methodology): “In such a case, we optimize where we recall that α i is a decreasing sequence and r ( i, j ) is the rank of j in the ordered list of neighbors to i : In all our ranking experiments, we use Equation (6) and choose α i = 1 /i (following [47]).”
Semantic Frame Identification with Distributed Word Representations (2014, methodology): “Choosing L ( η ) = Cη for any positive constant C optimizes the mean rank, whereas a weighting such as L ( η ) = P ηi =1 1 /i (adopted here) optimizes the top of the ranked list, as described in (Usunier et al., 2009).”
Generalized Low Rank Models (2014, background): “These losses include the area under the curve loss [138], ordered weighted average of pairwise classification losses [148], the weighted approximate-rank pairwise loss [153], the k-order statistic loss [154], and the accuracy at the top loss [14].”