Needles and Haystacks: a search engine for personal information collections

Information retrieval systems can be partitioned into two main classes: large-scale systems that make use of an inverted index or some other auxiliary data structure, intended for massive volumes of data; and the small-scale systems based upon sequential pattern matching that most computer users employ when hunting for missing email and news items. In this paper we describe a hybrid approach that offers the ranked queries and similarity matching of a genuine information retrieval system, but does so without any need for an index to be precomputed. This software tool, which we call seft, offers performance that in a retrieval effectiveness sense matches conventional information retrieval systems, and in a resource efficiency sense, while considerably slower than grep-like tools, is fast enough to be useful on hundreds of megabytes of text.