Virtual nodes and relationships are used for quite a number of things.
You can create visual representation of data that are not in the graph, e.g. from other databases, that's also what the apoc.bolt.load procedure supports.
Also the virtual nodes and relationships are used by db.schema and apoc.meta.graph
You can use it to visually project data, e.g. aggregate relationships into one, or collapse intermediate nodes into virtual relationships etc. For instance, you can project a citation graph into a virtual author-author or paper-paper graph with aggregated relationships between them.
Or turn twitter data into an user-user mention graph.
This is already automated in apoc.nodes.group which automatically groups nodes and relationships by grouping properties, read more about that here.
You can combine real and virtual entities, e.g. connecting two real nodes with a virtual relationship or connecting a virtual node via a virtual relationship to a real node.
Apoc has already some means to also create "virtual graphs" which can also be used for export.
Some more uses of virtual entities:
- return only a few properties of nodes/rels to the visualization, e.g. of you have huge text properties
- visualize clusters found by graph algorithms
- aggregate information to a higher level of abstraction
- skip intermediate nodes in a longer path
- hide away properties or intermediate nodes/relationships for security reasons
- graph grouping
- visualization of data from other sources (computation, RDBMS, document-dbs, CSV, XML, JSON) as graph without even storing it
- projecting partial data
You can also create them yourself e.g. for projections.
One thing to keep in mind: as you cannot look up already created virtual nodes from the graph you have to keep them in your own lookup structure. Something that works well for it is apoc.map.groupBy(Multi) which creates a map from a list of entities, keyed by the string value of a given property.
Virtual entities so far work across all surfaces, Neo4j-Browser, Bloom, neovis, and all the drivers, which is really cool, even if it was not intended.
They are mainly used for visualization as Cypher itself can't access them. That's why I added a bunch of functions to access their properties, labels, and rel-types.
In some future, they might be subsumed by graph views, the ability to return graphs and composable Cypher queries in Cypher 10.
We also allow graph projections to be used as inputs for graph algorithms, so you don't actually have to change your data to run an algorithm on a different shape but you can just specify node- and relationship-lists with two Cypher statements.
This turned out much more than I originally intended, almost a blog post :)