Document Retrieval Hacks

Abstract

Given a collection of strings, document listing refers to the problem of finding all the strings (or documents) where a given query string (or pattern) appears. Index data structures that support efficient document listing for string collections have been the focus of intense research in the last decade, with dozens of papers published describing exotic and elegant compressed data structures. The problem is now quite well understood in theory and many of the solutions have been implemented and evaluated experimentally. A particular recent focus has been on highly repetitive document collections, which have become prevalent in many areas (such as version control systems and genomics – to name just two very different sources). The aim of this paper is to describe simple and efficient document listing algorithms that can be used in combination with more sophisticated techniques, or as baselines against which the performance of new document listing indexes can be measured. Our approaches are based on simple combinations of scanning and hashing, which we show to combine very well with dictionary compression to achieve small space usage. Our experiments show these methods to be often much faster and less space consuming than the best specialized indexes for the problem.