This primitive supports both dense and sparse mode. In sparse mode, the input should be an undirected graph in compressed sparse row (CSR) format with its vertex and edge numbers. In dense mode, the input should be a set of vertex list with the known size. The result returns a vertex list with each vertex corresponding to a similarity value. The config contains several boolean values to control the graphType (0:Sparse Mode, 1:Dense Mode), similarityType (0:Jaccard Similarity, 1:Cosine Similarity), dataType (0:uint, 1:float).
