Large-scale targeted marketing by supervised pagerank with seeds

Abstract

In targeted marketing, the key is to spend the limited resources on as relevant a group of customers as possible to the campaign objective. We consider the business problem of selecting a group of standard/non-premier service customers and new customers to whom promotional means, e.g. issuing discount coupons are directed, with the goal of upgrading them to core premier service users. We develop a solution framework based on utilizing the anonymized interaction of user activities, within which the users are scored by their relevance to the marketing campaign objective. The links between two users are weighted, with the weights learnt in a supervised setting to ensure high relevance to the score prediction task. We modified a seeded variant of the PageRank algorithm to adept to this framework while maintaining convergence property. We demonstrate through real-world data that our framework can significantly improve the prediction relevance over conventional methods with regard to the marketing problem under consideration.