Discovering A Domain Knowledge Representation For Image Grouping: Multimodal Data Modeling, Fusion, And Interactive Learning