A belief framework for similarity evaluation of textual or structured data

Abstract

This paper discovers the major shortcomings of the Levenshtein Distance method, the longest common subsequence (LCS) method, and other general approaches to finding common parts, including the unjustified fragmentation of selected parts, the lack of sensitivity to transposition of large blocks, and no mechanisms to prevent accidental matches. The belief function theory leads to a flexible framework for similarity evaluation.The framework is aimed on new similarity models which are free of described shortcomings and can be effectively calculated. A sketch of better sequence alignment algorithm illustrates the framework’s utility.