I applied FastRP to create node embeddings for a graph consisting of 6,301 nodes. These embeddings are stored as node properties to be used in clustering using the K-means algorithm. I noticed that although the nodes are nearby to each other, they are assigned to different clusters.
Notes:
-
For FastRP, I set the embedding dimensions to 256.
-
For K-means, I set K to 2.
-
I tried a smaller embedding equals 2, 4, etc, The same results occurred.
-
In addition, I tried another graph size with 8,846 nodes. Similar incomprehensible results occurred.
-
I didn't specify a random seed for FastRP. I didn't know how to set a preferred value for this parameter. I it related to the graph size like the node embedding?