To Do

1. Write algorithms and implement temporal sets
   • Marriages are perfect examples, see May 9 meeting
2. Build table structure for extended marriages (tribes?) so we don’t have to build them on the fly from the DB every time.
3. Build the database and scripts to import Jill’s data
4. Write DHF abstract (see last week and week before notes)
5. Directed graph viz with a person, the marriages they are linked to (adoption, birth, marriage)
6. Add levels of separation to sankey (0 is just the marriage requested, 1 would be their parents’ marriages and children’s marriages, etc)

Discussion Points

1. Requirements for graduation?
2. Honeymoon time off, Nov 17-25
3. Moving offices (532: 4, 8 or 434: 3, 7)
4. Daniel sharing the Mormon viz on snac list
5. Digital Humanities Forum abstract
   • Will try to construct abstract draft today from Kathleen’s IATH application
6. Building the database
   • Will start constructing database and scripts today
7. Temporal Sets (Time-Dependent Sets)
   • New ideas for using timespans in the structure (insert for a given time span). Is the performance hit worth it?
   • See example below .
8. Measures on Temporal Networks: Time-dependent Data Structures (generalized to temporal pointers with temporal aspects on the nodes)
   • Using partial computation
     • Degree (in and out) can be easily done by storing the degree at each node. Increment/decrement when new time points come in (faster degree computation).
       • For directed graphs, this will be more important when wanting to calculate in-degree.
     • Connectivity - can we denote connected components or the size thereof?
     • Farness (1/closeness) - can we store the sum of distance to all other nodes? How often need to recompute?
9. To Note: When visualizing social networks, knowing the future structure is very good (see Visualization Methods for Longitudinal Social Networks and Stochastic Actor-Oriented Modeling, fig 2)
10. Some papers
    • See notes on website for 3 paper summaries

Discussion

Network Metrics

1. Degree (in- and out-)
2. Betweenness Centrality
3. Closeness (Farness) Centrality
4. Average Path Length
5. Average Cycle Length
6. Characteristic Path Length
7. Global Efficiency
8. Clustering Coefficient
9. Cliqueishness
10. Homogeneity
11. E-to-I Ratio
12. Density
13. Connectivity

Visualization

• There are some interesting visualization problems that still need attention
  • n-ary connections, n-ary relations. That is, how to show that more than 2 objects are connected
    • Ex: hypergraphs and hyperedges
  • 3-way connection (chord) in a chord diagram. Can we show a connection between 3 people or objects in a meaningful way?

Temporal Graphs

We really want to start looking at aggregate information. Not looking at snapshots of a temporal graph (or structure) at a certain timepoint, but over time, across time, etc.

• Measures
  • In- and out-degree
    • We know what this looks like for a snapshot. Or over over different snapshots.
    • What does it look like for a whole temporal network??
      • Is it the average over all time? Do we talk about standard deviation?
        • Importance of anode over all time: average degree of the node over all time? average connectivity of the node over all time?
        • Graph-wide: average degree of the network?
          • is this important? It’s the average (of all nodes) of averages (of all time for each node).
  • Connectivity over time
• Not just snapshots!
  • We need to push against the current ideas that a temporal network is just a series of snapshots that we can (or should) analyze individually.
  • You can’t always perform the normal metrics over a snapshot of the temporal graph
  • It could be that the network didn’t exist at the particular snapshot time in it’s true form, but was only in that form at that point as it transitioned from one point to another.
    • Ex: archive of the web. There’s not true entire web snapshot at a given time point, but it’s constructed using pieces from older/newer web crawls around that time.
• Derivates of the temporal graph
  • Maybe we consider changes as either clock ticks (metronome-like, such as minutes/days) or events (time only changes when something happens to cause the network to change).
  • Then we could look at a “snapshot” and +/- 4 changes (ticks/events)
    • What does the change in the network look like? (What does it’s change look like?)
    • Is it a time of rapid change?
    • What metrics should we perform?
    • Are groups formed quickly during this time?
    • Do groups change?
    • Does the connectivity of the network change and how?
  • We can think about this as the derivative of the graph. Given a timeframe, what’s the trend at this time? What’s the change of the graph look like?
  • Example: (twitter)
    • Justin Bieber has many followers. A system’s candidate hypothesized that followers connected to a well-connected hub are less likely to re-share any statuses from the hub, since all their friends are likely also connected and will get the original message. Therefore, as JB tweets, his followers are less likely to retweet, since they know their friends are also connected to JB and will also see the original tweet. This is the dampening effect, where as less-well-known tweeters may have their messages retweeted more heavily.
      • Can we see this by looking at measures over the changing temporal network (looking at the network change before, during, and after his tweet)?
      • What measures are important?
  • This needs to be about the entire network, and changes within the network (structure, etc)
    • There are graphs out there that show information over time, but not really measures of the network itself:
      • We’ve seen (insert link) 2-D graphs of tweets/sec of emergencies, disasters, or events, and how they have a general shape
    • Can we use those points (at the uptick, max, and bend in the downfall) of popularity change in those graphs/viz as a starting point to look at the actual network and how it changes over that time?
      • Use those times to pic up the state of the network, then look at network derivative through those times
    • We’d like to look at more comprehensive things of the network as they change through time (derivative across a point)
      • How do groups from in time?
      • How does the network change?
      • Is it a time of fast growth? Or is it stagnant?
      • What’s important that changes?
      • Do important nodes change (which nodes have highest centrality measures across this time, are they the same person)?