Visualisation for Digital Humanities

Thomas Wielfaert

KU Leuven - Quantitative Lexicology and Variational Linguistics

Digital Humanities Spring Event

29 April 2015

Why?

Increasingly bigger data sets: 3 V's

source: gtcorp.com

The fourth research paradigm (Jim Gray)

Rather than finding data to test a hypothesis, find a hypothesis that can be tested on the existing data.

1. Empirical: describing natural phenomena
2. Theoretical: laws i.e. Newton, Kepler, ...
3. Computational: modelling, simulation
4. Data exploration: data processing a.k.a. Big Data

Black box algorithms

Visual Analytics can reveal properties of algorithms that were not detected before.

source: visual-analytics.eu

What is a good visualisation?

What is not?

Caveats of visual perception

Hermann grid

Checker shadow illusion

Properties of good visualisations

• Visual perception theory: research field proper
  • Even 'simple' standard visualisations such as boxplots and histograms can easily mislead. (see Lem et al. 2013)

Semiology of graphics (Bertin)

Reminder: types of variables

1. Numerical (quantitative): 23.4; 0.2; 5e-1
2. Categorical (nominal): man/woman/X
3. Ordinal (ordered): low/medium/high

Ranking of perceptual tasks

Mackinlay (1986)

colorbrewer2.org

Never ever pick colors yourself!

Different kinds of data

• Geo(spatial) data: maps
• Multivariate/high-dimensional data: line, point, region-based
• Graphs/networks

source: D3.js gallery

Get inspired

Don't try to reinvent the wheel...

Good starting point: D3.js Gallery

https://github.com/mbostock/d3/wiki/Gallery

So what is a good visualisation?

Interplay between several factors:

• Type and quality of the data
• Human cognition (visual perception theory)
• Usability (interactions fast enough?)

Evaluation:

• Interaction between programmer/designer and user/domain expert
• Discussions during visualisation work shops

Is this for me?

Empirical DH fields

• Computational linguistics
  • Corpus linguistics
  • Also literature
• Social sciences
• Philosophy
• Musicology
• ....

How?

Use commercial software

i.e. Tableau (also free version available)

Use commercial software

Pro:

• You can get started immediately; no need to program yourself.

Cons:

• Step into a closed ecosystem
• For technical support you will have to pay
• Depend on a company rather than a community for new features

​Program it yourself

Programming languages designed for the job:

1. Processing: based on Java, runs as a stand-alone program on your computer.
2. D3.js (Data Driven Documents): Javascript, works in any modern browser; easy to publish online; the future.

Bottom line: extremely flexible and versatile, but comes with a steep learning curve.

Middle ground

R data frame

D3.js

Google Charts

...

R libraries:

• rCharts
• networkD3
• googleVis
• ...

Some DH visualisations

DoubleTreeJS (Chris Culy)

Slash/A (Todorova and Chinkina)

Very basic introduction to D3.js

Today 14:45-15:45, MSI 02.18

Hands on: bring your own laptop or befriend someone with a laptop with Chrome or Firefox installed (pretty please).

Topics:

• The very basics of D3
• Import a dataset (!)
• Make a basic visualisation of the data set

Comments? Questions?

thomas.wielfaert@kuleuven.be

References

Bertin, J., (1967). Sémiologie graphique, Mouton/Gauthier-Villars, Paris.

Collins, J.P., (2009). The Fourth Paradigm: Data-Intensive Scientific Discovery. Microsoft Research, Redmond, Washington.

Lem, S., Onghena, P., Verschaffel, L., Van Dooren, W. (2013). On the misinterpretation of histograms and box plots. Educational Psychology: An International Journal of Experimental Educational Psychology, 33 (2), 155-174.

Mackinlay, J.D., (1986). Automating the Design of Graphical Presentations of Relational Information, ACM Transactions on Graphics, 5, 110-141.