Large Language Models (LLMs) often struggle with temporally grounded reasoning because their static parametric memory is misaligned with the dynamic evolution of real-world knowledge. Although RAG and Parameter-Efficient Fine-Tuning (PEFT) provide practical ways to enhance LLMs, they remain insufficient for temporal reasoning. Existing retrieval methods typically prioritize semantic relevance while overlooking temporal validity, event order, and evolutionary dependencies, resulting in evidence that is relevant in isolation but temporally fragmented. Meanwhile, conventional PEFT methods usually employ fixed and input-agnostic low-rank updates, making it difficult to capture time-dependent entity semantics and relation drift. To address these limitations, we propose Temporal Evidence-Parameter Co-Alignment (TEPCA), a framework for LLM reasoning over TKGs. TEPCA aligns LLM reasoning with temporal knowledge at both the evidence and parameter levels. At the evidence level, it introduces time-constrained Dynamic Relevance-Guided Path (DRGP) pruning to preserve globally coherent temporal paths, together with Mutual-Information (MI)-based pruning to retain query-informative local facts under temporal constraints. At the parameter level, it integrates Dynamic-Gated LoRA (DG-LoRA), a timestamp-conditioned low-rank adaptation mechanism that dynamically modulates the effective adaptation subspace according to temporal context. This design enables LLMs to better capture fine-grained entity evolution and time-dependent relation semantics while preserving parameter efficiency. Experimental results on two TKG benchmarks demonstrate that TEPCA achieves state-of-the-art performance. Additional evaluations of DG-LoRA on eight commonsense QA datasets further verify its general input-adaptive capacity, scalability across LLM backbones, and interpretability.
Figures & tables
Fig. 1 : Effect of TKGs constraints on LLM reasoning.
Fig. 2 : The framework of TEPCA. TEPCA fine-tunes LLMs via Dynamic-Gated LoRA (DG-LoRA), which injects time-adaptive low-rank updates modulated by query-conditioned gating. Query Expansion then identifies temporally relevant event chains to enhance contextual understanding. The Graph Retrieval module applies Dynamic Relevance-Guided Path (DRGP) Pruning and Mutual Information (MI)–based Pruning to distill globally consistent paths and locally informative facts from TKGs. Global paths are selected based on reverse chronological constraints (e.g., ti<t3 and t3>t2>t1 ), semantic relevance, and temporal decay, ensuring temporal consistency and causal validity. Finally, LLM Reasoning integrates the retrieved temporal subgraph into the fine-tuned LLM to produce time-sensitive and context-aware predictions.
Dataset
Entities
Relations
#Train
#Val
#Test
Granularity
ICEWS14
6,869
230
74,845
8,514
7,371
24 hours
GDELT
5,850
238
79,319
9,957
9,715
15 mins
TABLE I : Statistical information of datasets.
Method Type
Methods
ICEWS14
GDELT
Hit@1
Hit@3
Hit@10
Hit@1
Hit@3
Hit@10
Embed/Rule-based
xERTE [ 39 ]
0.330
0.454
0.570
0.085
0.159
0.265
RE-GCN [ 36 ]
0.313
0.473
0.626
0.084
0.171
0.299
TLogic [ 25 ]
0.332
0.476
0.602
0.113
0.212
0.351
TR-Rules [ 27 ]
0.339
0.485
0.611
–
–
–
RE-H 2 AN [ 65 ]
0.312
0.458
0.603
0.114
0.199
0.331
TABLE II : Link prediction results on ICEWS14 and GDELT. Bold values indicate the best results; “–” denotes unavailable data.
Methods
ICEWS14
GDELT
Hit@1
Hit@3
Hit@10
Hit@1
Hit@3
Hit@10
TEPCA (-) LoRA
0.347
0.524
0.748
0.132
0.220
0.342
TEPCA w/o QE
0.359
0.550
0.759
0.135
0.228
0.347
TEPCA w/o DRGP
0.339
0.519
0.727
0.124
0.213
0.332
TEPCA w/o MI
0.318
0.541
0.751
0.118
0.219
0.344
TEPCA
0.371
0.559
0.768
0.143
0.239
0.363
TABLE III : Ablation study on ICEWS14. The best results are in bold. “ (-) ” denotes replacing DG-LoRA with original LoRA. “ w/o ” indicates “without”.
Fig. 3 : Impact of DG-LoRA rank on both datasets.
Fig. 4 : Comparison of varying numbers of fine-tuning on ICEWS14 dataset within Hit@10 under different matching strategies. The left subplot is “ TEPCA w/ DG-LoRA ” model, and the right subplot is “ TEPCA w/ LoRA ” model.
LLMs
Methods
BoolQ
PIQA
SIQA
HellaSwag
WinoGrande
ARC-e
ARC-c
OBQA
Avg.
