All modern search engines attempt to detect and correct spelling errors in users‘ search queries. Google, for example, was one of the first to offer such a facility, and today we barely notice when we are asked "Did you mean x?" after a slip on the keyboard. This article shows you one way of adding a "did you mean" suggestion facility to your own search applications using the Lucene Spell Checker, an extension written by Nicolas Maisonneuve and David Spencer.
Automatic spell checking has a long history. One important early paper was F. Damerau‘s A Technique for Computer Detection and Correction of Spelling Errors, published in 1964, which introduced the idea of minimum edit distance. Briefly, the concept of edit distance quantifies the idea of one string being "close" to another, by counting the number of character edit operations (such as insertions, deletions and substitutions) that are needed to transform one string into the other. Using this metric, the best suggestions for a misspelling are those with the minimum edit distance.
Another approach is the similarity key technique, in which words are transformed into some sort of key so that similarly spelled and, hopefully, misspelled words have the same key. To correct a misspelling simply involves creating the key for the misspelling and looking up dictionary words with the same key for a list of suggestions. Soundex is the best-known similarity key, and is often used for phonetic applications.
A combination of minimum edit distance and similarity keys (metaphone) is at the heart of the successful strategy used by Aspell, the leading open source spell checker. However, it is a third approach that underlies the implementation of the "did you mean" technique described in this article: letter n-grams.
A letter n-gram is a sequence of n letters of a word. For instance, the word "lucene" can be divided into four 3-grams, also known as trigrams: "luc", "uce", "cen", and "ene.". Why is it useful to break words up like this? The intuition is that misspellings typically only affect a few of the constituent n-grams, so we can recognize the intended word just by looking through correctly spelled words for those that share a high proportion of n-grams with the misspelled word. There are various ways of computing this similarity measure, but one powerful way is to treat it as a classic search engine problem with an inverted index of n-grams into words. This is precisely the approach taken by Lucene Spell Checker. Let‘s see how to use it.
聯(lián)系客服
