Aqua Phoenix
     >>  Research >>  Actor Interaction Vis  


3. Visualization/Tool

The visualization tool contains two display modalities: 1. an interaction graph for the entire episode or film, and 2. interaction graphs for each scene-location-pair. A separate button panel contains tools to interact with the display, and a file chooser provides functions for loading and shuffling between different shot spreadsheets.

Figure 1.3: Actor Interaction Visualization Tool
Click image to enlarge, or click here to open

3.1 Actor Graph

The ActorGraph highlights interactions between actors over the entire duration of a sitcom episode (or film). Scenes and locations are not encoded in this representation. Actors are represented by stylized head icons (which should be replaced by actual headshots), and interactions between actors are represented by colored links with edge weights.

Actor interaction is encoded via 2-dimensional placement of actor nodes (represented by stylized face icons), and colored edges. The n-dimensional actor graph (n = number of actors) is projected into two-dimensional space via Multidimensional Scaling, a data reduction technique which encodes the most significant similarity (or, inversely, distance) in the first dimension, and then progressively less significant similarity in higher dimensions. Edge color and numerical edge labels signify the actual strength of interaction between actors. "Major" interaction is represented by the color in the edge. When "minor" interaction is available, it is represented by the thinner color band in the center of the edge. The color map is based on the jet set (dark blue - blue - cyan - green - yellow - red - dark red), with dark red representing the highest degree of interaction.

Figure 1.4: Actor Graph

3.2 Actor Scene Location Graph

The ActorSceneLocationGraph contains graphs ordered in a 2-dimensional table that separately encode the spatial dimension of location (rows) and the temporal dimension of scenes (columns). Each table cell highlights interactions between actors within a scene at a location (living room, coffee shop, etc.).

Row and column headers collapse the respective other dimension and depict interactions along a single dimension. Row headers summarize actor interactions within a location, whereas column headers summarize actor interactions within a scene.

The ActorSceneLocationGraph is sparsely populated, because not every location is used in every scene, and not every actor appears in every scene. Scene/location pairs without interaction are left as blank cells in the table. Scene/location pairs with interaction contain all actor nodes, but only relevant edges to the scene/location. The presence of all actor nodes establishes a spatial reference for comparison between graphs.

Figure 1.5: Actor Scene/Location Graph
Click image to enlarge, or click here to open

3.3 Functions

The "Palette" contains tools to manipulate and tune the graphs. With the "node" function, we can pick a node in the graph, move it, and change the node label. With the "edge" function, we can select an edge and modify its major and minor weights via a color slider.

The isolation tool enables custom selection of nodes, such that only selected actors and their relevant interactions are displayed in the scene/location graph. In "Isolation Mode", all actor nodes are by default hidden (drawn in half-transparency). Upon selecing (clicking on) individual nodes, they are once again included in the scene/location graphs in the ActorSceneLocationGraph. Only relevant edges between visible nodes are drawn, emphasizing specific interactions between actors across several dimensions.

Figure 1.6: Isolate selected actors
Click image to enlarge, or click here to open