SiameseXML: Siamese Networks meet Extreme Classifiers with 100M Labels

Kunal Dahiya, Ananye Agarwal, Deepak Saini, Gururaj K, Jian Jiao, Amit Singh, Sumeet Agarwal, Purushottam Kar, Manik Varma
Proceedings of the 38th International Conference on Machine Learning, PMLR 139:2330-2340, 2021.

Abstract

Deep extreme multi-label learning (XML) requires training deep architectures that can tag a data point with its most relevant subset of labels from an extremely large label set. XML applications such as ad and product recommendation involve labels rarely seen during training but which nevertheless hold the key to recommendations that delight users. Effective utilization of label metadata and high quality predictions for rare labels at the scale of millions of labels are thus key challenges in contemporary XML research. To address these, this paper develops the SiameseXML framework based on a novel probabilistic model that naturally motivates a modular approach melding Siamese architectures with high-capacity extreme classifiers, and a training pipeline that effortlessly scales to tasks with 100 million labels. SiameseXML offers predictions 2–13% more accurate than leading XML methods on public benchmark datasets, as well as in live A/B tests on the Bing search engine, it offers significant gains in click-through-rates, coverage, revenue and other online metrics over state-of-the-art techniques currently in production. Code for SiameseXML is available at https://github.com/Extreme-classification/siamesexml

Cite this Paper

BibTeX
@InProceedings{pmlr-v139-dahiya21a, title = {SiameseXML: Siamese Networks meet Extreme Classifiers with 100M Labels}, author = {Dahiya, Kunal and Agarwal, Ananye and Saini, Deepak and K, Gururaj and Jiao, Jian and Singh, Amit and Agarwal, Sumeet and Kar, Purushottam and Varma, Manik}, booktitle = {Proceedings of the 38th International Conference on Machine Learning}, pages = {2330--2340}, year = {2021}, editor = {Meila, Marina and Zhang, Tong}, volume = {139}, series = {Proceedings of Machine Learning Research}, month = {18--24 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v139/dahiya21a/dahiya21a.pdf}, url = {https://proceedings.mlr.press/v139/dahiya21a.html}, abstract = {Deep extreme multi-label learning (XML) requires training deep architectures that can tag a data point with its most relevant subset of labels from an extremely large label set. XML applications such as ad and product recommendation involve labels rarely seen during training but which nevertheless hold the key to recommendations that delight users. Effective utilization of label metadata and high quality predictions for rare labels at the scale of millions of labels are thus key challenges in contemporary XML research. To address these, this paper develops the SiameseXML framework based on a novel probabilistic model that naturally motivates a modular approach melding Siamese architectures with high-capacity extreme classifiers, and a training pipeline that effortlessly scales to tasks with 100 million labels. SiameseXML offers predictions 2–13% more accurate than leading XML methods on public benchmark datasets, as well as in live A/B tests on the Bing search engine, it offers significant gains in click-through-rates, coverage, revenue and other online metrics over state-of-the-art techniques currently in production. Code for SiameseXML is available at https://github.com/Extreme-classification/siamesexml} }
Endnote
%0 Conference Paper %T SiameseXML: Siamese Networks meet Extreme Classifiers with 100M Labels %A Kunal Dahiya %A Ananye Agarwal %A Deepak Saini %A Gururaj K %A Jian Jiao %A Amit Singh %A Sumeet Agarwal %A Purushottam Kar %A Manik Varma %B Proceedings of the 38th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2021 %E Marina Meila %E Tong Zhang %F pmlr-v139-dahiya21a %I PMLR %P 2330--2340 %U https://proceedings.mlr.press/v139/dahiya21a.html %V 139 %X Deep extreme multi-label learning (XML) requires training deep architectures that can tag a data point with its most relevant subset of labels from an extremely large label set. XML applications such as ad and product recommendation involve labels rarely seen during training but which nevertheless hold the key to recommendations that delight users. Effective utilization of label metadata and high quality predictions for rare labels at the scale of millions of labels are thus key challenges in contemporary XML research. To address these, this paper develops the SiameseXML framework based on a novel probabilistic model that naturally motivates a modular approach melding Siamese architectures with high-capacity extreme classifiers, and a training pipeline that effortlessly scales to tasks with 100 million labels. SiameseXML offers predictions 2–13% more accurate than leading XML methods on public benchmark datasets, as well as in live A/B tests on the Bing search engine, it offers significant gains in click-through-rates, coverage, revenue and other online metrics over state-of-the-art techniques currently in production. Code for SiameseXML is available at https://github.com/Extreme-classification/siamesexml
APA
Dahiya, K., Agarwal, A., Saini, D., K, G., Jiao, J., Singh, A., Agarwal, S., Kar, P. & Varma, M.. (2021). SiameseXML: Siamese Networks meet Extreme Classifiers with 100M Labels. Proceedings of the 38th International Conference on Machine Learning, in Proceedings of Machine Learning Research 139:2330-2340 Available from https://proceedings.mlr.press/v139/dahiya21a.html.

Related Material