A highly-efficient ANSI-C facility is described for intelligently comparing a query string with a series of database strings. The bipartite weighted matching approach taken tolerates ordering violations that are problematic for simple automaton or string edit distance methods---yet common in practice. The method is character and polygraph based and does not require that words are properly formed in a query. Database characters are processed at a rate of approximately 2.5 million per second using a 200MHz Pentium Pro processor. A subroutine-level API is described along with an simple executable utility supporting both command-line and Web interfaces. An optimized Web interface is also reported consisting of a daemon that preloads multiple databases, and a corresponding CGI stub. The daemon may be initiated manually or via inetd. Keywords: String Comparison/Similarity, Text/Database Search/Retrieval, Bipartite Matching/Assignment, Edit Distance. Both authors are with the NEC Research I...

Approximate string comparison and search is an important part of applications that range from natural language to the interpretation of DNA. This paper presents a bipartite weighted graph matching approach to these problems, based on the authors' linear time matching algorithms # . Our approach's tolerance to permutation of symbols or blocks, distinguishes it from the widely used edit distance and finite state machine methods. A close relationship with the earlier related `proximity comparison' method is established.