A Sequential Recursive Implementation of Dead-Zone Single Keyword Pattern Matching

Abstract

Earlier publications provided an abstract specification of a family of single keyword pattern matching algorithms [18] which search unexamined portions of the text in a divide-and-conquer fashion, generating dead-zones in the text as they progress. These dead zones are area of text that require no further examination. Here the results are described of implementing in C++ a sequential recursive version of the algorithm family, where all instances of a single keyword p in a text S are sought—the online keyword matching problem where S may not be precomputed.

We show that each step may involve a window shift of up to $2 \times |p|\index{set cardinality}-1$ characters—almost twice as much (and therefore potentially almost twice as fast) as the maximum of $|p|\index{set cardinality}$ characters possible with the Boyer-Moore family of algorithms. Our counterintuitive improvement over Boyer-Moore algorithms is achieved by simultaneously shifting left and right. Ongoing benchmarking shows [12] that such bidirectional shifts are highly efficient—and we make specific comparisons here to Horspool’s algorithm [9], regarded as one of the most efficient algorithms of the Boyer-Moore family.