For each user, the items for which a rating is predicted are ranked in descending order based on the ratings, then the topn items are selected. Group ratings are predicted by merging the top-n items of each users with a union of the ratings. If an item appears in the list of more members of the same group, the average of the predicted ratings for that item is calculated (Additive Utilitarian strategy), in order to derive the preference of that group for the item.