Scalable Multidimensional Hierarchical Bayesian Modeling on Spark

Robert Ormandi, Hongxia Yang, Quan Lu
; Proceedings of the 4th International Workshop on Big Data, Streams and Heterogeneous Source Mining: Algorithms, Systems, Programming Models and Applications, PMLR 41:33-48, 2015.

Abstract

We consider the problem of estimating occurrence rates of rare events for extremely sparse data using pre-existing hierarchies and selected features to perform inference along multiple dimensions. In particular, we focus on the problem of estimating click rates for {Advertiser, Publisher, User} tuples where both the Advertisers and the Publishers are organized as hierarchies that capture broad contextual information at different levels of granularities. Typically, the click rates are low and the coverage of the hierarchies and dimensions is sparse. To overcome these difficulties, we decompose the joint prior of the three-dimensional Click-Through-Rate (CTR) using tensor decomposition and propose a Multidimensional Hierarchical Bayesian framework (abbreviated as MadHab). We set up a specific framework of each dimension to model dimension-specific characteristics. More specifically, we consider the hierarchical beta process prior for the Advertiser dimension and for the Publisher dimension respectively and a feature-dependent mixture model for the User dimension. Besides the centralized implementation, we propose a distributed algorithm through Spark for inference which make the model highly scalable and suited for large scale data mining applications. We demonstrate that on a real world ads campaign platform our framework can effectively discriminate extremely rare events in terms of their click propensity.

Cite this Paper


BibTeX
@InProceedings{pmlr-v41-ormandi15, title = {Scalable Multidimensional Hierarchical Bayesian Modeling on Spark}, author = {Robert Ormandi and Hongxia Yang and Quan Lu}, booktitle = {Proceedings of the 4th International Workshop on Big Data, Streams and Heterogeneous Source Mining: Algorithms, Systems, Programming Models and Applications}, pages = {33--48}, year = {2015}, editor = {Wei Fan and Albert Bifet and Qiang Yang and Philip S. Yu}, volume = {41}, series = {Proceedings of Machine Learning Research}, month = {10 Aug}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v41/ormandi15.pdf}, url = {http://proceedings.mlr.press/v41/ormandi15.html}, abstract = {We consider the problem of estimating occurrence rates of rare events for extremely sparse data using pre-existing hierarchies and selected features to perform inference along multiple dimensions. In particular, we focus on the problem of estimating click rates for {Advertiser, Publisher, User} tuples where both the Advertisers and the Publishers are organized as hierarchies that capture broad contextual information at different levels of granularities. Typically, the click rates are low and the coverage of the hierarchies and dimensions is sparse. To overcome these difficulties, we decompose the joint prior of the three-dimensional Click-Through-Rate (CTR) using tensor decomposition and propose a Multidimensional Hierarchical Bayesian framework (abbreviated as MadHab). We set up a specific framework of each dimension to model dimension-specific characteristics. More specifically, we consider the hierarchical beta process prior for the Advertiser dimension and for the Publisher dimension respectively and a feature-dependent mixture model for the User dimension. Besides the centralized implementation, we propose a distributed algorithm through Spark for inference which make the model highly scalable and suited for large scale data mining applications. We demonstrate that on a real world ads campaign platform our framework can effectively discriminate extremely rare events in terms of their click propensity.} }
Endnote
%0 Conference Paper %T Scalable Multidimensional Hierarchical Bayesian Modeling on Spark %A Robert Ormandi %A Hongxia Yang %A Quan Lu %B Proceedings of the 4th International Workshop on Big Data, Streams and Heterogeneous Source Mining: Algorithms, Systems, Programming Models and Applications %C Proceedings of Machine Learning Research %D 2015 %E Wei Fan %E Albert Bifet %E Qiang Yang %E Philip S. Yu %F pmlr-v41-ormandi15 %I PMLR %J Proceedings of Machine Learning Research %P 33--48 %U http://proceedings.mlr.press %V 41 %W PMLR %X We consider the problem of estimating occurrence rates of rare events for extremely sparse data using pre-existing hierarchies and selected features to perform inference along multiple dimensions. In particular, we focus on the problem of estimating click rates for {Advertiser, Publisher, User} tuples where both the Advertisers and the Publishers are organized as hierarchies that capture broad contextual information at different levels of granularities. Typically, the click rates are low and the coverage of the hierarchies and dimensions is sparse. To overcome these difficulties, we decompose the joint prior of the three-dimensional Click-Through-Rate (CTR) using tensor decomposition and propose a Multidimensional Hierarchical Bayesian framework (abbreviated as MadHab). We set up a specific framework of each dimension to model dimension-specific characteristics. More specifically, we consider the hierarchical beta process prior for the Advertiser dimension and for the Publisher dimension respectively and a feature-dependent mixture model for the User dimension. Besides the centralized implementation, we propose a distributed algorithm through Spark for inference which make the model highly scalable and suited for large scale data mining applications. We demonstrate that on a real world ads campaign platform our framework can effectively discriminate extremely rare events in terms of their click propensity.
RIS
TY - CPAPER TI - Scalable Multidimensional Hierarchical Bayesian Modeling on Spark AU - Robert Ormandi AU - Hongxia Yang AU - Quan Lu BT - Proceedings of the 4th International Workshop on Big Data, Streams and Heterogeneous Source Mining: Algorithms, Systems, Programming Models and Applications PY - 2015/08/31 DA - 2015/08/31 ED - Wei Fan ED - Albert Bifet ED - Qiang Yang ED - Philip S. Yu ID - pmlr-v41-ormandi15 PB - PMLR SP - 33 DP - PMLR EP - 48 L1 - http://proceedings.mlr.press/v41/ormandi15.pdf UR - http://proceedings.mlr.press/v41/ormandi15.html AB - We consider the problem of estimating occurrence rates of rare events for extremely sparse data using pre-existing hierarchies and selected features to perform inference along multiple dimensions. In particular, we focus on the problem of estimating click rates for {Advertiser, Publisher, User} tuples where both the Advertisers and the Publishers are organized as hierarchies that capture broad contextual information at different levels of granularities. Typically, the click rates are low and the coverage of the hierarchies and dimensions is sparse. To overcome these difficulties, we decompose the joint prior of the three-dimensional Click-Through-Rate (CTR) using tensor decomposition and propose a Multidimensional Hierarchical Bayesian framework (abbreviated as MadHab). We set up a specific framework of each dimension to model dimension-specific characteristics. More specifically, we consider the hierarchical beta process prior for the Advertiser dimension and for the Publisher dimension respectively and a feature-dependent mixture model for the User dimension. Besides the centralized implementation, we propose a distributed algorithm through Spark for inference which make the model highly scalable and suited for large scale data mining applications. We demonstrate that on a real world ads campaign platform our framework can effectively discriminate extremely rare events in terms of their click propensity. ER -
APA
Ormandi, R., Yang, H. & Lu, Q.. (2015). Scalable Multidimensional Hierarchical Bayesian Modeling on Spark. Proceedings of the 4th International Workshop on Big Data, Streams and Heterogeneous Source Mining: Algorithms, Systems, Programming Models and Applications, in PMLR 41:33-48

Related Material