Browsing by Author "Hermann, Matthias"

Filter results by typing the first few letters
Now showing 1 - 2 of 2
  • Thumbnail Image
    ItemOpen Access
    Einsatz von Machine-Learning-Methoden zur adaptiven Darstellung von Software-Metriken
    (2017) Hermann, Matthias
    Auf manchen SonarQube-Instanzen wird die verfügbare Fläche der Webseite nicht effizient genutzt und große Teile der Seite enthalten Leerflächen. Damit diese Flächen genutzt werden können, um genau die Informationen darzustellen, weswegen der Benutzer die Webseite aufgerufen hat, wurde im Rahmen dieser Arbeit mit DeepSonar eine adaptive Benutzeroberfläche für die Codeanalyse-Plattform SonarQube entwickelt. Diese erlernt mittels Machine-Learning die für den aktuellen Benutzer und Nutzungskontext relevantesten Informationen, d. h. die aus einer Programmcodeanalyse resultierenden Software-Metriken. Anhand der Ergebnisse des Machine-Learnings wird die Weboberfläche von SonarQube angepasst, sodass diese Metriken in der davor ungenutzten Fläche auf der Startseite angezeigt werden.
  • Thumbnail Image
    ItemOpen Access
    Graph sparsification techniques for triangle counting
    (2020) Hermann, Matthias
    The triangle count of a graph is a key metric in graph analysis. Especially for social networks, the triangle count is important to assess connectedness of vertices in the graph. However, these social networks in particular can produce large graphs with trillions of edges. In fact, the size of graphs appears to grow faster than the computational resources to analyze and process these graphs. Confronted with similar problems in the past, the solutions for developers of algorithms were oftentimes approximating algorithms. Research in approximate triangle counting algorithms has led to a multitude of various approximating algorithms. Comparing and understanding the differences in the mechanisms they use to provide faster and more accurate results has therefore become complicated. This work presents an analysis on existing triangle counting algorithms to improve understanding of which mechanisms work best for fast triangle count approximations. In order to further this understanding even more, an analysis on graph structures and their influence on triangle counts is presented, as well. Results of this analysis include a method for decentralized coordination and reducing communication in distributed computations as well as a method for estimating a triangle count of a graph by using a small sample of vertices and their degree values.