FHGE: A Fast Heterogeneous Graph Embedding with Ad-hoc Meta-paths

TL;DR

This paper tackles the challenge of producing fast, meta-path-guided embeddings for heterogeneous graphs under ad-hoc queries, where traditional HGNNs incur prohibitive retraining costs. It introduces FHGE, a two-module framework that segments graph information using Meta-Path Units (MPUs) to capture local/global structure, and reconstructs embeddings through type-specific transformations and MPU-based semantics with a dual attention mechanism. FHGE demonstrates strong effectiveness and substantially higher efficiency compared to state-of-the-art baselines on meta-path-guided tasks, while maintaining competitive performance on downstream link prediction and node classification. The approach enables real-time, ad-hoc semantic querying over HetGs and shows potential for real-world deployment in dynamic graph analytics.

Abstract

Graph neural networks (GNNs) have emerged as the state of the art for a variety of graph-related tasks and have been widely used in Heterogeneous Graphs (HetGs), where meta-paths help encode specific semantics between various node types. Despite the revolutionary representation capabilities of existing heterogeneous GNNs (HGNNs) due to their focus on improving the effectiveness of heterogeneity capturing, the huge training costs hinder their practical deployment in real-world scenarios that frequently require handling ad-hoc queries with user-defined meta-paths. To address this, we propose FHGE, a Fast Heterogeneous Graph Embedding designed for efficient, retraining-free generation of meta-path-guided graph embeddings. The key design of the proposed framework is two-fold: segmentation and reconstruction modules. It employs Meta-Path Units (MPUs) to segment the graph into local and global components, enabling swift integration of node embeddings from relevant MPUs during reconstruction and allowing quick adaptation to specific meta-paths. In addition, a dual attention mechanism is applied to enhance semantics capturing. Extensive experiments across diverse datasets demonstrate the effectiveness and efficiency of FHGE in generating meta-path-guided graph embeddings and downstream tasks, such as link prediction and node classification, highlighting its significant advantages for real-time graph analysis in ad-hoc queries.

Figures (7)

• Figure 1: Example of HetG. (a) depicts a movie network with heterogeneous nodes and relations, while (b) outlines its meta-paths. For instance, the path "a1-m1-a2" in (a) indicates that actor a1 and a2 co-starred in the movie m1, represented as the meta-path"AMA" in (b).
• Figure 2: Comparison of time and memory consumption for HGNNs on "LastFM" and "DBLP". The area of the circles indicates the memory consumption of each model, while the time costs are recorded on a log scale.
• Figure 3: (a) The overall architecture of FHGE: it first segments graph information into local and global components, samples and groups heterogeneous neighbors to balance the graph, and then reutilizes local information during the reconstruction process to facilitate meta-path-guided node embedding; (b) sampling and grouping process; (c) dual attention mechanism.
• Figure 4: Two types of embedding integration based on user-defined meta-paths.
• Figure 5: The time cost comparison on different meta-paths for the "Academic" and "IMDB" datasets. The time costs are presented on a log scale. (Blank bars indicate that the method is unable to generate meta-path-guided graph embeddings.)