利用网络结构对节点进行向量化表示
矩阵分解-传统方法
Laplacian Eigenmaps(LE)
Graph Factorization)(GF)
High-Order Proximity preserved Embedding(HOPE)
Random Walk based(随机游走)
DeepWalk:
https://blog.csdn.net/weixin_39915444/article/details/80666037
https://blog.csdn.net/qq_26919935/article/details/76574845
node2vec:
https://blog.csdn.net/sinat_26917383/article/details/54406344
LINE(Large-scale Information Network Embeddings)
参考:https://blog.csdn.net/u013527419/article/details/76017528
Deep Learning based
SDNE(Structural Deep Network Embeddings)-通过auto-encoder降维
DNGR(Deep Neural Graph Representations)
Neighborhood Aggregation Algorithms-将节点本身及其邻居节点的属性(比如文本信息)
或特征(比如统计信息)编码进向量中,引入了更多特征信息,并且在邻居节点间共享了一些特征或参数。
GCN(Graph Convolutional Networks)
GraphSAGE
相似性指标:Kate Index, Root Page Rank, Common Neighbors, Adamic-Adar Score等
数据集 dataset
Synthetic:SYN-SBM
Social Network:Karate Youtube BlogCatalog
Biloogy Network: PPI
Collaboration Network: HEP-TH ASRTO-PH