I am trying to calculate edit distances of a string against a collection to find the closest match. My current problem is that the collection is very large (about 25000 item
If your criteria for 'similar' define a total ordering, you should be able to define a Comparator and use a TreeSet to find the closest matches (eg using the ceiling and floor methods).
Sounds like a BK-tree might be what you want. Here's an article discussing them: http://blog.notdot.net/2007/4/Damn-Cool-Algorithms-Part-1-BK-Trees. A quick Google yields some Java implementations.
Levenshtein Automata allow for fast selection of a set of words from a large dictionary such that they are within the given Levenshtein distance from a given word.
See: Schulz K, Mihov S. (2002) Fast String Correction with Levenshtein-Automata.