OLGA: One-cLass Graph Autoencoder

TL;DR

The paper tackles one-class node classification on graphs by identifying gaps in existing approaches: non-customized representations, unconstrained hypersphere learning, and limited interpretability. It introduces OLGA, an end-to-end graph autoencoder that simultaneously learns node embeddings and classifies interest nodes using a novel hypersphere loss ${\mathcal L}_1$ plus reconstruction losses ${\mathcal L}_2$ and ${\mathcal L}_3$ within a multi-task objective ${\mathcal L}(\mathbf{W}) = \alpha {\mathcal L}_1 + \beta {\mathcal L}_2 + \delta {\mathcal L}_3$. OLGA enables low-dimensional, interpretable representations while maintaining strong classification performance, demonstrated across eight diverse datasets with statistically significant improvements over several baselines. The work emphasizes interpretability and visualization for one-class graph learning and shows OLGA’s potential for practical OCL applications on graphs.

Abstract

One-class learning (OCL) comprises a set of techniques applied when real-world problems have a single class of interest. The usual procedure for OCL is learning a hypersphere that comprises instances of this class and, ideally, repels unseen instances from any other classes. Besides, several OCL algorithms for graphs have been proposed since graph representation learning has succeeded in various fields. These methods may use a two-step strategy, initially representing the graph and, in a second step, classifying its nodes. On the other hand, end-to-end methods learn the node representations while classifying the nodes in one learning process. We highlight three main gaps in the literature on OCL for graphs: (i) non-customized representations for OCL; (ii) the lack of constraints on hypersphere parameters learning; and (iii) the methods' lack of interpretability and visualization. We propose One-cLass Graph Autoencoder (OLGA). OLGA is end-to-end and learns the representations for the graph nodes while encapsulating the interest instances by combining two loss functions. We propose a new hypersphere loss function to encapsulate the interest instances. OLGA combines this new hypersphere loss with the graph autoencoder reconstruction loss to improve model learning. OLGA achieved state-of-the-art results and outperformed six other methods with a statistically significant difference from five methods. Moreover, OLGA learns low-dimensional representations maintaining the classification performance with an interpretable model representation learning and results.

Figures (6)

• Figure 1: OLGA illustration with our new hypersphere and GAE loss functions.
• Figure 2: Our new hypersphere loss function ($\mathcal{L}_{1}$) Illustration.
• Figure 3: Critical difference diagram of Friedman's statistical test with Nemenyi post-test considering $f_1$-macro for low and high dimensional scenarios.
• Figure 4: Two-dimensional representations of OLGA last layer consider the learned representations in three datasets. The colors indicate the interest class (blue) and the non-interest class (green).
• Figure 5: Relations between the hypersphere volume and the dimension of the representation used.