As has been noted, there isn't a terribly efficient* way to do it with the datastructure you have specified. However, if you add an additional Map<Integer, List<String>>
to keep track of the mapping from string length to the list of all keys with that length, then you will be able to do this very efficiently.
*Using just the Map<String, Integer>, you would need to iterate through the entire capacity of the larger map, whereas adding this supplemental datastructure would impose an O(1) lookup (assuming you used a HashMap) followed by iteration through just the result set, which is the fastest possible outcome.