Clustering and Analysis of User Behaviors utilizing a Graph Edit Distance Metric.

Jürgensen, Lars (2019) Clustering and Analysis of User Behaviors utilizing a Graph Edit Distance Metric. (Bachelor thesis), Kiel University, Kiel, 48 pp.

[thumbnail of thesis.pdf]
thesis.pdf - Published Version

Download (654kB) | Preview


Understanding the utilization of software systems by users is helpful for utilization forecasts used in performance optimization and improving user experience. Therefore, it is necessary to understand the utilization and be able to identify typical user behaviors. One possibility to find typical behaviors is cluster analysis. Various clustering algorithms, like X-Means, EMClustering and BIRCH, were implemented in iObserve. The user actions were transformed into vectors, which were then compared using the Euclidean distance metric. Evaluation showed that these approaches were not able to find all expected behaviors in simple
We propose an approach, which models the user behaviors as graphs. The Behavior Models are compared using a Graph Edit Distance based distance function. We utilize the M-Tree index structure to make nearest neighbor searches more efficient. The OPTICS algorithm is then utilized to identify the clustering structure of the behaviors.
We implemented our approach in the iObserve project using the pipe and filter architecture. The implementation is evaluated by clustering fixed and randomized workloads of the JPetStore and comparing the clustering results with manually created expected results. The evaluation shows it is possible to detect all expected behaviors in both fixed and randomized workloads.

Document Type: Thesis (Bachelor thesis)
Keywords: Clustering, User Behavior
Research affiliation: Kiel University > Software Engineering
Projects: iObserve
Date Deposited: 11 Nov 2019 12:43
Last Modified: 11 Nov 2019 12:43

Actions (login required)

View Item View Item