Elamur Research⌕
☰
home / papers / 2026 / From Implicit Graph Encoding to Explicit Evidence: A Training-Free LLM Framework for Temporal Knowledge Graph Reasoning

From Implicit Graph Encoding to Explicit Evidence: A Training-Free LLM Framework for Temporal Knowledge Graph Reasoning

Paper · 2026 · Guo Tang, Ke Cheng, Huiming Fan et al.

Summary

Forecasting future facts in Temporal Knowledge Graphs (TKGs) is difficult for parametric models. They degrade under distribution shift and cannot handle unseen entities or relations without retraining. Existing LLM-based alternatives have their own problems: some flatten graphs into long text and lose information, while others depend on soft prompts or adapters co-trained with GNNs.

The authors propose ExE-LLM, a training-free framework that separates graph computation from language reasoning. A lightweight heuristic module computes structural, temporal and semantic signals and rewrites them as natural-language evidence. A frozen LLM then weighs this evidence to choose the missing tail entity. The pipeline has three parts:

  • a Task-Aware Scheduler that sets per-query recall budgets and fusion weights;
  • a Heuristic Evidence Synthesizer that retrieves and ranks candidates;
  • a Self-Diagnosis Module that turns wrong predictions into reusable cases.

On ICEWS14, ICEWS05-15, ICEWS1819 and GDELT, the method reaches state-of-the-art inductive results. It outperforms fully trained GNNs without updating any LLM parameters. This points to explicit graph-to-text evidence as a viable way to reason over evolving graphs.

Key points

  • The scheduler derives three signals for each query node: structural popularity (log of degree + 1), temporal recency (time since last activity) and semantic ambiguity (spread of similarity to top semantic neighbours). From these it sets the budgets k_str, k_tem, k_sem and the weights α, β.
  • Candidates come from three recall paths: nodes with the most common neighbours, recently activated nodes, and nearest neighbours in a pretrained embedding space. They are then scored bidirectionally, with the query node's history weighted by exponential time decay.
  • The evaluation covers link prediction and node retrieval (following Zhang et al., 2024), measured with Hit@1 and Hit@10. Ablations that remove the LLM reasoner (Tables 5–6) isolate its contribution.
  • Stated limitation: on ranking tasks, the non-finetuned LLM shows high output variance. Averaging runs and using larger models helped little, so the authors suggest a post-hoc classifier head and gating as future work.
  • Code is released at github.com/JENLISA4EVER/ExE-LLM.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Tsinghua UniversityBeijing · China333 works · 87,019 citations🎓Harbin Institute of TechnologyHarbin · China158 works · 28,841 citations🎓Beihang UniversityBeijing · China65 works · 8,452 citations🔬Peng Cheng LaboratoryShenzhen · China38 works · 2,504 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 3Research subjects and languages the work is about.Language models & pretraining · 3,320 worksReasoning · 1,426 worksKnowledge graphs & ontologies · 192 works

Cite

Cite as Guo Tang, Ke Cheng, Huiming Fan, Heng Chang, Wenxiang Zheng, Xianhao Ou, Junjia Xiang, Ming Liu, Yujun Zhou, Li Lanyu and Bing Qin. 2026. From Implicit Graph Encoding to Explicit Evidence: A Training-Free LLM Framework for Temporal Knowledge Graph Reasoning. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.266

@inproceedings{tang2026implicit,
  title = {{From Implicit Graph Encoding to Explicit Evidence: A Training-Free LLM Framework for Temporal Knowledge Graph Reasoning}},
  author = {Guo Tang, Ke Cheng, Huiming Fan, Heng Chang, Wenxiang Zheng, Xianhao Ou, Junjia Xiang, Ming Liu, Yujun Zhou, Li Lanyu and Bing Qin},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.266},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.266}
}

