Full-text and Keyword Indexes for String Searching

String searching consists in locating a substring in a longer text, and two strings can be approximately equal (various similarity measures such as the Hamming distance exist). Strings can be defined very broadly, and they usually contain natural language and biological data (DNA, proteins), but they can also represent other kinds of data such as music or images. One solution to string searching is to use online algorithms which do not preprocess the input text, however, this is often infeasible due to the massive sizes of modern data sets. Alternatively, one can build an index, i.e. a data structure which aims to speed up string matching queries. The indexes are divided into full-text ones which operate on the whole input text and can answer arbitrary queries and keyword indexes which store a dictionary of individual words. In this work, we present a literature review for both index categories as well as our contributions (which are mostly practice-oriented). The first contribution is the FM-bloated index, which is a modification of the well-known FM-index (a compressed, full-text index) that trades space for speed. In our approach, the count table and the occurrence lists store information about selected q-grams in addition to the individual characters. Two variants are described, namely one using O(n log n) bits of space with O(m + logm log logn) average query time, and one with linear space and O(m log log n) average query time, where n is the input text length and m is the pattern length. We experimentally show that a significant speedup can be achieved by operating on q-grams (albeit at the cost of very high space requirements, hence the name “bloated”). In the category of keyword indexes we present the so-called split index, which can efficiently solve the k-mismatches problem, especially for 1 error. Our implementation in the C++ language is focused mostly on data compaction, which is beneficial for the search speed (by being cache friendly). We compare our solution with other algorithms and we show that it is faster when the Hamming distance is used. Query times in the order of 1 microsecond were reported for one mismatch for a few-megabyte natural language dictionary on a medium-end PC. A minor contribution includes string sketches which aim to speed up approximate string comparison at the cost of additional space (O(1) per string). They can be used in the context of keyword indexes in order to deduce that two strings differ by at least k mismatches with the use of fast bitwise operations rather than an explicit verification.