[Back]

B. Alsallakh:
"Interactive Visual Analysis of Relational Data and Applications in Event-Based Business Analytics";
Supervisor: E. Gröller, M. Suntinger; Institut für Computergraphik und Algorithmen, TU Wien, 2009; final examination: 2009-07-20.

In this work, a framework for interactive visual analysis of attributed graphs has been developed. An attributed graph is an extension of the standard graph of a binary relation, which attaches a set of attributes to the nodes and edges. The implemented visual analysis techniques aim at the local level at enabling an intuitive navigation in the graph which reveals both the structure of the selected part of the graph and the attributes of the nodes and edges in this part. At the global level these techniques aim at understanding the distributions of the attributes in the graph as a whole or in specific parts in it and at spotting meaningful associations between the attributes and the relations. The work presents several extensions to the attributes such as graph‐theoretic features, values aggregated over the relations, and hierarchical grouping. All attributes are treated in a unified manner which helps performing elaborate analysis tasks using the existing tools. Additionally, novel graph drawing techniques are proposed. They are designed to understand attribute distributions and associations in the graph. These techniques can be additionally used to visualize results of queries in the data, which can be also visually defined using the attribute analysis tools. Finally, the work addresses several types of association analysis in relational data, along with visual analysis methods for them. It presents a perceptual enhancement for the well‐known parallel sets technique for association analysis in categorical data, and proposes extensions for employing it in relational data. Also, novels methods for other types of association analysis are introduced. The relational data in this work were defined upon typed events in an event‐based system, which offers a flexible architecture for real‐time analysis. Nevertheless, the presented analysis methods are generic and have been tested on two real‐world datasets. In the first dataset, entities for customers and products are derived from the purchase events, and various meaningful associations were found between the attributes and the relation (for example, which types of products the female customers bought more frequently, or at which age customers have higher interest for books). In the second dataset, events in an issue‐tracking system are analyzed to find out ticket assignment patterns and forwarding patterns between the support offices.

http://publik.tuwien.ac.at/files/PubDat_180432.pdf