The simplest way to express this, beginning with a raw text file, would be to do the two steps below: 1. First, use MR to form a word count. Split each line into words and count them as 1 each in Mapper, and aggregate they counts by word as key in Reducer. 2. Second, use yet another subsequent MR job to read and invert the key to sort it in the opposite form, wherein the key is now the count and the value are the words that match the count. Run either with a TotalOrderPartitioner or a single reducer to get your final result.
... View more
