How to read these slides?

Navigate via arrow keys between slides Overview

Click on the menu bar items to navigate to chapters

Click here for PDF version
SCIENCE PASSION TECHNOLOGY

Journal Club: Time Curves

Julian Rakuschek

23.01.2025

Motivation

Temporal Patterns

Article Revision

Brain Activity

Climate

Domain specific visualizations exist

Revisions of the "Chocolate" Wikipedia article

The Problem: They are not easy to adapt for other domains!

The goal: Development of dataset agnostic visualization method.

The Paper

2016 TVCG Paper with 247 citations (Google Scholar)

Time Curves

What do we need?

First: Snapshots of your dataset at different time points

Second: Similarity Measure

Example

Snapshots of Wikipedia articles

Similarity measure: edit distance

Arranging Points by Similarity

Multidimensional Scaling

\[ \text{Stress}_D (x_1, x_2, \ldots, x_N) = \sqrt{\sum_{i \neq j = 1, \ldots, N} (d_{ij} - \|x_i - x_j\|)^2} \]

$x_i \ldots$ data points in projected space

$d_{ij} \ldots$ similarity between data points in high dimensional space

$\|x_i - x_j\| \ldots$ distance between projected points

Goal: Minimise Stress

First Step

Apply MDS to the dataset

Second Step

Connect the points in their temporal ordering

Third Step

Remove overlaps and color the points

Like Folding a Timeline

Important choices

  1. Similarity metric
  2. Dimensionality reduction method, e.g. MDS
  3. Curve drawing algorithm, e.g. Catmull-Rom
  4. Extras:
    • Remove overlap
    • Coloring
    • Node size

Applications

Document Histories

The Wikipedia chocolate page edit war

Document Histories

Alternative coloring highlights user groups

Document Histories

Palestine Wikipedia Page (Caution: 2016)

Document Histories

Vandalism

Document Histories

Time Curves as fingerprints

Video Time Curves

Surveillance

Video Time Curves

Precipitation

Taxonomy of patterns and characteristics

Limitations

Limitations

  • Quantitative aspect of time is lost.
  • High dependence on distance metric.
  • MDS expensive to compute.
  • Curve might not be legible due to high complexity.
  • Not resistant to noise.

Why did I choose this paper?

Because it might be relevant for your projects!

  • PRESENT
  • Hereditary
  • A+CHIS
  • OpenReassembly: Possible Puzzle Wars?

Can you think of possible use cases?

Could you show an "edit war" in some datasets provided in your projects?