Graph embedding, aiming to learn low-dimensional representations for nodes in graphs, has attracted increasing attention due to its critical application including node classification, link prediction and clustering in social network analysis. Most existing algorithms for graph embedding only rely on the topology information and fail to use the copious information in nodes as well as edges. As a result, their performance for many tasks may not be satisfactory. In this thesis, we proposed a novel and general framework for graph embedding with rich text information (GERI) through constructing a heterogeneous network, in which we integrate node and edge content information with graph topology. Specially, we designed a novel biased random walk to explore the constructed heterogeneous network with the notion of flexible neighborhood. Our sampling strategy can compromise between BFS and DFS local search on heterogeneous graph. To further improve our algorithm, we proposed semi-supervised GERI (SGERI), which learns graph embedding in an discriminative manner through heterogeneous network with label information. The efficacy of our method is demonstrated by extensive comparison experiments with 9 baselines over multi-label and multi-class classification on various datasets including Citeseer, Cora, DBLP and Wiki. It shows that GERI improves the Micro-F1 and Macro-F1 of node classification up to 10%, and SGERI improves GERI by 5% in Wiki.

Graphs are commonly employed to model complex data and discover latent patterns and relationships between entities in the real world. Canonical graph learning models have achieved remarkable progress in modeling and inference on graph-structured data that consists of nodes connected by edges. Generally, they leverage abundant labeled data for model training and thus inevitably suffer from the label scarcity issue due to the expense and hardship of data annotation in practice. Data-efficient graph learning attempts to address the prevailing data scarcity issue in graph mining problems, of which the key idea is to transfer knowledge from the related resources to obtain the models with good generalizability to the target graph-related tasks with mere annotations. However, the generalization of the models to data-scarce scenarios is faced with challenges including 1) dealing with graph structure and structural heterogeneity to extract transferable knowledge; 2) selecting beneficial and fine-grained knowledge for effective transfer; 3) addressing the divergence across different resources to promote knowledge transfer. Motivated by the aforementioned challenges, the dissertation mainly focuses on three perspectives, i.e., knowledge extraction with graph heterogeneity, knowledge selection, and knowledge transfer. The purposed models are applied to various node classification and graph classification tasks in the low-data regimes, evaluated on a variety of datasets, and have shown their effectiveness compared with the state-of-the-art baselines.