Learning predictive models from massive, semantically disparate data رسالة دكتوراه