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

TL;DR

RADAR transforms knowledge graph reasoning with LLMs from surface-level pattern matching to discriminative relational reasoning, significantly improving out-of-distribution generalization and robustness in link prediction and triple classification tasks.

Contribution

RADAR introduces a discriminative entity selection framework using reinforcement learning, enhancing relational reasoning and representation separability in LLM-based knowledge graph inference.

Findings

Achieves 5-6% relative gains on benchmarks.

Increases task-relevant mutual information by 62.9%.

Enhances out-of-distribution generalization.

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.