This paper describes the formulation of a new model for unintrusive targeted advertising on the web, extending the linear programming approach taken by Langheinrich et al.[10] A feature of our model is that it avoids unrealistic solutions of the type which may show ads to only a too-narrow group of users. This is accomplished by using a statistically derived entropy maximization model, which combines a form of randomization in associating advertisements with targetable groups of users, as well as considering click-through probability. It is then shown that this nonlinear entropy model can be embedded in larger models for the purpose of optimal management of web advertisement portfolios by agencies or brokerages.
John A. Tomlin