Integrating visual and semantic contexts for topic network generation and word sense disambiguation