future work

Graph data Quality
Unknown Cluster Number
Stability
Scalability
graph partition like random work
questions
classical methods like kmeans can't get good preformance
many methods may not fit batch training techniques
包括DEC ，node2vec ，DGI ，MVGRL ，BGRL ，S3GC 和Dink—Net 。
For the first part, i.e., encoder training on the large-scale graph, the possible techniques include the better sub-graph extraction method, more efficient batch training strategy, etc. For the second part, i.e., node clustering on the large-scale graph, the potential solutions contain the mini-batch k-Means , finding cluster centers via the neural network , calculating the clustering assignments in a minibatch manner , and bi-part graph clustering , etc.



Discriminative Capability
Unknown Cluster Number