Learning multimodal entity representations and their ensembles, with applications in a data-driven advisory framework for video game players