March 2017

Meeting on 22.4.2017 at Uni LU with Mohammad Ghoniem, Fintan McGee, Mickael Steffas, Simone Zorzan, Antoine Laumond

Proposed dimensions for Hiveplots (as discussed :

  • Persons
  • Institutions
  • ePublications
  • Time
  • new: Topics

June 2017

  • Persons (Position: frequency with which they occur in the corpus to position them on the axis? Most prominent nodes outside to ease vis?)
  • Institutions (see Persons)
  • ePubs (Position: By number of associated resources? Color code individual ePubs)
  • Time (decades as axis attribute?)
  • Topics (see Persons)

Edge weight = frequency of co-occurrence between nodes on the axis

For the CVCE data, I expect that we will use hive plots not only to see patterns but just as much as gateways to the content.

  • Reading the graph: I can discover nodes which have certain properties but how can I get an overview of which nodes are actually displayed? Seeing the node symbol highlighted is not enough. The simplest solution to me is a complimentary list view. This can serve two purposes: 1) tell me which nodes are on an axis, 2) let me select nodes I care about and see where they are in the Hive Plot. Lists should be sortable by alphabet, degree?
  • How can I systematically investigate relationships? How can we pin them down temporarily? In Bostock’s demo (, any movement of the cursor makes a selection disappear
  • Display relevance of a node through node size - degree
  • How can we display relationships within an axis? Split it up and create a new one? Matrices? Is it feasible to have curved edges between nodes on an axis?
  • How can I see relations between two axes which are NOT neighboring axes? Could I move an axis next to another and thereby make them neighbors, i.e. by this movement update the data on display?

In my understanding, Hive Plots are not really made to display paths, clusters and components. Which is fine since with the histograph data we care more about individual links between entities rather than structural properties of the graphs.

June 26th 2017

Discussion between Marten During and Fintan McGee concerning the roadmap for DH visualization Interesting example app. Facet based visualization with a good timeline and timespan implementation. Very popular with DH researchers.

Interactive Nodelist

  • Highlight corresponding Node in graph (Mouse over / click)
  • On click maybe fade out nodes outside of neighbourhood? (try both, depending on context)
  • Filterable by text search
  • Orderable by attribute ( degree…. Co occurrence count)
  • Make it hideable ( low priority )
  • Include small data summary chart (histogram of years of mention, or attribute summaries, degree ( between modes) )
  • Reusable across views
  • Essentially a graph where only nodes are displayed
  • Similar to jigsaw query
  • Adapt this list so a set of them can be used to query the master graph

Time filtering

  • Remove nodes from master view based on time attribute
  • For DH use case the only time attribute is creation data of a resource
  • Resources will be connected to entities of interest…. most nodes will have a date ( 8000 resource node Don’t)
  • So dates will be collected by looking at resource node neighbours
  • Option include histograms with filters, including time nodes within graphs ( e.g. hive plot)
  • Need to clarify the role of time filtering in master view and layers ( is there a difference .e.g. delete node vs visually fade)

Person Search Co-occurrence View

  • Answer with whom does the person co-occur
  • There is an existing edge projection in the database (“ mentioned in document” and it may track the number of co-occurrences ( under “intersection”)
  • See Appendix A
  • Look into doing the same projection in the Blizaar for organizations etc
  • Use the hive plot to show each projection as a layer
  • Projections may have a large number of edges, so include an edge filter for a minimum projection edge strength, due to over drawing of edges
  • Need to decide between Visual edge removal vs removal in the underlying data.
  • Projections will only ever be on the resource nodes for the moment

Clustering is unlikely to be accepted by DH people.

  • Low priority for the moment

Perhaps a warning when a large amount of data is to be added to the graph

Topics to be visualized in the future

Results of topic mapping will be forthcoming soon

  • Search for a person P, get suggestions and select the right person
  • With whom does this person co-occur (list view on the side (names) / graph view (top connectors))
  • Which time periods are covered: With whom does P co-occur when?
  • Show me the documents between P and a selected person Y (links to new tabs)
  • How are P’s neighbours connected with each other? Are there clusters?
  • Repeat the above for institutions
  • View institutions and person related to node of interest in Hive Plot.