Between 2 nodes, 2 relationships are created in both directions instead of 1 in one direction
> CALL gds.knn.write('myGraph', {
> writeRelationshipType: 'SIMILAR',
> writeProperty: 'score',
> topK: 1,similarityCutoff: 0.9,
> randomSeed: 42,
> concurrency: 1,
> nodeProperties: ['htc']
> })
> YIELD nodesCompared, relationshipsWritten
Result:
I think the results make sense. The algorithm calculates a distance metric between each pair of nodes and then finds each nodes k nearest neighbors, creating a relationship between the node and its k nearest neighbors. In your case, you have specified k as 1, through the topK parameter. As such each node would have a relationship to its closest neighbor. The pairs with two relationships indicates that the two nodes are closest to each other.
In contrast HTC 2013 is closest to HTC 2014, but HTC 2014 is closest to HTC 2008 and not HTC 2013.
You should be able to verify this by looking at the relationship property ‘score’ written.
As a sanity check, for each node you should get a relationship to every other node if topK >=5
