Fast Phonetic Similarity Search over Large Repositories

Today there is a large amount of unstructured data produced by information systems from different domains. These sources may be analyzed for different purposes. Existing approaches use string similarity methods to search for valid words within a text, with a supporting dictionary. However, they have two main drawbacks. First, they are not rich enough to encode phonetic information to assist the search. Second, the solutions may be inefficient in the presence of spelling errors. In this paper, we present a novel approach for efficiently perform phonetic similarity search over large data sources. We present a data structure called PhoneticMap, which encodes language-specific phonetic information. The phonetic maps are used by a novel fast similarity search algorithm to find words with spelling errors. We validate our approach through an experiment over a data set using a Portuguese variant of a well-known repository, to automatically correct words with spelling errors.