A similarity-based approach to generate edge bundles

• Ano de defesa: 2016
• Autor(a) principal: Sikansi, Fábio Henrique Gomes
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Agrupamento visual de arestas; Edge bundling; Graph visualization; Information visualization; Visualização de grafos; Visualização de informação
• Link de acesso: http://www.teses.usp.br/teses/disponiveis/55/55134/tde-27032017-083934/

Graphs have been successfully employed in avariety of problems and applications, being the object of study in modeling, analysis and construction of visual representations. While different approaches exist for graph visualization,most of them suffer from the severe clutter when the number of nodes or edges is large. Among the approaches that handle such problem, edge bundling techniques attained relative success on improving the quality of the visual representations by bending and aggregating edges in order to produce an organized layout. Despite this success, most of the exiting techniques create edge bundles based only on the visual space information, that is, there is no explicit connection between the edge bundling layout and the original data. There fore, these techniques generates less meaningful bundles and may lead users to misinterpret the data. This masters research presents a novel edge bundling technique based on the similarity relationships among vertices. We developed such technique based on two assumptions. First, it supports the hypothesis that edge bundling can better represent the data when there is an inherent connection between the proximity among the elements in the information space and the proximity between edges in the edge bundling layout. We address this question by presenting a similarity bundling framework, that considers the similarity between vertices when performing the edges bending. To guide the bundling, we create a similarity hierarchy, called backbone. This is based on a multilevel partition of the data, which groups edges of similar vertices. Second, we also support that a multiscale representation improves the visual and complexity scalability of bundling layouts. We present a multiscale edge bundling, which allows an overview plus detailed exploration, coarsening or revealing the bundling at different levelsof the same visualization. Our evaluation framework shows that our backbone produces a balanced hierarchy with a good representation of similarity relationships among vertices. Moreover, the edge bundling layout guided by the backbone reduces the visual clutter and surpass state-of-the-art techniques in displaying global and local edge patterns.