Extreme Multi-label Loss Functions for Recommendation, Tagging, Ranking &amp; Other Missing Label Applications

Jain, Himanshu; Prabhu, Yashoteja; Varma, Manik

doi:10.1145/2939672.2939756

Cited by 240 publications

(286 citation statements)

References 21 publications

Supporting

Mentioning

285

Contrasting

Unclassified

Order By: Relevance

“…Baselines: XReg was compared to leading extreme classifiers such as PfastreXML [26], Parabel [46], DiSMEC [5] and ProXML [6], traditional multivariate regressors such as one-vs-all least-squares regression (1-vs-all-LS) and LEML [67], and a popular pairwise Table 2: XReg achieves the best or close to the best ranking and regression performance in both pointwise ("-p") and labelwise ("-l") prediction settings. Re-ranking with tail classifiers (XReg-t) further improves the performance in many cases.…”

Section: Methodsmentioning

confidence: 99%

“…Slice only works on low-dimensional embeddings and does not scale to high-dimensional bag-of-words features used in this paper. Some extreme classifiers like PfastreXML [26] and LEML [67] can be easily adapted to learn from any relevance weights, but they tend to be inaccurate and inefficient since they train a large ensemble of weak trees and inaccurate low-dimensional projections with linear reconstruction time, respectively.…”

Section: Related Workmentioning

confidence: 99%

“…Regression & ranking: Most of the conventional regression approaches [7,17,53,56,67] learn a separate regressor for each outcome variable and hence do not scale to millions of labels. This problem is mitigated to some extent in the multi-objective decision tree based approaches [4,26,32] which scale sublinearly in the number on outcome variables. However, these approaches suffer from low accuracy issues despite learning a large ensemble of weak trees.…”

Section: Related Workmentioning

confidence: 99%

“…However, these approaches suffer from low accuracy issues despite learning a large ensemble of weak trees. As seen from experiments, XReg can be significantly more scalable and accurate than the naive 1-vs-all least squares regressor [56], the more efficient LEML regressor with low-rank assumption on the parameter space [67] and the decision tree based PfastreXML [26]. The performance accuracy in regression have traditionally been measured by error metrics such as Mean Absolute Deviation (MAD) and Root Mean Square Error (RMSE) [10], but these are not appropriate for XR.…”

Section: Related Workmentioning

confidence: 99%

“…Re-ranking the relevance estimates could significantly improve the final ranking quality, especially when the XMAD errors are small. An example of re-ranking is to combine these estimates with the scores from tail classifiers (see [26]) to improve the recommendation accuracies over rare labels. It is worth noting that bad relevance estimates, despite inducing a good initial ranking, could hurt the subsequent filtering or re-ranking performance.…”

Section: Extreme Regression Metricsmentioning

confidence: 99%

See 4 more Smart Citations

Extreme Regression for Dynamic Search Advertising

Prabhu

Kusupati

Gupta

et al. 2020

Proceedings of the 13th International Conference on Web Search and Data Mining

Self Cite

View full text Add to dashboard Cite

This paper introduces a new learning paradigm called eXtreme Regression (XR) whose objective is to accurately predict the numerical degrees of relevance of an extremely large number of labels to a data point. XR can provide elegant solutions to many large-scale ranking and recommendation applications including Dynamic Search Advertising (DSA). XR can learn more accurate models than the recently popular extreme classifiers which incorrectly assume strictly binary-valued label relevances. Traditional regression metrics which sum the errors over all the labels are unsuitable for XR problems since they could give extremely loose bounds for the label ranking quality. Also, the existing regression algorithms won't efficiently scale to millions of labels. This paper addresses these limitations through: (1) new evaluation metrics for XR which sum only the k largest regression errors; (2) a new algorithm called XReg which decomposes XR task into a hierarchy of much smaller regression problems thus leading to highly efficient training and prediction. This paper also introduces a (3) new labelwise prediction algorithm in XReg useful for DSA and other recommendation tasks.Experiments on benchmark datasets demonstrated that XReg can outperform the state-of-the-art extreme classifiers as well as large-scale regressors and rankers by up to 50% reduction in the new XR error metric, and up to 2% and 2.4% improvements in terms of the propensity-scored precision metric used in extreme classification and the click-through rate metric used in DSA respectively. Deployment of XReg on DSA in Bing resulted in a relative gain of 27% in query coverage. XReg's source code can be downloaded from [1]

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Extreme Regression Metricsmentioning

confidence: 99%

See 3 more Smart Citations

Extreme Regression for Dynamic Search Advertising

Prabhu

Kusupati

Gupta

et al. 2020

Proceedings of the 13th International Conference on Web Search and Data Mining

Self Cite

View full text Add to dashboard Cite

show abstract

Conclusions and Future Research Directions

Naik

Rangwala

2018

Large Scale Hierarchical Classification: State of the Art

View full text Add to dashboard Cite

PParabel: Parallel Partitioned Label Trees for Extreme Classification

Lu¹,

Zheng²,

Hu³

2019

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Extreme classification consists of extreme multi-class or multilabel predictions, whose objective is to learn classifiers that can label each data point with the most relevant labels. Recently, some approaches such as 1-vs-all method have been proposed to accomplish the task. However, their training time is linear with the number of classes, which makes them unrealistic in real-world applications such as text and image tagging. In this work, we are motivated to present a two-stage thread-level parallelism which is based on Partitioned Label Trees for Extreme Classification (Parabel). Our method is able to train the tree nodes in different parallel ways according to their number of labels. We compare our algorithm with recent state-of-the-art approach on some publicly available real-world datasets which have up to 670,000 labels. The experimental results demonstrate that our algorithm achieves the shortest training time.

show abstract

Extreme Multi-label Loss Functions for Recommendation, Tagging, Ranking & Other Missing Label Applications

Cited by 240 publications

References 21 publications

Extreme Regression for Dynamic Search Advertising

Extreme Regression for Dynamic Search Advertising

Conclusions and Future Research Directions

PParabel: Parallel Partitioned Label Trees for Extreme Classification

Contact Info

Product

Resources

About