ChatGPT
-
73.1
85.4
68.5
78.5
66.1
89.8
79.9
74.8
77.0
Llama3.2-1B
LoRA
63.9
75.7
72.0
64.7
67.4
71.0
53.6
67.2
66.9
DG-LoRA (Ours)
63.7
75.9
72.9
74.1
68.4
71.1
54.9
68.2
68.7
Llama3.2-3B
LoRA
71.5
83.2
75.4
86.3
79.2
82.5
70.1
79.6
78.5
DG-LoRA (Ours)
71.5
84.5
78.8
92.3
81.2
83.8
72.4
80.0
80.5
Llama2-7B
LoRA
69.8
79.9
79.5
83.6
82.6
79.8
64.7
81.0
77.6
TABLE IV : Experimental results on eight commonsense reasoning datasets using various LoRA methods for Llama3.2-1B/3B, Llama2-7B, Qwen2.5-7B, and Llama3-8B. Notably, the rank value is 32 ( r=32 ) and best indicates the highest average metric value among all methods.
Fig. 5 : The low-rank subspace heatmap (the number of gatec=0 ) of the Query ( qproj ) and Value ( vproj ) mappings for ARC, BoolQ, and OBQA question texts, based on the LLaMA-3.2-3B models ( r=32 ).
Temporal knowledge graphs (TKGs) support reasoning over time-evolving facts, yet state-of-the-art models are often computationally heavy and costly to deploy. Existing compression and distillation techniques are largely designed for static graphs; directly applying them to temporal settings may overlook time-dependent interactions and lead to performance degradation. We propose an LLM-assisted distillation framework specifically designed for temporal knowledge graph reasoning. Beyond a conventional high-capacity temporal teacher, we incorporate a large language model as an auxiliary instructor to provide enriched supervision. The LLM supplies broad background knowledge and temporally informed signals, enabling a lightweight student to better model event dynamics without increasing inference-time complexity. Training is conducted by jointly optimizing supervised and distillation objectives, using a staged alignment strategy to progressively integrate guidance from both teachers. Extensive experiments on multiple public TKG benchmarks with diverse backbone architectures demonstrate that the proposed approach consistently improves link prediction performance over strong distillation baselines, while maintaining a compact and efficient student model. The results highlight the potential of large language models as effective teachers for transferring temporal reasoning capability to resource-efficient TKG systems.
Wang Xing, Wei Song, Siyu Lin +2
School of Computer Science and Technology, Xidian University · School of Computing and Artificial Intelligence, Southwest Jiaotong University · School of Information Science and Engineering, Chongqing Jiaotong University
Large language models have demonstrated strong reasoning capabilities in general knowledge question answering. However, their ability to handle temporal information remains limited. To address this limitation, existing approaches often involve external tools or manual verification and are tailored to specific scenarios, leading to poor generalizability. Moreover, these methods apply a fixed pipeline to all questions, overlooking the fact that different types of temporal questions require distinct reasoning strategies, which leads to unnecessary processing for simple cases and inadequate reasoning for complex ones. To this end, we propose AdapTime, an adaptive temporal reasoning method that dynamically executes reasoning steps based on the input context. Specifically, it involves three temporal reasoning actions: reformulate, rewrite and review, with an LLM planner guiding the reasoning process. AdapTime integrates seamlessly with state-of-the-art LLMs and significantly enhances their temporal reasoning capabilities without relying on external support. Extensive experiments demonstrate the effectiveness of our approach.
Yimin Deng, Yejing Wang, Zhenxi Lin +8
Xi’an Jiaotong University · City University of Hong Kong · Tencent Jarvis Lab +1
Retrieval-Augmented Generation (RAG) has shown strong effectiveness in grounding Large Language Models (LLMs) with external knowledge. However, existing RAG and Graph RAG frameworks largely treat knowledge as static or associate time with coarse-grained timestamps or metadata, failing to capture rich temporal structures such as duration, overlap, and containment. We propose IA-RAG, a hierarchical temporal RAG framework that models knowledge as time intervals and performs retrieval under formal temporal constraints. IA-RAG represents facts as Interval Event Units (IEUs) and organizes them into a hierarchical Thematic Forest, where temporal dependencies are governed by Allen's Interval Algebra. To handle incomplete or uncertain temporal boundaries, IA-RAG further introduces a Sub-graph Time Tightening mechanism that refines fuzzy intervals through logical constraints within connected event subgraphs. In addition, IA-RAG supports implicit temporal semantic retrieval through interval-algebra-guided traversal. Experiments on multiple temporal question answering benchmarks, including TimeQA, TempReason, and ComplexTR, demonstrate that IA-RAG achieves strong temporal retrieval and reasoning performance, particularly on complex compositional temporal reasoning tasks. Our code is released at https://github.com/xiaoAugenstern/LogicalRAG_TemporalQA.
Xiaoman Wang, Yaoze Zhang, Wenzhuo Fan +7
East China Normal University · Shanghai Artificial Intelligence Laboratory · University of Shanghai for Science and Technology +1