Bibliometric Analysis of Symbolic Knowledge Extraction and Explainable Artificial Intelligence in Hybrid Neuro-Symbolic Systems: Trends, Themes, and Trajectories (1991–2026)

Abstract

The application of symbolic knowledge extraction techniques to neural networks is a promising approach to combating deep learning opacity, which is particularly important in applications that require predictive and human-interpretable reasoning abilities. An extensive bibliometric analysis of 83 peer-reviewed articles indexed in the Web Of Science (WOS) Core Collection that was published from 1991 to 2026 are presented in this study. Performance metrics and science mapping techniques provided the analytical basis. The analysis shows a low annual growth rate (6.12%) with a dramatic upsurge in scientific production between 2022 and 2026. Further, the year 2025 saw the highest production of scientific output (16 articles). Moreover. top contributors are the institutions from Italy, China and the USA. Similarly, the authors who produced the most scientific output include Garcez A.A., Omicini A., Wang Z. show lasting impact. Proposed thematic clusters include Explainable AI, symbolic knowledge extraction pipelines, neuro-symbolic integration, and domain-specific applications. The paper shows that hybrid systems in which logical rules are dynamically derived from black-box models are now maturing. This review offers a longitudinal and quantitative synthesis of the literature on neuro-symbolic AI and outlines directions for future studies of scalable, explainable hybrid architectures.