Bottom-k document retrieval

Abstract

We consider the problem of retrieving the k documents from a collection of strings where a given pattern P appears least often. This has potential applications in data mining, bioinformatics, security, and big data. We show that adapting the classical linear-space solutions for this problem is trivial, but the compressed-space solutions are not easy to extend. We design a new solution for this problem that matches the best-known result when using $2|\mathsf{CSA}|+o(n)$ bits, where $\mathsf{CSA}$ is a Compressed Suffix Array. Our structure answers queries in the time needed by the $\mathsf{CSA}$ to find the suffix array interval of the pattern plus $O(k\lg k{\lg }^{ϵ}n)$ accesses to suffix array cells, for any constant $ϵ>0$.