Succinct Data Structures

by Gupta, Ankur

Abstract (Summary)
The world is drowning in data. The recent explosion of web publishing, XML data, bioinformation, scientific data, image data, geographical map data, and even email communications has put a strain on our ability to manage the information contained there. In general, the influx of massive data sets for all kinds of data present a number of difficulties with storage, organization of information, and data accessibility. A primary computing challenge in these cases is how to compress the data but still allow it to be queried quickly.In real-life situations, many instances of data are highly compressible, presenting a major opportunity for space savings. In mobile applications, such savings are critical, since space and the power to access information are at a premium. In a streaming environment, where new data are being generated constantly, compression can aid in prediction as well. In the case of bioinformatics, understanding succinct representations of DNA sequences could lead to a more fundamental understanding of the nature of our own "data stream," perhaps even giving hints on secondary and tertiary structure, gene evolution, and other important topics.In this thesis, we focus our attention on the important problem of compressed text indexing<\i>, where the goal is to compress a text document and allow arbitrary searching for patterns in the best possible time without first decompressing the text<\i>. We develop a number of compressed data structures that either solve this problem directly, or are used as smaller components of an overall text indexing solution. Each component has a number of applications beyond text indexing as well. For each structure, we provide a theoretical study of its space usage and query performance on a suite of operations crucial to access the stored data. In each case, we relate its space usage to the compressed size of the original data and show that the supported operations function in near-optimal or optimal time. We also present a number of experimental results that validate our theoretical findings, showing that our methodology is competitive with the state-of-the-art.
Bibliographical Information:

Advisor:Vitter, Jeffrey S

School:Duke University

School Location:USA - North Carolina

Source Type:Master's Thesis

Keywords:computer science compression text indexing dictionaries dynamic optimal


Date of Publication:12/14/2007

© 2009 All Rights Reserved.