RADAR: Reasoning as Discrimination with Aligned Representations for LLM-based Knowledge Graph Reasoning

2026-02-25

AI Summary

Current large language models often rely on superficial co-occurrence patterns in knowledge graph reasoning, struggling to capture genuine relational semantics, which limits their out-of-distribution generalization. This work proposes a discriminative relation reasoning framework that, for the first time, formulates knowledge graph reasoning as an entity selection task. By aligning representation spaces and enhancing entity separability through reinforcement learning, the model performs reasoning directly in this discriminative space, thereby avoiding hallucinatory generations. The approach substantially improves the robustness and transferability of relational semantics, achieving relative performance gains of 5–6% in link prediction and triple classification across four benchmark datasets, along with a 62.9% increase in task-relevant mutual information in intermediate representations.

Technology Category

• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Reasoning under Uncertainty: Relational Probabilistic Models
• Machine Learning: Statistical Relational/Logic Learning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Knowledge graph reasoning (KGR) infers missing facts, with recent advances increasingly harnessing the semantic priors and reasoning abilities of Large Language Models (LLMs). However, prevailing generative paradigms are prone to memorizing surface-level co-occurrences rather than learning genuine relational semantics, limiting out-of-distribution generalization. To address this, we propose RADAR, which reformulates KGR from generative pattern matching to discriminative relational reasoning. We recast KGR as discriminative entity selection, where reinforcement learning enforces relative entity separability beyond token-likelihood imitation. Leveraging this separability, inference operates directly in representation space, ensuring consistency with the discriminative optimization and bypassing generation-induced hallucinations. Across four benchmarks, RADAR achieves 5-6% relative gains on link prediction and triple classification over strong LLM baselines, while increasing task-relevant mutual information in intermediate representations by 62.9%, indicating more robust and transferable relational reasoning.

Problem

• Knowledge Graph Reasoning
• Large Language Models
• Out-of-Distribution Generalization
• Relational Semantics
• Generative Paradigms

Innovation

• discriminative reasoning
• representation alignment
• reinforcement learning
• knowledge graph reasoning
• large language models