This paper reviews recent advancements in graph representation learning, focusing on methods for embedding nodes and subgraphs. It discusses various approaches, including matrix factorization, random-walk based algorithms, and graph neural networks. The paper highlights the importance of encoding graph structure into low-dimensional embeddings to enable effective machine learning. It reviews methods for embedding individual nodes and entire (sub)graphs, and develops a unified framework to describe these approaches. The paper also emphasizes important applications and directions for future work. It discusses the encoder-decoder framework, which is used to learn embeddings that capture structural information about the graph. The paper reviews various node embedding methods, including shallow embedding approaches such as DeepWalk and node2vec, and more recent approaches such as neighborhood autoencoder methods and graph convolutional networks. It also discusses the incorporation of task-specific supervision and extensions to multi-modal graphs. The paper concludes with a discussion of the importance of graph representation learning in various applications, including social networks, biological networks, and recommendation systems.This paper reviews recent advancements in graph representation learning, focusing on methods for embedding nodes and subgraphs. It discusses various approaches, including matrix factorization, random-walk based algorithms, and graph neural networks. The paper highlights the importance of encoding graph structure into low-dimensional embeddings to enable effective machine learning. It reviews methods for embedding individual nodes and entire (sub)graphs, and develops a unified framework to describe these approaches. The paper also emphasizes important applications and directions for future work. It discusses the encoder-decoder framework, which is used to learn embeddings that capture structural information about the graph. The paper reviews various node embedding methods, including shallow embedding approaches such as DeepWalk and node2vec, and more recent approaches such as neighborhood autoencoder methods and graph convolutional networks. It also discusses the incorporation of task-specific supervision and extensions to multi-modal graphs. The paper concludes with a discussion of the importance of graph representation learning in various applications, including social networks, biological networks, and recommendation systems.