Abstract:Recently, the Network Representation Learning (NRL) techniques, which target at learning the low-dimension vector representation of graph structures, have attracted wide attention due to the effectiveness on various social-oriented application. Though large efforts have been made on the joint analysis combining node attributes with the network structure, they may usually fail to summarize the weighted correlations within nodes and attributes, especially when the nodes suffer extremely sparse attributes. To that end, in this paper, we propose a novel Sparse Attributed Network Embedding (SANE) framework to learn the network structure and sparse attribute information simultaneously in a united approach. Specifically, we first embed the nodes and attributes into a low-dimensional vector space. Then we introduce the pairwise method to capture the interaction between nodes and sparse attributes, and aggregate the attribute information of neighbors to alleviate sparsity for obtaining a better vector representation of node embeddings, which will be used in following network representation learning task. Along this line, we maintain the network structure by maximizing the probability of predicting the center node according to surrounding context nodes. Different from previous work, we introduce an attention mechanism to adaptively weigh the strength of interactions between each context node and the center node, according to the node attribute similarity. Furthermore, we combine the attention network with CBOW model to learn the similarity of the network structure and node attributes simultaneously. Extensive experiments on public datasets have validated the effectiveness of our SANE model with significant margin compared with the state-of-the-art baselines, which demonstrates the potential of adaptively attribute analysis in network embedding.

LDSNE: Learning Structural Network Embeddings by Encoding Local Distances

Learning Network Embedding with Community Structural Information

Enhancing the Network Embedding Quality with Structural Similarity

Local Neighborhood Embedding for Unsupervised Nonlinear Dimension Reduction.

Deep Recursive Network Embedding With Regular Equivalence

Evolutionary Network Embedding Preserving Both Local Proximity and Community Structure

Hierarchical Community Structure Preserving Network Embedding: A Subspace Approach

SPINE: Structural Identity Preserved Inductive Network Embedding.

Decentralized Framework for Embedding Large-Scale Networks

Local Embedding Learning Via Landmark-Based Dynamic Connections.

Local Graph Embeddings Based on Neighbors Degree Frequency of Nodes

Sim2vec: Node Similarity Preserving Network Embedding.

LEARN: Enhanced Network Embedding Via Similarity Learning.

Preserving Local and Global Information for Network Embedding

Local Structural Aware Heterogeneous Information Network Embedding Based on Relational Self-Attention Graph Neural Network

Lower Order Information Preserved Network Embedding Based on Non-Negative Matrix Decomposition

BDNE: A Method of Bi-Directional Distance Network Embedding

Locality Preserving Projection with Symmetric Graph Embedding for Unsupervised Dimensionality Reduction.

A Survey of Structural Representation Learning for Social Networks

Attributed Network Embedding for Learning in a Dynamic Environment

A United Approach to Learning Sparse Attributed Network Embedding