A field study of a visual controllable talk recommender