Influencing learning

If you are using the Interact built-in learning module, you can influence the learning output beyond the standard learning configurations such as the list of learning attributes or the confidence level. You can override components of the learning algorithm while using the remaining components.

You can override learning using the LikelihoodScore and AdjExploreScore columns of the default offers and score override tables. You can add these columns to the default offers and score override tables using the aci_scoringfeature feature script. To properly use these overrides, you need a thorough understanding of Interact built-in learning.

The learning module takes the list of candidate offers and the marketing score per candidate offer and uses them in the final calculations. The offer list is used with the learning attributes to calculate the likelihood (accept probability) that the customer will accept the offer. Using these probabilities and the historical number of presentations to balance between exploration and exploitation, the learning algorithm determines the offer weight. Finally, the built-in learning takes the offer weight, multiplies it by the final marketing score and returns a final score. The offers are sorted by this final score.