Designing Sketches for Similarity Filtering

D 2016

Designing Sketches for Similarity Filtering

MÍČ, Vladimír, David NOVÁK and Pavel ZEZULA

Basic information

Original name

Designing Sketches for Similarity Filtering

Authors

MÍČ, Vladimír (203 Czech Republic, guarantor, belonging to the institution), David NOVÁK (203 Czech Republic, belonging to the institution) and Pavel ZEZULA (203 Czech Republic, belonging to the institution)

Edition

USA, 2016 IEEE 16th International Conference on Data Mining Workshops (ICDMW), p. 655-662, 8 pp. 2016

Publisher

IEEE

Other information

Language

English

Type of outcome

Stať ve sborníku

Field of Study

10201 Computer sciences, information science, bioinformatics

Country of publisher

United States of America

Confidentiality degree

není předmětem státního či obchodního tajemství

Publication form

printed version "print"

References:

URL

RIV identification code

RIV/00216224:14330/16:00088645

Organization unit

Faculty of Informatics

ISBN

978-1-5090-5472-5

ISSN

DOI

http://dx.doi.org/10.1109/ICDMW.2016.0098

UT WoS

000401906900090

Keywords in English

Algorithm;Similarity search;Similarity filtering;Bit strings;Sketches;Hamming distance

Abstract

V originále

Abstract: The amounts of currently produced data emphasize the importance of techniques for efficient data processing. Searching big data collections according to similarity of data well corresponds to human perception. This paper is focused on similarity search using the concept of sketches – a compact bit string representations of data objects compared by Hamming distance, which can be used for filtering big datasets. The object-to-sketch transformation is a form of the dimensionality reduction and thus there are two basic contradictory requirements: (1) The length of the sketches should be small for efficient manipulation, but (2) longer sketches retain more information about the data objects. First, we study various sketching methods for data modeled by metric space and we analyse their quality. Specifically, we study importance of several sketch properties for similarity search and we propose a high quality sketching technique. Further, we focus on the length of sketches by studying mutual influence of sketch properties such as correlation of their bits and the intrinsic dimensionality of a set of sketches. The outcome is an equation that allows us to estimate a suitable length of sketches for an arbitrary given dataset. Finally, we empirically verify proposed approach on two real-life datasets.

Links

GBP103/12/G084, research and development project

Name: Centrum pro multi-modální interpretaci dat velkého rozsahu

Investor: Czech Science Foundation

Citovat

MÍČ, Vladimír, David NOVÁK and Pavel ZEZULA. Designing Sketches for Similarity Filtering. In Carlotta Domeniconi, Francesco Gullo, Francesco Bonchi, Josep Domingo-Ferrer, Ricardo Baeza-Yates, Zhi-Hua Zhou, Xindong Wu. 2016 IEEE 16th International Conference on Data Mining Workshops (ICDMW). USA: IEEE, 2016, p. 655-662. ISBN 978-1-5090-5472-5. Available from: https://dx.doi.org/10.1109/ICDMW.2016.0098.

@inproceedings{1371735,
   author = {Míč, Vladimír and Novák, David and Zezula, Pavel},
   address = {USA},
   booktitle = {2016 IEEE 16th International Conference on Data Mining Workshops (ICDMW)},
   doi = {http://dx.doi.org/10.1109/ICDMW.2016.0098},
   editor = {Carlotta Domeniconi, Francesco Gullo, Francesco Bonchi, Josep Domingo-Ferrer, Ricardo Baeza-Yates, Zhi-Hua Zhou, Xindong Wu},
   keywords = {Algorithm;Similarity search;Similarity filtering;Bit strings;Sketches;Hamming distance},
   howpublished = {tištěná verze "print"},
   language = {eng},
   location = {USA},
   isbn = {978-1-5090-5472-5},
   pages = {655-662},
   publisher = {IEEE},
   title = {Designing Sketches for Similarity Filtering},
   url = {http://ieeexplore.ieee.org/document/7836729/},
   year = {2016}
}

TY  - JOUR
ID  - 1371735
AU  - Míč, Vladimír - Novák, David - Zezula, Pavel
PY  - 2016
TI  - Designing Sketches for Similarity Filtering
PB  - IEEE
CY  - USA
SN  - 9781509054725
KW  - Algorithm;Similarity search;Similarity filtering;Bit strings;Sketches;Hamming distance
UR  - http://ieeexplore.ieee.org/document/7836729/
N2  - Abstract: The amounts of currently produced data emphasize the importance of techniques for efficient data processing. Searching big data collections according to similarity of data well corresponds to human perception. This paper is focused on similarity search using the concept of sketches – a compact bit string representations of data objects compared by Hamming distance, which can be used for filtering big datasets. The object-to-sketch transformation is a form of the dimensionality reduction and thus there are two basic contradictory requirements: (1) The length of the sketches should be small for efficient manipulation, but (2) longer sketches retain more information about the data objects. First, we study various sketching methods for data modeled by metric space and we analyse their quality. Specifically, we study importance of several sketch properties for similarity search and we propose a high quality sketching technique. Further, we focus on the length of sketches by studying mutual influence of sketch properties such as correlation of their bits and the intrinsic dimensionality of a set of sketches. The outcome is an equation that allows us to estimate a suitable length of sketches for an arbitrary given dataset. Finally, we empirically verify proposed approach on two real-life datasets.
ER  -

MÍČ, Vladimír, David NOVÁK and Pavel ZEZULA. Designing Sketches for Similarity Filtering. In Carlotta Domeniconi, Francesco Gullo, Francesco Bonchi, Josep Domingo-Ferrer, Ricardo Baeza-Yates, Zhi-Hua Zhou, Xindong Wu. \textit{2016 IEEE 16th International Conference on Data Mining Workshops (ICDMW)}. USA: IEEE, 2016, p.~655-662. ISBN~978-1-5090-5472-5. Available from: https://dx.doi.org/10.1109/ICDMW.2016.0098.

Detailed Information on Publication Record