Please use this identifier to cite or link to this item:
|Title:||Quantitative Evaluation of Generalizations|
|Authors:||Mohamed, Habab Musa|
|Publisher:||The British University in Dubai (BUiD)|
|Abstract:||Inspired by the explosive growth of complex networks and the extraction of common patterns from varied complex networks' features, mining and analyzing networks have become a recent eld of signi cant interest for many researchers with the primary focus on network measures. The relative ease of computation of unweighted measures leads them to be widely used in analyzing real world networks, although they ignore important network information: the weights. Despite many real world networks arise in the form of weighted networks, a few number of network measures take the weights into account. From this prospective, the last few years have witnessed the attempts of some researchers to generalize di erent unweighted network measures. With several possible generalizations for di erent measures, the issue of evaluating these generalizations and quantifying their e ectiveness becomes increasingly important. Up until now, such generalizations comparison relied primarily on visual inspection of different plots and informal articulation on how a particular generalization is more informative than the original unweighted measure. In this thesis, we provide a comparative automated methodology for quantitative evaluation of di erent generalizations of unweighted degree measure. We conduct a comparative study between two state-of-art generalizations, the unweighted degree generalization based on elective cardinality  and the α-degree generalization , based on the quantitative evaluation of their productive power of classifying networked nodes. We show that some generalizations of unweighted degree measure outperform other generalizations and even the original degree measure. We study the elect of the type of the network involved and classier used on the e ectiveness of generalizations.|
|Description:||DISSERTATION WITH DISTINCTION|
|Appears in Collections:||Dissertations for Informatics (Knowledge and Data Management)|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.