MemoTime: Memory-Augmented Temporal Knowledge Graph Enhanced Large Language Model Reasoning

TL;DR

MemoTime presents a memory-augmented temporal knowledge graph framework that enhances large language model reasoning over time by grounding questions in a TKG, decomposing complex queries via a Tree of Time, performing operator-aware evidence retrieval, and continually updating an experience memory with verified reasoning traces. The approach enforces temporal fidelity through monotonic timestamps and global synchronization across multiple entities, while adaptively invoking temporal toolkits to handle diverse operators. Extensive experiments on MultiTQ and TimeQuestions show state-of-the-art performance across backbone LLMs, including strong gains over baselines and enabling smaller models to approach GPT-4-Turbo capabilities. The work also provides an ablation analysis that reveals the pivotal roles of graph-based retrieval, hierarchical decomposition, and memory, and demonstrates the framework’s efficiency and generality as a plug-in module for temporal reasoning tasks.

Abstract

Large Language Models (LLMs) have achieved impressive reasoning abilities, but struggle with temporal understanding, especially when questions involve multiple entities, compound operators, and evolving event sequences. Temporal Knowledge Graphs (TKGs), which capture vast amounts of temporal facts in a structured format, offer a reliable source for temporal reasoning. However, existing TKG-based LLM reasoning methods still struggle with four major challenges: maintaining temporal faithfulness in multi-hop reasoning, achieving multi-entity temporal synchronization, adapting retrieval to diverse temporal operators, and reusing prior reasoning experience for stability and efficiency. To address these issues, we propose MemoTime, a memory-augmented temporal knowledge graph framework that enhances LLM reasoning through structured grounding, recursive reasoning, and continual experience learning. MemoTime decomposes complex temporal questions into a hierarchical Tree of Time, enabling operator-aware reasoning that enforces monotonic timestamps and co-constrains multiple entities under unified temporal bounds. A dynamic evidence retrieval layer adaptively selects operator-specific retrieval strategies, while a self-evolving experience memory stores verified reasoning traces, toolkit decisions, and sub-question embeddings for cross-type reuse. Comprehensive experiments on multiple temporal QA benchmarks show that MemoTime achieves overall state-of-the-art results, outperforming the strong baseline by up to 24.0%. Furthermore, MemoTime enables smaller models (e.g., Qwen3-4B) to achieve reasoning performance comparable to that of GPT-4-Turbo.

Figures (9)

• Figure 1: Representative workflow of four LLM reasoning paradigms.
• Figure 2: Overview of the MemoTime framework. Temporal Grounding: Topic entities and temporal operators are extracted from the input question to construct a question-specific subgraph. Tree of Time (Hierarchical Reasoning): Recursively decomposes the question into sub-questions guided by temporal dependencies, adaptively reusing experience or invoking toolkits for new evidence. Temporal Evidence Retrieval and Pruning: Performs operator-aware retrieval under monotonic time constraints, followed by semantic-temporal re-ranking and LLM-based sufficiency verification. Experience Memory: Verified reasoning traces are stored, updated, and retrieved for cross-type reuse, enabling continual self-improvement across reasoning cycles.
• Figure 3: The accuracy of MemoTime on MultiTQ by varying different $D_{\max}$ and different LLM backbones.
• Figure 4: TemporalGrounding
• Figure 5: TreeofTimeReasoning

Theorems & Definitions (4)

• Definition 1: Temporal Path
• Example 1: Temporal Path
• Definition 2: Temporal Reasoning Path
• Example 2: Temporal Reasoning Path