Graph visualization¶

Graphiy is a graph management tool. Naturally its navigation view is a graph visualization.

Criteria¶

There are some important problems in well-known interface we all are used to.

important files appear to be buried deep in the structure
cannot put a file in multiple folders
cannot name a file with multiple names (tags)
erroneous structure emerge when file is in related folder, but doesn’t relate to its nesting hierarchy

Graph layout resolves all above. But it is harder to visualize. Existing graph visualizations share some common problems:

Too many nodes. This usually makes sense only to show a big picture, like on mindmap. But labels quickly become unreadable with scale.
Too many neighbors. There are some nodes highly interconnected with others, like group definition. Showing all its linked nodes always makes those crowded inflorescences.
Too many links. Unlike hierarchies, graph nodes are free to connect to any other nodes. Visually this always becomes a hairball, in which crosslinks goes as strikes across other nodes.
Scattered related nodes. Good layouts may resolve some issues from above revealing clusters of information. But it’s not rare when node is related to multiple clusters.
Spatial stability in dynamics. Once graph is drawn and navigation starts, user memorizes the spatial mental map. That’s how our brain works. Revealing new nodes, collapse / expand, hide actions change the graph substantially. And if layout of nodes which stay changes too much graph would seem to be a new one with complete loss of context and relatedness with previous state.

We addressed these issues with the corresponding set of techniques:

define context: any item can be found with no more than 5 links to others and usually 2 or 3. So working in certain context wouldn’t need many nodes to show.
filtering by links to group nodes

show only first N neighbors (connected nodes) ordered by link weight (nodes relatedness). It may be any other user defined criteria, like popularity or rank.
show neighbors in separate scrollable list view

highlight selection links (partial solution)
show graph as a tree starting at current context (selected node / nodes)
place related nodes nearby and duplicate those which need to be placed in multiple clustered areas

persist nodes positions
let the user manually change position and fix it
layout nodes depending on previous position, preserving overal shapes of clusters and minimizing individual nodes shift.