Explore visualization/representation methods for site connections

[cory-skeers]

Issue

We have a wide web ecosystem, and currently our main form of analysis is siloed at the host domain level. How these domains connect, or how this looks once you get into a site, is less clear.

Look into how to "map" user journeys, and explore other visualization and site connection methods. Likely some form of node graphs, but exploration of other possibilities.

Proposed solution

• Parse directories/pages to create similarity scores for pages within an individual site, to create forked/branching paths
• Create bigrams of page referrers to measure site "distance" and connection points
  • Node graph with stretchy connections weighted by a sort of TF-IDF measure

[cory-skeers]

Started looking into grabbing referrer -> page connections, and connecting an undirected graph based on where connections are or not. Next steps are looking at weighting these.

Here you can see the upper arc of non-uiowa hosts, and a lower cluster of uiowa hosts (sample of 1000 connections, labels removed).

Want to look more into weighting, as well as a pre-structured approach. This approach goes at end-user first, where looking at page/menu structures would be a content-editor first approach. The differences between the two may be interesting.