A cross-modal, tag-cleansing algorithm addresses tag uncertainty by automatically addressing the visual similarities of weakly tagged images with the semantic similarities of their tags.