Let be a set of textual queries representing universal concepts, and the set of images from different cultures. Given a query , the goal is to retrieve a ranked list of images that maximizes both relevance and cultural diversity.
Relevance: refers to how well the image matches the query captured by the standard precision@k.
Diversity: measures the cultural diversity of the retrieved images using the formula,
where is the proportion of images from the -th culture in the top retrieved images ,
and is the total number of cultures in the top .
A high entropy value (∼ 100) indicates high diversity, retrieved images are well-distributed across different cultures.
Conversely, a low entropy value (∼ 0) indicates low diversity, retrieved images are biased towards specific cultures.