HGBER: Heterogeneous Graph Neural Network With Bidirectional Encoding Representation

Heterogeneous graphs with multiple types of nodes and link relationships are ubiquitous in many real-world applications. Heterogeneous graph neural networks (HGNNs) as an efficient technique have shown superior capacity of dealing with heterogeneous graphs. Existing HGNNs usually define multiple meta-paths in a heterogeneous graph to capture the composite relations and guide neighbor selection. However, these models only consider the simple relationships (i.e., concatenation or linear superposition) between different meta-paths, ignoring more general or complex relationships. In this article, we propose a novel unsupervised framework termed Heterogeneous Graph neural network with bidirectional encoding representation (HGBER) to learn comprehensive node representations. Specifically, the contrastive forward encoding is firstly performed to extract node representations on a set of meta-specific graphs corresponding to meta-paths. We then introduce the reversed encoding for the degradation process from the final node representations to each single meta-specific node representations. Moreover, to learn structure-preserving node representations, we further utilize a self-training module to discover the optimal node distribution through iterative optimization. Extensive experiments on five open public datasets show that the proposed HGBER model outperforms the state-of-the-art HGNNs baselines by 0.8%–8.4% in terms of accuracy on most datasets in various downstream tasks.

I. Introduction

Heterogeneous graphs, which are capable of modeling various types of nodes and diverse interactions between them, also known as heterogeneous information network, have become ubiquitous in real-world scenarios, ranging from bibliographic networks [1], social networks [2] to biological networks [3]. For example, as shown in Fig. 1, a bibliographic network (i.e., academic network) contains three types of nodes (author, paper, and venue) and two types of edges (author-write-paper and conference-publish-paper). Meanwhile, these basic relations can be further derived for more complex semantics over the heterogeneous graph (e.g., author-write-paper-conference-publish-paper). It has been well recognized that heterogeneous graphs are powerful models that are able to embrace rich semantics and structural information in real world data. Recently, heterogeneous graph neural networks (HGNNs) have received considerable research attention, because they are able to effectively combine the mechanism of message passing with complex heterogeneity, so that the complex structures and rich semantics can be well captured. So far, HGNNs have significantly promoted the development of heterogeneous network analysis toward real-world applications, e.g., recommend system [4], security system [5], and information retrieval [6]. Fig. 1.

Illustrative example of a heterogeneous citation network. (a) It consists of three types of nodes and two types of link relationships (heterogeneous citation network). (b) Author and its neighbors based on two meta-paths, Author-Paper-Author (APA) and Author-Paper-Venue-Paper-Author (APVPA) (meta-path-based node neighbors).