Oblikovanje sažetih prikaza povezanosti elemenata u transakcijskim podacima

doctoral thesis

• Title (english): Designing concise representation of correlations among elements in transactional data
• Author: Mihaela Vranić
• Mentor: Zoran Skočir (mentor)
• Committee member: Dragan Gamberger (predsjednik povjerenstva)
• Committee member: Boris Vrdoljak (član povjerenstva)
• Committee member: Bruno Blašković (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; (Department of Electrical Engineering Fundamentals and Measurements); Zagreb
• Defense date and country: 2011-11-04, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing
• Universal decimal classification (UDC): 004 - Computer science and technology. Computing. Data processing

Abstract

U doktorskom radu je obrađena problematika sažetih prikaza povezanosti transakcijskih elemenata. U svrhu dobivanja sažetih prikaza povezanosti korištene su dvije sasvim različite poznate deskriptivne metode dubinske analize podataka: stvaranje asocijacijskih pravila i hijerarhijsko grupiranje. Metoda stvaranja asocijacijskih pravila je iskorištena za izoliranje zatvorenih čestih kolekcija elemenata. Spomenute kolekcije se povezuju u stablastu strukturu pomoću dvije razvijene strategije koje su i implementirane. Da bi gubitak informacije u prikazanoj strukturi bio minimalan, napravljena je poveznica s teorijom grafova te je iskorišten algoritam za razapinjanje minimalnog stabla. S druge strane problemu se prišlo razradom različitih mjera zanimljivosti vezanih uz povezivanje elemenata u transakcijama. Ove su mjere prilagođene hijerarhijskom grupiranju te je za njihovo testiranje razvijena i programska podrška koja uz korištenje različitih mjera omogućuje i obrezivanje na osnovu podrške. Dva navedena pristupa su testirana nad umjetno konstruiranim podacima koji bi trebali analitičaru omogućiti lakši odabir pogodne mjere udaljenosti te nad referentnim skupovima podataka. Tu su se pokazale vrlo dobrima za jednostavan i brzi uvid u odnose među elementima transakcija. Razvijene metode su također korištene nad podacima vezanim uz visokoškolsko obrazovanje te pokazale potencijal za sažeti prikaz odnosa među elementima ispita – među pitanjima i uz njih vezanih koncepata

Abstract (english)

This doctoral thesis proposes new approaches in designing concise representation of correlations among elements in transactional data. For this purpose two different known data mining methods are used: association rules generation and hierarchical clustering. Method of association rules generation is used to isolate closed frequent itemsets which are further connected to form a tree structure. For this purpose two strategies are developed and implemented. To minimize information loss, graph theory was used along with its existing algorithms for minimal spanning tree creation. Additionally, different interest measures that demonstrate certain relationships between transaction elements were modulated to serve in hierarchical clustering. Software implementation of these measures along with support based pruning ability is also made. Two different approaches are tested both on artificial datasets constructed to serve analysts in selecting the most appropriate measure as well as on known and easily reachable reference datasets. They proved to be very useable as descriptive methods that provide quick and easily interpretable structures that represent relationships between transactional elements. Developed methods were also used on the real-life educational data and showed great potential in displaying relationships between test elements – questions and related concepts.

• Keywords: transakcijski podaci; asocijacijska pravila; česte kolekcije elemenata; stablaste strukture; hijerarhijsko grupiranje; mjere udaljenosti; mjere zanimljivosti; dubinska analiza podataka u edukaciji
• Keywords (english): transactional data; association rules; frequent itemsets; tree structures; hierarchical clustering; distance measures; interest measures; educational data mining
• Language: croatian
• URN:NBN: urn:nbn:hr:168:528568
• Project: Number: 036-0362027-1638; Title: Umrežena ekonomija; Leader: Zoran Skočir; Jurisdiction: Croatia; Funder: MZOS; Funding stream: ZP
• Study programme: Title: Computer Science; Study programme type: university; Study level: postgraduate; Academic / professional title: Doktor znanosti (Doktor znanosti)
• Catalog URL: http://lib.fer.hr/cgi-bin/koha/opac-detail.pl?biblionumber=37185
• Type of resource: Text
• Extent: 177 str. ; 30 cm
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2019-07-10 07:52:40