This is a novel generative model which ranks where, geographically, it is most opportunistic to promote certain television programs. Here is a full description of the model and the math behind it.
- computes the expected lift in television ratings of a program by geography if it is promoted there
- ranks the geographies by expected lift in order to decide where to promote
- uses a generative model to realistically model ratings in the presence of promos
- uses the Pomegranate API for maximum likelihood estimation (using the EM algorithm)
- incorporates realistic correlation between viewing behavior when promos are shown and when they are not shown
- uses Facebook's Prophet API to forecast the number of television viewers when the promotional campaign will occur
- is robust from program-to-program and geography-to-geography