“Although the total size of the text corpus is large, each team’s corpus is relatively small and thus allows us to devote more computational resources to each message during ranking.”