Real-time comparable phrases searching via the levenshtein distance with the use of CUDA technology

Abstract

The paper presents a real-time method for finding strings similar to a given pattern. The method is based on the Levenshtein metric with the Wagner–Fischer algorithm being adopted. An improvement is proposed to this well-known technique, a histogram-based approach which resulted in significant reduction of calculation time without a noticeable loss of correctness. Additionally, the used Wagner–Fischer algorithm has been massively parallelized with CUDA technology. The presented method is very flexible as one can define a task-suitable vocabulary, even for abstract elements, far beyond applications relevant to alphanumeric objects. The presented approach seems to be promising for networking and security applications as it is suitable for real-time analysis of data streams.