| Resumo: | Static aggregations of network activity can unravel attributes of the complex systems they represent. However, they fall short when the structure of the systems changes over time. In some cases, changes are sluggish, such as in power grids, where lines enjoy a lengthy temporal permanence. In others, a high frequency of change is observed, such as on a network of online messages, social contacts, pathogen transmission or ball passing in a soccer game. In these cases, reducing what is inherently a temporal network to a static one, leads necessarily to a loss of information, such as causal relationships, precedence or reachability rules. Temporal networks are thus the main subject of this thesis, centered on the study of network evolution from the point of view of its clusters as significant meso-structures. The thesis has two interrelated parts. In the first, theoretical challenges are addressed and original algorithms, methods and tools are developed that can further the study of network theory. In the second, these developments are applied to the analysis of team invasion sports. A measurement of game dynamics was created based on a temporal network representation of a match, with nodes clustered by spatial proximity. These measurements were found to correlate with match events of known dynamics. Moreover, they reveal unique, multi-level, aspects of the game, from the individual players contributions, to the clusters of interacting players, to their teams and their matches, which is useful for game analysis, training and strategy development.
|
|---|