A Comparative Analysis of Neuro-symbolic Methods for Link Prediction

Link prediction on knowledge graphs is relevant to various applications, such as recommendation systems, question answering, and entity search. This task has been approached from different perspectives: symbolic methods leverage rule-based reasoning but struggle with scalability and noise, while knowledge graph embeddings (KGE) represent entities and relations in a continuous space, enabling scalability but often neglecting logical constraints from ontologies. Recently, neuro-symbolic approaches have emerged to bridge this gap by integrating embedding-based learning with symbolic reasoning. This paper provides a structured review of state-of-the-art neuro-symbolic methods for link prediction. Beyond a qualitative analysis, a key contribution of this work is a comprehensive experimental benchmarking, where we systematically compare these methods on the same datasets using a broader set of evaluation metrics. This unified experimental setup allows for a fair assessment of their strengths and limitations, bringing elements of answers to key questions: How accurate are these methods? How scalable are they? To what extent do they respect logical formulas? Additionally, we introduce a novel metric Contradicts@k that evaluates how well model predictions align with ontological rules, further assessing the integration of symbolic reasoning in link prediction.