Goal-driven Look-alike Modelingfor Mobile Consumers

Look-alike modeling (LAM) is a well-adopted practice in the marketing industry to reach the right audience or target users of specific interests who fits the profile of a potential customer. In this paper, we propose a goal-driven LAM (GLAM) approach which not only caters to the audience extension problem but also gives businesses the flexibility to choose the basis for similarity. In addition, the proposed solution has been implemented at a large scale with sublinear execution time - an essential requirement as the order of complexity of such problems can be computationally high. We tested the efficacy of our solution by statistically computing and comparing features
⁹
distribution between different sets of audiences. Our experimental results show that there is close proximity between seed audience and GLAM selected look-alike audience compared to the initial search population.