Event Detection in ARTigo Data

In the last decade the ARTigo platform collected descriptive tags for historic artworks through the help of different games with a purpose (GWAP). Using the additional provided creation date of the artworks, the goal of this thesis is to detect events relevant in an art historic context. To achieve this goal the relative amount of distinct tags is analyzed over time. Locating the events is accomplished with change point detection. This thesis discusses the necessary steps to format the collected data into an input format suited for change point detection algorithms. The main focus lies on the CUSUM change point detection algorithm and off-line change point detection. Furthermore the assumption of a normal and log-normal distribution as well as the influence of variance is evaluated. The results are compared to the trend detection algorithm SigniTrend. In the end art historians get the option to analyze the data themselves. For that the used algorithms are implemented into the ELKI framework and a new component is provided in the ARTigo Analytics Center.