, , , ,

Uncertainty is a state that cannot be entirely described because of limited knowledge. It can arise in the data when they are acquired, processed or visualized (which is actually called uncertainty of visualization).

This short post will be focusing on the visualization of uncertainty, which is defined by the use of special variables (e.g. colors, size or texture…) in order to include uncertainty in a diagram. This is very important because people tend to treat data differently when they can visualize them instead of simply reading them. Data are not questioned much when visualised.[2] How uncertainty will be included in a diagram depends mostly on its properties. For some kind of diagram, there are quite a few simple variables that can be used to include uncertainty whereas for others, it will be a lot more challenging. In term of modeling, the inclusion of uncertainty in a visualization results in most cases to the addition of a dimension.[4] In this short post, 3 examples are given to show how uncertainty can be visualized.

In “Digging into Human Rights Violations: phrase mining and trigram visualization”, the data that is studied comes from multiple testimonies. In this case, temporal, locative and entity uncertainties are introduced at the start, during the acquisition. The authors chose to use a trigram-based visualization called an event trigraph. Shortly, an event trigraph is a 2-D planar diagram consisting of events as its buildings blocks. The weights on the line represent the confidence values and they have a peak at 1. This representation makes it possible to visually associate events that are in different documents at the same time.[1]


Figure 1: Event trigraph with uncertainty values and voids

Instead of using a special kind of graph to show uncertainty visually, the authors of [2] propose an approach that only modifies a little bit the original representation. One of the most natural ways to visualize uncertain data is probably to use blurriness. The analogy is simple but strong and just like location, size, texture, color, orientation, shape, color saturation or transparency (…), blurriness is a good variable because those features correspond to the basic feature channels in our primary visual cortex, being thus perceptually completely distinct features (Ware 2012) [1].

In the paper “Visualizing Uncertainty: How to Use the Fuzzy Data of 550 Medieval Texts?”, the approach is very different: a web-based tool permits a geospatial versus temporal visualization that is robust and includes uncertainty. It is possible to change dynamically the thresholds for the different type of uncertainties and more traditionally select the genres and time range to choose the amount of data that is displayed. The uncertainty in this study seems to be higher than for the first two examples.

Uncertainty visualization is challenging because most visualisations are used with the assumption that the data are accurate. The problem is that data exists in different forms and can have different components. It is not always possible to simply draw an error bar or to add a spatial dimension. With the examples given in this post, it should be a bit easier to see how to get around those difficulties or at least realise that they exist. Also, after reading through this post, one should see that including uncertainty into a dataset does not only allow the use of all the data but can also lead to very interesting or unsuspected results.


[1] Digging into Human Rights Violations: phrase mining and trigram visualization. Miller, Ben ; Li, Fuxin ; Shrestha, Ayush ; Umapathy, Karthikeyan. http://dh2013.unl.edu/abstracts/ab-368.html

[2] Bindings of Uncertainty. Visualizing Uncertain and Imprecise Data in Automatically Generated Bookbinding Structure Diagrams. Campagnolo Alberto ; Velios Athanasio. http://dh2013.unl.edu/abstracts/ab-187.html

[3] Visualizing Uncertainty: How to Use the Fuzzy Data of 550 Medieval Texts? Jänicke, Stefan ; Wrisley, David Joseph. http://dh2013.unl.edu/abstracts/ab-158.html


[4] A Review of Uncertainty in Data Visualization. Ken Brodlie, Rodolfo Allendes Osorio and Adriano Lopes. http://www.comp.leeds.ac.uk/kwb/publication_repository/2012/uncert.pdf