Técnicas avançadas para análise visual de redes temporais

Abstract

Temporal networks represent interactions among entities of a given domain with additional information about when such interactions occur. The visualization of temporal networks plays a key role in the recognition of properties that would be difficult to perceive without an adequate visualization strategy. Due to a large amount of information provided by these networks, more attention has been given to issues related to the visual scalability associated with the produced layouts, but this still represents an unsolved problem and lacks effective techniques. We propose in this thesis novel techniques to enhance the visualization of temporal networks. Specifically, a scalable node reordering technique for temporal network visualization, named Community-based Node Ordering (CNO), combining static community detection with node reordering techniques, along with a taxonomy to categorize activity patterns. In addition, a visualization method that allows the comparison of two community detection algorithms is presented in order to decide which one is better for visual analysis of communities. Another contribution is the analysis of dynamic processes, as spreading rumors, diseases, applied in the visualization of temporal networks. Furthermore, we conducted a user experiment consisting of the application of different tasks in temporal networks, in order to find the relation of the layouts with the most appropriate tasks. Finally, the Dynamic Network Visualization (DyNetVis) system demonstrates the software specifications, examples, functionalities, and impact in the study field. We performed experiments with qualitative and quantitative analyses using real networks in several fields to show that the proposed layouts and categorization helped in the identification of patterns that would otherwise be difficult to see.