Layered Hypernetwork Models for Cross-Modal Associative Text and Image Keyword Generation in Multimodal Information Retrieval