Generating effective representations

Generating effective representations of potential advertisements, commercials, or other messaging such as public safety or health messages, providing a quantitative indication of effectiveness, an effectiveness score, of different portions of the representations. Inputting a user desired property for a representation, training a machine learning model with a data set of representations and indications of effectiveness relative to different properties to output effectiveness scores that indicate effectiveness relative to the desired property, analysing the data set to determine a contribution to the predicted effectiveness scores arising from the components of representations. Properties may be colours, images, words, video or sound. The correlation between contribution to effectiveness and desired property for different components may be obtained using a machine learning model. This correlation data may be statistically analysed to determine said contribution to predicted effectiveness by the components. A generative machine learning model is used with the set of representations, the predicted effectiveness scores, and the component contributions to generate a candidate set of advertisements. The set can be processed by the trained machine learning model and then the analysing module; predicting effectiveness scores and the contribution arising from the components.