A Multiple Source based Transfer Learning Framework for Marketing Campaigns

Abstract

The rapid growing number of marketing campaigns demands an efficient learning model to identify prospective customers to target. Transfer learning is widely considered as a major way to improve the learning performance by using the generated knowledge from previous learning tasks. Most recent studies focused on transferring knowledge from source domains to target domains which may result in knowledge missing. To avoid this, we proposed a multiple source based transfer learning framework to do it reversely. The data in target domains is transferred into source domains by normalizing them into the same distributions and then improving the learning task in target domains by its generated knowledge in source domains. The proposed method is general and can deal with supervised and unsupervised inductive and transductive learning simultaneously with a compatibility to work with different machine learning models. The experiments on real-world campaign data demonstrate that the proposed method outperform state-of-theart methods.