jbktree

jbktree provides a generic BK-tree implemented as a java.util.Collection. A BK-tree is a kind of metric tree designed for use in discrete metric spaces. BK-trees are generally used to efficiently conduct k-nearest neighbor searches.

A common use case for BK-trees (but certainly not the only use case) is "fuzzy matching" for strings, or "spell-checking." In this use case, we add a list of known words into a BK-tree, and can then search the tree for words that are within a certain edit distance of a query term. To demonstrate this use case, we can start by loading a list of words into a List.

final List<String> words;

// Under macOS, /usr/share/dict/words contains a list of 235,886 English words
try (final BufferedReader reader = new BufferedReader(new FileReader("/usr/share/dict/words"))) {
    words = reader.lines().collect(Collectors.toList());
}

Next, we define the distance function we'd like to use to calculate the edit distance between words in the list. In this case, we'll use the Levenshtein distance implementation from Commons Text:

final DiscreteDistanceFunction<String> distanceFunction = (first, second) ->
        LevenshteinDistance.getDefaultInstance().apply(first, second);

With a distance function and a collection of words, constructing a BK-tree is staightforward:

final BKTree<String> bkTree = new BKTree<>(distanceFunction, words);

Now, let's say we have a word that we think is misspelled ("exaple"), and we want to find some possible replacements. We can search the BK-tree for all other words that are within a certain edit distance ("radius") of our query term to get some suggestions:

final PriorityQueue<String> results = bkTree.getNearestNeighbors("exaple", 2);

That gives us the following results:

Neighbor	Distance
example	1
hexapla	2
vexable	2
exciple	2
exile	2
exhale	2
staple	2
epaule	2
enable	2
elapse	2
eagle	2
saple	2
maple	2
exalt	2

jchambers/jbktree

jbktree