Week 5-6

New Idea

A heatmap for node, link level distribution over a certain local measure.

• x-axis: time (foldable according to aggregation level, suitable for O(N) measures)

• y-axis: bins for the measure

• hue: number of links / nodes

drawback: ineffective to discover the changes of active nodes.

use a pillar to encode distribution?

convert the box plot to a streamgraph? make the timeline brushable

implementation on degree distribution over time

A radar chart for group level (global) measures (updated at each slide)

drawback: less effective to show values -> table / parallel coordinate

Data Loading Strategy

• Pre-computed

  • High complexity static local measures of every timestamp

    • when brushed, encode the derived value ~ max, min, ave, std

    • hypothesis: single measure on aggregated graph is of no use; we brush for a closer look at changes during the period

• Compute on demand

  • Dynamic local measures at brushed period

  • Global measures at each brushed period

    • selected groups

    • whole network

Issue

is link-specific measures wanted? will there be difference between nodes and links?

do we need a measure table/list?

is it worthwhile to aggregate nodes into a super-node and re-analyze on measures?

To Do

1. Group (subgraph) selection (lasso), global statics radar), subgraph node-link

2. heatmap view

3. re-arrange the web page

4. think about a clearer transition from global to subgraph then to local