Hybrid Neuro-Symbolic Reasoning for Context-Aware Decision Support Systems

Abstract

Decision support systems must increasingly combine statistical, learned judgment with explicit, auditable rules, yet neural and
symbolic AI have historically developed as separate research programs. This article synthesizes the literature on hybrid
neuro-symbolic reasoning for context-aware decision support: architectures that pair a neural component, responsible for
perception and pattern recognition, with a symbolic or differentiably-relaxed logical component, responsible for rule-governed,
explainable inference. We trace the technique from its roots in rule-based expert systems, case-based reasoning, and fuzzy
logic, through the neural learning advances that exposed the limits of purely connectionist reasoning, to concrete coupling
mechanisms including differentiable first-order logic, neural probabilistic logic programming, neural theorem proving,
neuro-symbolic concept learning, abductive learning, and neurosymbolic knowledge graph reasoning. We connect this
architectural literature to the largely independent tradition of context-aware computing, and examine the applied evidence,
strongest in healthcare, for where hybrid reasoning has actually reached deployed or pilot-scale decision support. Read
together, the evidence supports hybrid neuro-symbolic reasoning as a coherent and technically mature pattern for
context-aware decision support, while leaving symbolic-component scalability, the distinction between jointly-trained and
post-hoc hybrids, and large-scale real-world validation as an open research agenda.