Limitations

As stated at the beginning, we were inspired by time series data for a daily monitoring task.

Especially CLO and STA with their 24 hour clock metaphor profit from this data arrangement.

The performance may change with different lengths of time series.

The same is true for the aspect ratio and the size of the single glyphs. The aspect ratio was chosen in order not to greatly disadvantage the circular designs in terms of display space used.

However, especially STR would profit from an aspect ratio with more horizontal space. With varying sizes of glyphs, the performance of the designs could change. In our setting we used the minimal space possible to be able to assign one pixel to one data value for the higher data density.

3.3.6 Conclusion

The goal of this experiment was to compare the performance of the clock glyph against well-established alternative data glyph designs. Therefore, we quantitatively measured accuracy and efficiency, and qualitatively surveyed user confidence and preferences for four glyph types based on three tasks important to our domain experts: peak detection, peak detection at a certain point in time, and trend detection. The results show that depending on tasks and data density, the chosen

glyphs performed differently. We show that the line glyph is generally a good choice for peak and trend detection tasks but that radial encodings of time (star glyph andclock glyph) were more effective when one had to find a particular temporal location. Participants’ subjective preferences support these findings and underline the fact that the clock metaphor helped in detecting specific temporal locations. Thus, our study shows that both accuracy and efficiency of tasks such as ours can be boosted when carefully choosing the most appropriate design.

3.4 Summary

In this chapter 3, the literature about different data glyph designs has been carefully reviewed for time-series data in various settings. Structuring the glyphs according to the basic visualization techniques they were combined with illustrated the great flexibility in positioning them on the screen.

Based on this related work I motivated the necessity for a metaphoric clock glyph design to convey temporal information in an easy to understand way. This design was implemented and applied to real world data in the network security domain. Three different prototypes were introduced (i. e., ClockView, ClockMap, andVistracer) usingclock glyphs to visualize complex temporal data structures. Each prototype was evaluated with a use case scenario to show the applicability of the design especially in combination with different visualization techniques.

To generalize the findings and get more concrete information about the performance of the clock glyph a controlled user study was conducted. Since there was a lack of quantitative ex-periments for data glyph designs using color saturation to encode data values this evaluation additionally closed some previously identified research gaps, which were revealed in chapter 2.

Based on the results of this experiment I further confirmed the usefulness of metaphoric designs for information visualization.

Chapter 4

Data Glyph Designs for Multi-Dimensional Data

Parts of this chapter appear in the following publications:

• Johannes Fuchs, Roman R¨adle, Dominik Sacha, Fabian Fischer, and Andreas Stoffel. Col-laborative Data Analysis with Smart Tangible Devices. InIS&T/SPIE Electronic Imaging, pages 90170C–90170C. International Society for Optics and Photonics, 2013¹

• Johannes Fuchs, Petra Isenberg, Anastasia Bezerianos, Fabian Fischer, and Enrico Bertini.

The Influence of Contour on Similarity Perception of Star Glyphs.IEEE TVCG, 20(12):2251–

2260, Dec 2014²

• Johannes Fuchs, Dominik J¨ackle, Niklas Weiler, and Tobias Schreck. Leaf Glyph - Visu-alizing Multi-dimensional Data with Environmental Cues. In Digital Library Scitepress, pages 195–206, March 2015³

• Johannes Fuchs, Dominik J¨ackle, Niklas Weiler, and Tobias Schreck. Leaf Glyphs: Story Telling and Data Analysis Using Environmental Data Glyph Metaphors, pages 123–143.

Springer International Publishing, Cham, 2016⁴

1The responsibilities for this joint publication were divided as follows: I spearheaded the writing of the paper, Roman R¨adle, Fabian Fischer and I did the programming and conducted the user study, Dominik Sacha and Andreas Stoffel formalized the problem and did the proofreading.

2The responsibilities for this joint publication were divided as follows: Petra Isenberg and I designed the user study. Fabian Fischer and I conducted the experiment. I was responsible for analyzing the results and writing the paper. Petra Isenberg, Anastasia Bezerianos and Enrico Bertini gave advice and did the proofreading.

3The responsibilities for this joint publication were divided as follows: I spearheaded the writing of the paper, Niklas Weiler did the programming, Dominik J¨ackle was involved in the writing, Tobias Schreck and I did the proofreading and gave advice.

4The responsibilities for this joint publication were divided as follows: I spearheaded the writing of the paper, Niklas Weiler and Dominik J¨ackle did the programming, Tobias Schreck and I did the proofreading and gave advice.

Multi-dimensional data can be considered as an n x m matrix with n being the different data points and m the corresponding attribute dimensions. In contrast to time-series data the relationship between the attributes may be dependent or independent from each other. This is reflected in the analysis tasks. Since attributes need not necessarily be related trend detection tasks across dimensions like in an intra-record comparison are not this likely. As can be seen in the systematic review of user studies in chapter 2 participants performing a trend detection task with multi-dimensional data compared single dimensions across different entities (inter-record comparison) and not within a single design (intra-record comparison) like with time-series data.

The design of the data glyph is, therefore, more flexible. Restrictions like the comparabil-ity of the different attribute dimensions like in time-series data are not necessarily given. The possibility for mapping data values to visual variables is, therefore, much more flexible. Based on Ward’s categorization [192] data glyph designs may comprise many-to-one, one-to-one, and one-to-many mappings, which results in a much bigger design space compared to temporal data.

Of course, the data glyph designs can also be combined with different visualizations (e.g., geo-graphic maps, node-link diagrams etc.), as well.

In the following section 4.1 I will review the literature about data glyphs for multi-dimensional data. Motivated from the related work section I will introduce a new data glyph design namely theleaf glyphmaking use of environmental cues to visualize multi-dimensional data. This design will be evaluated in a use case analyzing the forest fire data set form the UCI machine learning repository [44]. The dataset was carefully chosen to show the usefulness of the context related design and, therefore, the benefit of a metaphoric representation. After introducing this new data glyph I will focus on the well-known star glyph design with all its variations used in literature.

Since only little guidance exists, which star glyph variation works best for similarity search tasks this research gap will be closed by conducting a controlled user study and additionally trying to further improve on the design.

4.1 Related Work

Since data glyph designs for multi-dimensional data can be created in an entire flexible way, I will structure this section according to Ward’s classification of data glyphs [192]. In his cate-gorization he distinguishes between three different ways a data value can be mapped to a glyph representation.