References

46 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Yuxiang Cai, Qiao Liu, Yanglei Gan, Changlin Li, Xueyi Liu, Run Lin, Da Luo, and JiayeYang JiayeYang. 2024. Predicting the unpredictable: Uncertainty-aware reasoning over temporal knowledge graphs via diffusion process. In Findings of the Association for Computational Linguistics: ACL 2024, pages 57… find
  2. He Chang, Jie Wu, Zhulin Tao, Yunshan Ma, Xianglin Huang, and Tat-Seng Chua. 2025. Integrate temporal graph learning into llm-based temporal knowledge graph model. arXiv preprint arXiv:2501.11911. — [arXiv:2501.11911](https://arxiv.org/abs/2501.11911) find
  3. Zifeng Ding, Heling Cai, Jingpei Wu, Yunpu Ma, Ruotong Liao, Bo Xiong, and Volker Tresp. 2024. zrllm: Zero-shot relational learning on temporal knowledge graphs with large language models. Preprint, arXiv:2311.10112. find
  4. Alberto García-Durán, Sebastijan Dumančić, and Mathias Niepert. 2018. Learning sequence encoders for temporal knowledge graph completion. arXiv preprint arXiv:1809.03202. find
  5. Jinhao Jiang, Kun Zhou, Zican Dong, Keming Ye, Wayne Xin Zhao, and Ji-Rong Wen. 2023. Structgpt: A general framework for large language model to reason over structured data. Preprint, arXiv:2305.09645. find
  6. Zihao Jiang, Ben Liu, Miao Peng, Wenjie Xu, Yao Xiao, Zhenyan Shan, and Min Peng. 2025. Towards explainable temporal reasoning in large language models: A structure-aware generative framework. arXiv preprint arXiv:2505.15245. find
  7. Woojeong Jin, Meng Qu, Xisen Jin, and Xiang Ren. 2019. Recurrent event network: Autoregressive structure inference over temporal knowledge graphs. arXiv preprint arXiv:1904.05530. — [arXiv:1904.05530](https://arxiv.org/abs/1904.05530) find
  8. Srijan Kumar, Xikun Zhang, and Jure Leskovec. 2019. Predicting dynamic embedding trajectory in temporal interaction networks. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, page 1269–1278. ACM. find
  9. Julien Leblay and Melisachew Wudage Chekol. 2018. Deriving validity time in knowledge graph. In Companion of the The Web Conference 2018 on The Web Conference 2018 - WWW ’18. find
  10. Dong-Ho Lee, Kian Ahrabian, Woojeong Jin, Fred Morstatter, and Jay Pujara. 2023. Temporal knowledge graph forecasting without knowledge using in-context learning. arXiv preprint arXiv:2305.10613. find
  11. Kalev Leetaru and Philip A Schrodt. 2013. Gdelt: Global data on events, location, and tone, 1979–2012. In ISA annual convention, volume 2, pages 1–49. Citeseer. find
  12. Runlin Lei, Jiarui Ji, Haipeng Ding, Lu Yi, Zhewei Wei, Yongchao Liu, and Chuntao Hong. 2025. Exploring the potential of large language models as predictors in dynamic text-attributed graphs. Preprint, arXiv:2503.03258. — [arXiv:2503.03258](https://arxiv.org/abs/2503.03258) find
  13. Yujia Li, Shiliang Sun, and Jing Zhao. 2022. Tirgn: Time-guided recurrent graph network with local-global historical patterns for temporal knowledge graph reasoning. In IJCAI, pages 2152–2158. find
  14. Zixuan Li, Xiaolong Jin, Wei Li, Saiping Guan, Jiafeng Guo, Huawei Shen, Yuanzhuo Wang, and Xueqi Cheng. 2021. Temporal knowledge graph reasoning based on evolutional representation learning. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Ret… find
  15. Ruotong Liao, Xu Jia, Yangzhe Li, Yunpu Ma, and Volker Tresp. 2024. Gentkg: Generative forecasting on temporal knowledge graph with large language models. Preprint, arXiv:2310.07793. find
