Self-Exploring Language Models for Explainable Link Forecasting on Temporal Graphs via Reinforcement Learning

2025-08-31

AI Summary

Existing temporal graph neural networks lack interpretability and generalizability to unseen graph structures; meanwhile, large language models (LLMs) in temporal graph reasoning remain confined to static or synthetic small-scale graphs, with no systematic evaluation of reasoning trace quality. Method: We propose ReaL-TG—a novel framework that introduces reinforcement learning to temporal graph link prediction for the first time—enabling an LLM (Qwen3-4B) to autonomously discover interpretable, history-aware reasoning paths. It integrates structural state modeling with outcome-oriented reward design and employs a dual-track evaluation protocol combining ranking metrics and LLM-as-a-Judge assessment. Contribution/Results: ReaL-TG-4B achieves state-of-the-art predictive performance on real-world temporal graphs, outperforming larger models such as GPT-5 mini. Its generated explanations demonstrate high fidelity and low hallucination rates, validated by both automated and human evaluations.

Technology Category

• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Machine Learning: Graph-based Machine Learning
• Planning, Routing, and Scheduling: Temporal Planning

Application Category

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

Abstract

Forecasting future links is a central task in temporal graph (TG) reasoning, requiring models to leverage historical interactions to predict upcoming ones. Traditional neural approaches, such as temporal graph neural networks, achieve strong performance but lack explainability and cannot be applied to unseen graphs without retraining. Recent studies have begun to explore using large language models (LLMs) for graph reasoning, but most of them are constrained to static graphs or small synthetic TGs and lack the evaluation of the quality of reasoning traces generated by LLMs. In this work, we present Reasoning-Enhanced Learning for Temporal Graphs (ReaL-TG), a reinforcement learning framework that fine-tunes LLMs to perform explainable link forecasting on real-world TGs. ReaL-TG uses outcome-based reward to encourage models to self-explore reasoning strategies from graph structure and to produce explanations that directly justify their predictions. To enable evaluation on LLM-generated reasoning traces, we propose a new evaluation protocol combining ranking metrics with an LLM-as-a-Judge system that assesses both the quality of reasoning and the impact of hallucinations. Experiments with ReaL-TG-4B, obtained by fine-tuning Qwen3-4B under our framework, show that it outperforms much larger frontier LLMs, including GPT-5 mini, on ranking metrics, while producing high-quality explanations confirmed by both the LLM judge and human evaluation.

Problem

• Enabling explainable link forecasting on temporal graphs using reinforcement learning
• Addressing lack of explainability in traditional temporal graph neural networks
• Evaluating quality of reasoning traces and hallucinations in LLM predictions

Innovation

• Reinforcement learning fine-tunes LLMs
• Self-exploring reasoning strategies from graphs
• LLM-as-a-Judge evaluates reasoning quality