Glassbox Models for Graph Data: Taxonomy, Illustrative Analysis, and Perspectives on an Emerging Direction

Glassbox Models for Graph Data: Taxonomy, Illustrative Analysis, and Perspectives on an Emerging Direction

Soloviov M.V. (ISP RAS, Moscow, Russia; MIPT, Dolgoprudny, Moscow Region, Russia)
Drobyshevskiy M.D. (ISP RAS, Moscow, Russia; MIPT, Dolgoprudny, Moscow Region, Russia)
Moiseikin A.D. (ISP RAS, Moscow, Russia; MSU, Moscow, Russia)
Poltavtsev E.Y. (MIPT, Dolgoprudny, Moscow Region, Russia; FRCCSC, Moscow, Russia)
Poturilo E.N. (MIPT, Dolgoprudny, Moscow Region, Russia; FRCCSC, Moscow, Russia)
Khalko I.S. (MIPT, Dolgoprudny, Moscow Region, Russia; FRCCSC, Moscow, Russia)

Abstract

Despite the high effectiveness of graph neural networks across a wide range of tasks involving graph-structured data, their use in high-stakes domains remains limited due to insufficient interpretability. In this context, glassbox models are of particular interest, since the prediction process in such models is transparent and does not require external post-hoc explanation methods. However, while glassbox approaches for tabular data have an established classification and are commonly associated with additive, tree-based, and rule-based models, a comparable systematization for graph data is still emerging. We present a taxonomy and systematization of glassbox approaches for graph data, distinguishing additive, tree-based, and rule-based classes. We also present an illustrative analysis demonstrating that the transparency of the prediction mechanism in glassbox models for graph data makes it possible to trace the logic of decision-making and to use it for graph data analysis. The paper discusses the limitations of existing glassbox models for graph data and the prospects for further development of this direction. The analysis suggests that glassbox models for graph data can be regarded as a conceptually important and promising emerging direction of interpretable machine learning for graph data.

Keywords

graph machine learning; interpretable machine learning; glassbox models; graph neural networks; additive models; decision trees; rule-based models.

Edition

Proceedings of the Institute for System Programming, vol. 38, issue 6, part 2, 2026, pp. 93-116

ISSN 2220-6426 (Online), ISSN 2079-8156 (Print).

DOI: 10.15514/ISPRAS-2026-38(6)-27

For citation

Soloviov M.V., Drobyshevskiy M.D., Moiseikin A.D., Poltavtsev E.Y., Poturilo E.N., Khalko I.S. Glassbox Models for Graph Data: Taxonomy, Illustrative Analysis, and Perspectives on an Emerging Direction. Proceedings of the Institute for System Programming, vol. 38, issue 6, part 2, 2026, pp. 93-116 DOI: 10.15514/ISPRAS-2026-38(6)-27.

Full text of the paper in pdf (in Russian) Back to the contents of the volume