  16. Zhenghao Liu, Chenyan Xiong, Maosong Sun, and Zhiyuan Liu. 2018. Entity-duet neural ranking: Understanding the role of knowledge graph semantics in neural information retrieval. Cornell University -arXiv,Cornell University - arXiv. find
  17. Xiaodong Lu, Leilei Sun, Tongyu Zhu, and Weifeng Lv. 2024. Improving temporal link prediction via temporal walk matrix projection. Preprint, arXiv:2410.04013. — [arXiv:2410.04013](https://arxiv.org/abs/2410.04013) find
  18. Aldo Pareja, Giacomo Domeniconi, Jie Chen, Tengfei Ma, Toyotaro Suzumura, Hiroki Kanezashi, Tim Kaler, Tao B. Schardl, and Charles E. Leiserson. 2019. Evolvegcn: Evolving graph convolutional networks for dynamic graphs. Preprint, arXiv:1902.10191. — [arXiv:1902.10191](https://arxiv.org/abs/1902.10191) find
  19. Matthew E. Peters, Mark Neumann, Robert L. Logan IV, Roy Schwartz, Vidur Joshi, Sameer Singh, and Noah A. Smith. 2019. Knowledge enhanced contextual word representations. Preprint, arXiv:1909.04164. find
  20. Emanuele Rossi, Ben Chamberlain, Fabrizio Frasca, Davide Eynard, Federico Monti, and Michael Bronstein. 2020. Temporal graph networks for deep learning on dynamic graphs. Preprint, arXiv:2006.10637. — [arXiv:2006.10637](https://arxiv.org/abs/2006.10637) find
  21. Aravind Sankar, Yanhong Wu, Liang Gou, Wei Zhang, and Hao Yang. 2019. Dynamic graph representation learning via self-attention networks. Preprint, arXiv:1812.09430. — [arXiv:1812.09430](https://arxiv.org/abs/1812.09430) find
  22. Xiaoming Shi, Siqiao Xue, Kangrui Wang, Fan Zhou, James Y. Zhang, Jun Zhou, Chenhao Tan, and Hongyuan Mei. 2023. Language models can improve event prediction by few-shot abductive reasoning. Preprint, arXiv:2305.16646.
  23. Qingyu Tan, Hwee Tou Ng, and Lidong Bing. 2023. Towards benchmarking and improving the temporal reasoning capability of large language models. Preprint, arXiv:2306.08952. find
  24. Guo Tang, Zheng Chu, Wenxiang Zheng, Junjia Xiang, Yizhuo Li, Weihao Zhang, Ming Liu, and Bing Qin. 2025. AnRe: Analogical replay for temporal knowledge graph forecasting. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 4632–4… find
  25. Rakshit Trivedi, Mehrdad Farajtabar, Prasenjeet Biswal, and Hongyuan Zha. 2018. Representation learning over dynamic graphs. Preprint, arXiv:1803.04051. — [arXiv:1803.04051](https://arxiv.org/abs/1803.04051) find
  26. Rakshit Trivedi, Mehrdad Farajtabar, Prasenjeet Biswal, and Hongyuan Zha. 2019. Dyrep: Learning representations over dynamic graphs. International Conference on Learning Representations (ICLR). find
  27. Lu Wang, Xiaofu Chang, Shuang Li, Yunfei Chu, Hui Li, Wei Zhang, Xiaofeng He, Le Song, Jingren Zhou, and Hongxia Yang. 2021. Tcl: Transformer-based dynamic graph modelling via contrastive learning. Preprint, arXiv:2105.07944. — [arXiv:2105.07944](https://arxiv.org/abs/2105.07944) find
  28. Xiang Wang, Xiangnan He, Yixin Cao, Meng Liu, and Tat-Seng Chua. 2019. Kgat: Knowledge graph attention network for recommendation. In Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining, pages 950–958. find
  29. Yanbang Wang, Yen-Yu Chang, Yunyu Liu, Jure Leskovec, and Pan Li. 2022. Inductive representation learning in temporal networks via causal anonymous walks. Preprint, arXiv:2101.05974. find
  30. Yuwei Xia, Ding Wang, Qiang Liu, Liang Wang, Shu Wu, and Xiao-Yu Zhang. 2024. Chain-of-history reasoning for temporal knowledge graph forecasting. In Findings of the Association for Computational Linguistics: ACL 2024, pages 16144–16159, Bangkok, Thailand. Association for Computational Linguistics. find
