Cycle of Emotions : Composition
Interactive installation — projection, pedestal interface, webcam (FER)
Ars Electronica Festival 2026, Campus Exhibition · POSTCITY, Linz · September 2026
About the work
A continuation of Cycle of Emotions, G.MAP Gwangju, January–April 2026
This work continues the Cycle of Emotions project, shown at the Gwangju Media Art Platform as part of New Wave. It looks at how emotion is turned into data by algorithms, and how that data can be returned to art.
Its starting point is Wassily Kandinsky’s chain reaction: the artist’s emotion takes visual form as a work, which the viewer then senses and answers with an emotion of their own. Here the chain begins somewhere else. The viewer’s emotion comes first, and the data it leaves behind accumulates as the work itself — a circulation model in which the viewer composes rather than receives.
The installation has two parts: a projection screen, and a pedestal holding an interface and a webcam. At the pedestal you choose a color for the emotion you want to give, and it appears on screen straight away. At the same time the webcam reads your face and pulls out the emotion that dominates, and the screen answers — texture and sound shifting as that emotion changes.
The three layers
Three different times and places, stacked on one screen
The Korean Dataset
Emotion data from around 1,400 visitors to the Gwangju exhibition, held as representative colors. This is the ground everything else sits on — and the layer described in detail below.
The Cumulative Dataset
Colors left by visitors here in Linz, gathering over the run of the festival. It thickens as more people pass through.
The Individual Layer
The color you have just chosen, as a geometric form on top of the other two. It is yours only for a moment, then it joins the middle layer.
The Korean Dataset
What colors did Korean visitors give to each emotion?
Between January and April 2026, around 1,400 visitors to Cycle of Emotions at G.MAP in Gwangju assigned a color to each of seven emotions. They left 1,862 responses; after filtering out noise, 746 were used for this analysis. Nothing personal was recorded — each entry holds only a timestamp, an emotion, and a color.
The only emotion that never settles into one region. Choices spread across five color families at 11–22% each — happiness seems to mean different things to different people.
Blues dominate: sky blue (22%), cyan (14%) and blue (13%) together account for 49% of all choices. Darker in tone than most other emotions.
The sharpest single association in the dataset — red (25%) and pink (25%) make up half of all choices. Anger is also the most saturated emotion here (S=182). The cool tones sitting alongside the reds are worth a second look.
Hue spreads widely, but brightness does not — fear gathers at the dark end whatever the color. Here how dark seems to matter more than which color.
The darkest emotion in the dataset (V=137). No single hue takes over, but the tonal range stays consistently low.
Teal leads on its own (19%). Both brightness and saturation run high, yet red never surfaces — which is what separates surprise from anger in this data.
The least saturated of the seven (S=143). Teal comes first at 16%, but nothing dominates — no strong color seems to come to mind for neutral.
Where the choices fall
Every response sorted into one of twelve 30° hue bands (%)
Three things stand out. Anger concentrates in red far more sharply than any other emotion does anywhere on the wheel. Sadness sits almost entirely in the blue family. And surprise and neutral — two emotions with little else in common — both peak at teal.
The choices are not random
Chi-square test against an even spread across the twelve bands
If visitors had picked colors without regard to emotion, responses would fall evenly across the twelve hue bands. They do not.
Four emotions fall below p<0.001. Fear and disgust show weaker figures largely because their samples are smaller.
Still gathering
The colors above are one audience, in one place, over four months. A second set is planned from visitors in Linz, to sit beside the Korean dataset rather than replace it. What the two have in common, and where they part, is the part that has not been written yet.
JOANN LEE 