Lan Žagar (2008) Data representation and mining using multi-layered networks. EngD thesis.
Abstract
WE PRESENT A NEW TECHNIQUE FOR NETWORK VISUALIZATION AND NETWORK-BASED DATA MINING. STANDARD NETWORK VISUALIZATION TECHNIQUES MOST OFTEN FOCUS ON A SINGLE-TYPE RELATIONS AND ARE USED FOR VISUALIZATION OF A SINGLE DATA SET. IN PRACTICAL PROBLEM SOLVING, HOWEVER, ADDITIONAL DATA SETS AND RELATIONS THAT RELATE THEM ARE AVAILABLE. OUR SPECIFIC GOAL IN THIS THESIS WAS TO ADDRESS THE PROBLEM OF VISUALIZATION OF MULTIPLE DATASETS FROM A RELATIONAL DATABASE. OUR PROPOSED APPROACH IS BASED ON MULTI-LAYER NETWORKS. IN THIS STUDY WE USE ONLY TWO LAYERS REPRESENTING TWO DIFFERENT DATASETS. A METHOD FOR OPTIMIZING THE LAYOUT OF A MULTI-LAYER NETWORK WAS PROPOSED. SEVERAL OBJECTIVE CRITERIA FOR EVALUATION OF NETWORK VISUALIZATIONS WERE ALSO DEVELOPED. SIMULATIONS ON SYNTHETIC DATA SETS SHOWED THAT THE PROPOSED OPTIMIZATION TECHNIQUE PERFORMS WELL IN SIMULTANEOUS OPTIMIZATION OF TWO-LAYERED NETWORK WITH RESPECT TO THE STRUCTURE OF BOTH LAYERS. WE HAVE ALSO STUDIED THE PERFORMANCE OF THE TECHNIQUE IN BIOINFORMATICAL APPLICATION, WHERE A GENE NETWORK WAS SUCCESSFULLY COMPLEMENTED WITH A NETWORK OF MESH TERMS RESULTING IN AN INFORMATIVE TWO-LAYER NETWORK. AFTER THE OPTIMIZATION STEP, SEVERAL MESH TERMS WERE PLACED NEAR RELATED GENE CLUSTERS AND THUS PROVIDED ADDITIONAL INSIGHT INTO THE IDENTIFIED GENE SETS.
Actions (login required)