PAPERS
THE SIXTH INTERNATIONAL CONFERENCE ON FORENSIC COMPUTER SCIENCE
Print ISBN 978-85-65069-07-6 - Online ISBN 978-85-65069-05-2, pp 106-114
DOI: 10.5769/C2011011 and http://dx.doi.org/10.5769/C2011011
Computer Forensic Document Clustering with ART1 Neural Networks
By Georger Rommel Ferreira de Araújo, and Célia Ghedini Ralha
To download this paper, click here.
Print ISBN 978-85-65069-07-6 - Online ISBN 978-85-65069-05-2, pp 106-114
DOI: 10.5769/C2011011 and http://dx.doi.org/10.5769/C2011011
Computer Forensic Document Clustering with ART1 Neural Networks
By Georger Rommel Ferreira de Araújo, and Célia Ghedini Ralha
To download this paper, click here.
ABSTRACT
Computer forensic text corpora are usually very heterogeneous. While classification, by file type or other criteria, should be an aid in the exploration of such corpora, it does not help in the task of thematically grouping together documents. Adaptive Resonance Theory (ART) describes a number of self-organizing artificial neural networks that employ an unsupervised learning process and is specially designed to learn new patterns without forgetting what it has already learned, overcoming the important restriction defined by the stability/plasticity dilemma. In this direction, this paper applies the ART1 algorithm (ART with binary input vectors) to thematically cluster documents returned from query tools used with forensic text corpora. Documents that would previously be presented in a disorganized and often long list are thematically clustered, giving the examiner a faster and effective way of obtaining a general picture of document content during forensic examinations. Our experimental results are expressive to validate our approach, achieving high agreement between the clustering solution processed with our software package and the gold standard defined by domain area experts.
KEYWORDS
ART1; artificial neural networks; computer forensics; document clustering
Computer forensic text corpora are usually very heterogeneous. While classification, by file type or other criteria, should be an aid in the exploration of such corpora, it does not help in the task of thematically grouping together documents. Adaptive Resonance Theory (ART) describes a number of self-organizing artificial neural networks that employ an unsupervised learning process and is specially designed to learn new patterns without forgetting what it has already learned, overcoming the important restriction defined by the stability/plasticity dilemma. In this direction, this paper applies the ART1 algorithm (ART with binary input vectors) to thematically cluster documents returned from query tools used with forensic text corpora. Documents that would previously be presented in a disorganized and often long list are thematically clustered, giving the examiner a faster and effective way of obtaining a general picture of document content during forensic examinations. Our experimental results are expressive to validate our approach, achieving high agreement between the clustering solution processed with our software package and the gold standard defined by domain area experts.
KEYWORDS
ART1; artificial neural networks; computer forensics; document clustering
To return to the "Published Papers" main page, click here.