I had ranked data showing preferences from 1 to 7 where 1 was the most preferred and 7 the least, but on the Kruskal-Wallis I get mean ranks of 368, 168, and 155 etc.
Give the rank of 1 to the least preferred, 2 to the next least preferred, and so on. I mean you have to reverse your coding, then it will be clear. The out put is in 'Rank Sum' not a mean rank. So, the groups which takes the max value will be the most preferred in that case.