To do
- Continue working on getting data in
- Outline for the presentation/paper
- Get something to Worthy next week, even partial deliverables are okay
- Fix sliders
- Start thinking about dissertation proposal
- READ MORE PAPERS
Digital Humanities Forum
- Good place for feedback on some of these ideas
- Transaction time vs calendar time
- Are there good distinctions?
- What if the transactions did not happen in calendar time order?
- Example: narrative story, telling about past events
- Reading a book, there is page order, where things happen in order of the book, from page 1 to page $n$
- But, there is also chronological order, the time in which events occured. A later page might have a flashback to an earier time point
- What about a node being a conversation. It would grow as people are included in the conversation (cc-ed, mentioned in a tweet, retweeted, etc)
- The documents could be analyzed later, and the analysis could add other people into that conversation (perhaps people metioned in emails or tweets, but not originally linked into the conversation.
- The later analysis might add people into the conversation at an earlier time, so “analysis time” would be different than real time
- Analyizing documents in an archival box
- Analysis order will likely not be in chronological order of the documents. So they will go in at different times.
- There is an analogue here to storing a sparce matrix
- Event driven storage is like a sparce matrix, you only need to store points or the structure when an event occurs. So, the calendar time in between is empty.
- Wall-clock, calendar, time would be the full matrix
- How much do we store about each event? Do we store the entire state of the graph? Or use something like what the temporal graph structures use and store the updated portion?
- Idea: 2-D sparce matrix, x-axis = time ordered by telling, y-axis = time ordered by calendar, to store something where we have 2 different orderings of events.
- Outline for the talk
- Evolving graphs
- Evolving over time
- We’ve seen temporal graphs in literature. Go over what temporal graphs are.
- We’re doing something extra, different from these (looking at the evolution over time)
- variation
- distinction
- these are crucial for the dissertation
- Examples to show. Use this distinction between evolving graphs and temporal graphs to pick some good examples to show. What should we show them?
- What we’re doing for our example, marriages
- Complexity of the marriages, richness of the data
- …
Other Discussion
- We have these two different levels/evolving structures and we want to see how the interact
-
- Groups of people as nodes
- People can be in multiple groups, so we have multi-arcs or hyperedges for people connecting groups
-
- Marriages as nodes
- There’s a geneological connection here of people between families
- One arc per person, less hyperedges
- We’ve been hand wavy about knowing how one affects the other
- We want temporal/evolving measures that we can draw a correlation as to which one affects the other (and why?).
- One idea: pull measures out of each individual graph (say connectedness, degree, etc), which are time-dependent, then use those measures as dependent variables/values over time and compare the metrics to find some correlation.
- Desirable for dissertation
- Aggregate measures over one of these things
- the notion of :what is an evolving network? (structure, definition, data structure, theoretical notion, …)
- Necessary (sufficient?) conditions fo the graph that make the evolving measures make sense / possible
- Now, with two of them
- are there necessary conditions (or sufficient conditions), ie max degree, avg degree, etc, that make evolving comparison between the two evolving structure possible or make sense (in this context)?
- Conditions on each (independently) that make comparison possible. Necessary or sufficient.
- One has this property, the other has this other property, so therefore they won’t be comparable with these measures
- “it is this, but not that by definition”
- “when is the measure/structure applicable?”
- If timespan organization is not consistent (event-driven vs calendar-driven, or in different orders), then some measures might not make sense
- Event-time: we might need to have specific rates (when considered on calendar-time) for certain measures to make sense
- one or the other can not be bursting while the other is not, or vice versa. The rates might need to match for some measures to make sense.
- Queuing theory: queue lengths, consumer vs producer rates
- might have some relevance here, or find some good results in queuing theory
- Comparing two different queuing networks
- Worthy hasn’t seen anything comparing these different networks, which might have different conditions, makeups, parameters
- Could do something like get the throughput of each queuing network and compare their throughputs
- This might work, however throughput might not be a fair comparison of the two based on structure, type of the network, rate of events, network conditions, etc.