Integracija podatkovnih shem na osnovi analize podatkov z algoritmom arhetipske analize za povzemanje podatkovnih množic

magistrsko delo

Anton Zvonko Gazvoda (Author), Matjaž B. Jurič (Mentor)

Abstract

Ključna aktivnost v procesu integracije aplikacij na podatkovnem nivoju je iskanje preslikav med podatkovnimi shemami, kar je osnova za izvedbo ustreznih transformacij podatkov. V ta namen predlagamo novo metodo za integriranje shem, ki deluje na osnovi ocenjevanja podobnosti med podatkovnimi instancami. Metoda temelji na arhetipski analizi, s katero generiramo povzetke podatkov elementov sheme. Njihove približke opišemo s konveksnimi ovojnicami. Za izračun povzetkov definiramo različne pristope za transformacijo podatkov v vektorski prostor in metrike podobnosti. Preslikave iščemo s pomočjo dveh algoritmov za odkrivanje enostavnih in kompleksnih preslikav. Metodo smo ovrednotili na testnih podatkih, ki vključujejo pravilne preslikave med shemami, in jo primerjali z iskalnikom preslikav COMA CE. Uspešnost smo ocenili z občutljivostjo (91%), specifičnostjo (75%), točnostjo (87%) in natančnostjo (91%), pri čemer je naša metoda v povprečju za 20% boljša od COMA CE.

Keywords

integracija podatkovnih shem na osnovi instanc;iskanje preslikav;arhetipska analiza;konveksna ovojnica;povzetek podatkov;računalništvo;računalništvo in informatika;magisteriji;

Data

Language:	Slovenian
Year of publishing:	2014
Typology:	2.09 - Master's Thesis
Organization:	UL FRI - Faculty of Computer and Information Science
Publisher:	[A. Z. Gazvoda]
UDC:	004.6(043.2)
COBISS:	1536018115
Views:	1108
Downloads:	219
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	Data schemes integration with algorithms for data summarization via archetypal analysis
Secondary abstract:	Schema mapping discovery is key activity while performing data-level integration process and represents the basis for proper data transformation. For this purpose, we introduce novel instance-based schema matching method by using archetypal analysis in order to generate data summary for each schema element. Summary approximations are represented by convex hulls. We define several approaches for data transformation to vector space, as well as summary-similarity metrics. Two algorithms were developed in order to determine simple and complex matches. Our method was evaluated on the test data including proper mappings between schemas and compared with COMA CE schema matcher. Efficiency of our method was evaluated with sensitivity (91%), specificity (75%), accuracy (87%) and precision (91%). Compared with COMA CE, our method performs on average 20% better.
Secondary keywords:	instance-based schema matching;schema mapping;archetypal analysis;convex hull;data summary;computer science;computer and information science;master's degree;
File type:	application/pdf
Type (COBISS):	Master's thesis/paper
Study programme:	1000471
Thesis comment:	Univ. v Ljubljani, Fak. za računalništvo in informatiko
Pages:	91 str.
ID:	8739484