Electronic promotion to new customers using mkNN learning

In recent years, several techniques have been proposed to model electronic promotions for existing customers. However, these techniques are not applicable for new customers with no previous profile or behavior data. This study models promotions to new customers in an electronic marketplace. We introduce a multi-valued k-Nearest Neighbor (mkNN) learning capability for modeling promotions to new customers. In this modified learning algorithm, instead of a single product category, the seller sends the new customer a promotion on a variable set of m categories (where m is a variable) with the highest rank of desirability among the most similar previous customers. Previous studies consider sellers’ profits in promotion and marketing models. In addition to the sellers’ profits, three important factors – annoyance of customers, sellers’ reputations, and customers’ anonymity – are considered in this study. Without considering the customer’s profile, we minimize unrelated and disliked offers to reduce the customer’s annoyance and elevate the seller’s reputation. The promotion models are evaluated in two separate experiments on populations with different degrees of optimism: (1) with fixed number of customers; and (2) in a fixed period of time. The evaluation is based on the parameters of customer population size and behavior as well as time interval, seller payoff, seller reputation, and the number of promotions canceled by the customers. The simulation results demonstrate that the proposed mkNN-based promotion strategies are moderately efficient with respect to all parameters for providing services in a large population. In addition, purchasing preferences of past customers, which are based on periodic promotions that a seller sends to customers, can generate future rapidly expanding demands in the market. By using these approaches, an advertising company can send acceptable promotions to customers without having specific profile information.