I think that recommendation has four types:
1) recommend user with item (use case:content subscriber, generate the more interesting content for user )
2) recommend item to user (use case :advertisement pushing ,push the advertisement to the people that is interested with)
3) recommend item to item (use case; related content)
4) recommend user to user (use case : people connection )
now org.apache.mahout.cf.taste.hadoop.item can recommend user with item,also can recommend item to user .
we have no off-line implement for 3) and 4). although we already implement the code in 1) and 2).