Graph Edit Distance applied to diverse frameworks: Learning, matching and exploring techniques

Identificador:  TDX:4001

Handle: https://hdl.handle.net/20.500.11797/TDX4001

Autores:  Rica Alarcón, María Elena

Resumen:
Graphs are mathematical objects that depict the abstract representation of data when relations between elements are defined. When data is represented with graphs, nodes represent the main objects of the data and edges represent the relations between them. In this context, specific machine learning techniques need to be defined or adapted to obtain information from the data and predict characteristics or features that could be of interest for some applications. During the last 40 years, researchers have been analysing how to represent data with graphs and how to adapt machine learning methods to these structures or define new ones adapted to this framework. With this aim, the concept of Graph Edit Distance (GED) has been used during decades and several machine learning techniques use the GED as measure of dissimilarity between graphs to tackle the solution of different problems. This thesis presents a compendium of machine learning methods focused on graph-represented data. Diverse tasks have been faced representing the data as graphs and the majority of the presented methods make use of the concept of GED as a tool to analyse the structure of the data. The techniques developed on this thesis demonstrate that the graph representation of data is suitable to solve different situations and can be explored for future contexts.