In this repository, we present a slight modification to the knowledge graph (KG) with the domain model used as a basis in 10.1109/JBHI.2024.3417224. This new KG has been used in doi: TODO.
The changes over the KG relate to the spatial dimension used to describe a hospital layout. Specifically, we have changed the edges that link the Area nodes with their coordinates (Unit [row] and Block [column] nodes) over its Floor.
Below, we present some other related repositories that may be of interest to you:
- STeMECH: Code for
doi: TODO. - HospitalEdgeWeights: It is also linked to
doi: TODO. - HospitalGeneratorRDF_V2: It is also linked to
doi: TODO.
The following edges:
- placedInUnit between an Area and an Unit
- placedInBlock between an Area and a Block
- placedInFloor between a Block or Unit and a Floor
- nextTo between two Units or two Blocks
should not have a weight. Therefore, they couldn't be used to calculate the semantic distance between two Location nodes.
In the following image, we present a floor layout with 6 Areas organized in 2 rows (Units) and 3 columns (Blocks). Units are named with letters and Blocks with numbers. Each Area is named with the union of its Block and Unit.
In this floor layout, the Area 0A should only have two neighbouring Areas:
- 1A: Same Unit (A), consecutive Block
- 0B: Consecutive Unit, same Block (0)
However, if we represent the floor layout as a graph, we can trace several paths that drive to "fake neighbours":
- 0A -placedInUnit → A -nextTo → B ← placedIn- 2B
- 0A -placedInBlock → 0 -nextTo → 1 ← placedIn- 1B
- 0A -placedInUnit → A ← placedIn- 2A
(We can access to any _Area_ that is linked to the same _Block_/_Unit_ of another one)
As we can see in the previous section, when a shortest path algorithm is applied to search the path with the lowest weight between two Location nodes from the same Area, we might get a path that relates both nodes as "fake neighbours" and, therefore, the result distance between them would be lower than it should be.
Hence, we will add a new relation nextTo between Area nodes to define when two Area nodes are neighbours explicitly. For the shortest path algorithm between Location nodes, we will only use those edges with a weight property, so the edges between Area/Floor and Unit/Block won't be traversed.
We don't think that Unit and Block nodes have to be turned into attributes of Area nodes. As nodes, it is easier to retrieve all the Areas from the same Unit/Block, or get the number of columns or rows in which a Floor is organised.
In the following figure, we represent how these changes are implemented in the kG:
In the following figure, we represent the new KG that represents the domain model of a hospital (its layout and activity):