  31. Sheng Xiang, Dawei Cheng, Chencheng Shang, Ying Zhang, and Yuqi Liang. 2022. Temporal and heterogeneous graph neural network for financial time series prediction. find
  32. Peng Xiao, Chao Liu, Wei Jia, and Lijun Dong. 2025. Aligned-entities-based fusion embedding on hetero-field knowledge graphs. DATA INTELLIGENCE, 7(3):618–635. find
  33. Da Xu, Chuanwei Ruan, Evren Korpeoglu, Sushant Kumar, and Kannan Achan. 2020. Inductive representation learning on temporal graphs. Preprint, arXiv:2002.07962. — [arXiv:2002.07962](https://arxiv.org/abs/2002.07962) find
  34. Wenjie Xu, Ben Liu, Miao Peng, Xu Jia, and Min Peng. 2024. Pre-trained language model with prompts for temporal knowledge graph completion. Preprint, arXiv:2305.07912. find
  35. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jing Zhou, Jin… — Qwen3 Technical Report
  36. Lixiao Zheng, Jipeng Xiao, Shuai Ma, Zuxi Chen, and Xiangyu Luo. 2025. Temporal cycle enumeration for detecting financial fraud. DATA INTELLIGENCE, 7(3):567–590. find
  37. Linyao Yang, Hongyang Chen, Zhao Li, Xiao Ding, and Xindong Wu. 2023. Chatgpt is not enough: Enhancing large language models with knowledge graphs for fact-aware language modeling. arXiv preprint arXiv:2306.11489. find
  38. Linyao Yang, Hongyang Chen, Zhao Li, Xiao Ding, and Xindong Wu. 2024. Give us the facts: Enhancing large language models with knowledge graphs for fact-aware language modeling. Preprint, arXiv:2306.11489. find
  39. Junjie Ye, Nuo Xu, Yikun Wang, Jie Zhou, Qi Zhang, Tao Gui, and Xuanjing Huang. 2024. Llm-da: Data augmentation via large language models for few-shot named entity recognition. Preprint, arXiv:2402.14568. — [arXiv:2402.14568](https://arxiv.org/abs/2402.14568) find
  40. Le Yu, Leilei Sun, Bowen Du, and Weifeng Lv. 2023. Towards better dynamic graph learning: New architecture and unified library. Preprint, arXiv:2303.13047. — [arXiv:2303.13047](https://arxiv.org/abs/2303.13047) find
  41. Xuanqing Yu, Wangtao Sun, Jingwei Li, Kang Liu, Chengbao Liu, and Jie Tan. 2024. Onsep: A novel online neural-symbolic framework for event prediction based on large language model. Preprint, arXiv:2408.07840. find
  42. Chenhan Yuan, Qianqian Xie, Jimin Huang, and Sophia Ananiadou. 2023. Back to the future: Towards explainable temporal reasoning with large language models. Preprint, arXiv:2310.01074. find
  43. Chenhan Yuan, Qianqian Xie, Jimin Huang, and Sophia Ananiadou. 2024. Back to the future: Towards explainable temporal reasoning with large language models. In Proceedings of the ACM on Web Conference 2024, pages 1963–1974. find
  44. Jiasheng Zhang, Jialin Chen, Menglin Yang, Aosong Feng, Shuang Liang, Jie Shao, and Rex Ying. 2024. Dtgb: A comprehensive benchmark for dynamic text-attributed graphs. Preprint, arXiv:2406.12072. — [arXiv:2406.12072](https://arxiv.org/abs/2406.12072) find
  45. Shengyu Zhang, Linfeng Dong, Xiaoya Li, Sen Zhang, Xiaofei Sun, Shuhe Wang, Jiwei Li, Runyi Hu, Tianwei Zhang, Fei Wu, and Guoyin Wang. 2025. Instruction tuning for large language models: A survey. Preprint, arXiv:2308.10792.
  46. Dataset \\\# Entity \\\# Relation Train Valid Test ICEWS14 ICEWS05-15 ICEWS1819 GDELT 12,498 10,094 31,796 7,691 260 251 266 240 323,895 368,868 770,050 1,734,399 -46,302 165,011 238,765 341,409 46,159 165,010 305,241 *Table 7: Statistics of the datasets.* find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tang-2026-implicit-graph-encoding-explicit-evidence/