Graph Deep Clustering using Cluster Graph Conventional

Graph deep clustering aims to have an effective representation of the graph data structure into multiple groups. Most of the studies in graph clustering move to apply deep learning because of the effectiveness of the combination for feature extraction, dimensionality reduction and clustering. However, the graph area is very wide and always has a lot of varieties and possibilities for new approaches and models. The different types of graphs and networks in addition to the type of the data and graph structure cause that variety and the increase in the graph deep clustering models. Thus, it is important to gather all the new and latest approaches and compare the performance according to the most popular datasets in this field. This study focuses on the latest approaches in the area of graph deep learning clustering. Moreover, the best performance result approaches from 2016 until 2021 are compared according to the highest accuracy result. The comparison is done between Cora, Pubmed and Citeseer datasets for each approach. Since the data in the real world is complex, the study also focuses on the proposed approaches for an important type of graph which is a multiplex graph. The study investigates that most of the new approaches are using the node and graph structures in the learning method. Most of the approaches are evaluated on a single layer graphs, wheres few proposed using multiplex graph. Another main finding in this paper is the use of the features of the LSTM as an autoencoder techniques in the graph clustering. However, there are few studies in this field which can be an area for researchers as new studies. Finally, a cluster graph conventional model is proposed. Two datasets are used which are Cora and Pubmed. The best accuracy results in our experiment are 75.25% and it is shown when we use the Pubmed dataset. This result is the highest accuracy compared with other graph clustering models using Pubmed dataset.