Work‎ > ‎Projects‎ > ‎

ClickBoost - A Commerical Large-scale Framework of Click Models for

posted Oct 23, 2011, 9:12 PM by Botao Hu   [ updated Oct 24, 2011, 5:47 AM ]
Project Originator and Core Developer
Mar 2010 - Jun 2010
Mentored by Gang Wang
Microsoft Research Asia, Beijing

Probit click model framework is constructed based on the paper Learning Click Models via Probit Bayesian Inference written by Y. Zhang et al. They propose a novel inference approach which can be widely applied to existing click models. The new approach is based on the Bayesian framework. It replaces each probability variable in click models with a new variable following the Gaussian distribution through a probit link function, such that both the prior and the posterior distribution of the Bayesian learning can be approximated by Gaussians. 

This framework is a class library packed in a dynamic linked library file (DLL), which provides several functions and modules for data extraction from raw log, data trimming and filtering, click model training and testing, evaluation and so on. This framework is implemented on Scope cloud system, which is a Map-reduce computing system deployed over a network file system Cosmos. Thus, cloud implementation helps training and testing click model to be parallel and be able to handle large-scale click log data. Of course, this framework also supports to execute on local machine for debug.