Expanding Semantic Knowledge for Zero-shot Graph Embedding

TL;DR

This paper analyzes the RECT method for zero-shot graph embedding, revealing its core GNN-based semantic mapping mechanism and proposing label expansion strategies to improve discrimination and performance on unseen classes.

Contribution

It provides deep insights into RECT's working mechanism and introduces label expansion techniques to enhance zero-shot graph embedding performance.

Findings

RECT's core is a GNN prototypical model using class mean features.

Semantic space mapping connects raw features to seen and unseen classes.

Label expansion improves performance on real-world datasets.

Abstract

Zero-shot graph embedding is a major challenge for supervised graph learning. Although a recent method RECT has shown promising performance, its working mechanisms are not clear and still needs lots of training data. In this paper, we give deep insights into RECT, and address its fundamental limits. We show that its core part is a GNN prototypical model in which a class prototype is described by its mean feature vector. As such, RECT maps nodes from the raw-input feature space into an intermediate-level semantic space that connects the raw-input features to both seen and unseen classes. This mechanism makes RECT work well on both seen and unseen classes, which however also reduces the discrimination. To realize its full potentials, we propose two label expansion strategies. Specifically, besides expanding the labeled node set of seen classes, we can also expand that of unseen classes. Experiments on real-world datasets validate the superiority of our methods.