On the effectiveness of user manipulation in multidimensional projections

Abstract

With the advent of interactive techniques for multidimensional data visualization using dimensionality reduction, new possibilities of dealing with data complexity were introduced. The idea behind those methods is to allow the user to manually modify the mapping of a subset of the data, steering the mapping of the whole data set. Previous studies suggest that user manipulation is beneficial to the visualization as a means of effectively modifying the final result. Still, those studies focus only on the grouping of instances, such as group separation and compactness. This paper proposes a new view on effectiveness of user manipulation based on well-known evaluation measures. In addition, it provides initial experimental evidence on the effectiveness of user manipulation on state of the art interactive methods. Although the manipulation does affect results, we conclude that it does not lead to definitive improvements.