--- title: "故事启发大语言模型的时序知识图谱预测 — 陈娟 2024" description: "Temporal knowledge graphs (TKGs) record facts as (head, relation, tail, timestamp) quadruples." authors: ["陈娟", "赵新潮", "隋京言", "祁麟", "田辰", "庞亮", "方金云"] year: 2024 venue: "模式识别与人工智能 (Pattern Recognition and Artificial Intelligence)" source: "http://manu46.magtech.com.cn/Jweb_prai/CN/abstract/abstract12769.shtml" doi: "10.16451/j.cnki.issn1003-6059.202408005" licence: "Free access (manu46.magtech.com.cn), no open licence stated — readonly; © 《模式识别与人工智能》编辑部" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/papers/chen-2024-temporal-knowledge-graph/" --- [Elamur Research](https://md.research.elamur.ai/) · [this page in HTML](https://research.elamur.ai/free/papers/chen-2024-temporal-knowledge-graph/) [home](https://md.research.elamur.ai/) / [papers](https://md.research.elamur.ai/free/papers/) / [2024](https://md.research.elamur.ai/free/papers/2024/page/4/) / 故事启发大语言模型的时序知识图谱预测 # 故事启发大语言模型的时序知识图谱预测 **Authors:** 陈娟, 赵新潮, 隋京言, 祁麟, 田辰, 庞亮, 方金云 · **Year:** 2024 · **Venue:** 模式识别与人工智能 (Pattern Recognition and Artificial Intelligence) · **Licence:** Free access (manu46.magtech.com.cn), no open licence stated — readonly; © 《模式识别与人工智能》编辑部 **Links:** [Publisher's page](https://doi.org/10.16451/j.cnki.issn1003-6059.202408005) · [PDF](http://manu46.magtech.com.cn/Jweb_prai/CN/article/downloadArticleFile.do?attachType=PDF&id=12769) · [DOI 10.16451/j.cnki.issn1003-6059.202408005](https://doi.org/10.16451/j.cnki.issn1003-6059.202408005) ## Summary Temporal knowledge graphs (TKGs) record facts as (head, relation, tail, timestamp) quadruples. They are very sparse: many entities sit in a long tail, and many events have little interaction history, so forecasting missing entities at future timestamps is unreliable. The authors propose NarrativeGraph, which converts graph history into text so that a large language model can apply its world knowledge. First, a rule-based module builds a "key event tree" for each query. It scores past quadruples by how often the head entity and the head–relation pair appear, how many distinct related entities exist, and how recent the latest event is, and it keeps only a compact set of relevant events. Second, a fine-tuned LLaMA generator rewrites these events as a chronological, causally coherent story. Third, a separately fine-tuned LLaMA reasoner reads the story together with the query and predicts the missing entity. On ICEWS14, YAGO and ICEWS18, the method reports strong results, especially on Hit@3 and Hit@10. Further analyses indicate gains for long-tail entities and sparse events. The work shows that narrative serialization is a workable bridge between structured graphs and LLMs. ## Key points - Framework name: NarrativeGraph. It has two modules: key event tree extraction and temporal story reasoning, which is split into a LLaMA generator and a LLaMA reasoner, each fine-tuned separately. - The event-selection weight is W = f·k / (t_q − t_latest + 1). Here f is quadruple frequency, k is the number of distinct related entities, and the +1 prevents division by zero. This filtering also addresses the limits of LLM context windows (the paper cites GPT-3's 2,048 tokens as an example). - The story generator is prompted with an instruction to link the quadruples into a narrative, adding historical background and supplementary knowledge. - Evaluation uses three public datasets: ICEWS14, YAGO and ICEWS18. The clearest improvements are on the Hit@3 and Hit@10 metrics. - Related work spans graph-based TKG models (TTransE, TComplEx, RE-NET, RE-GCN, CyGNet, TeMP) and LLM-based approaches (KG-BERT, PPT, GenTKG, Chain of History, GPT-NeoX in-context forecasting). *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* Fields: [Linguistics & languages](https://md.research.elamur.ai/free/shelves/linguistics/) · [Natural language processing](https://md.research.elamur.ai/free/shelves/nlp/) ## Cite The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{2024, title = {{故事启发大语言模型的时序知识图谱预测}}, author = {陈娟 and 赵新潮 and 隋京言 and 祁麟 and 田辰 and 庞亮 and 方金云}, booktitle = {模式识别与人工智能 (Pattern Recognition and Artificial Intelligence)}, year = {2024}, doi = {10.16451/j.cnki.issn1003-6059.202408005}, url = {https://doi.org/10.16451/j.cnki.issn1003-6059.202408005} } ``` - [← previous 探索中文预训练模型的混合粒度编码和IDF遮蔽](https://md.research.elamur.ai/free/papers/shao-2024-exploring-chinese-pre/) - [next → 文档级神经机器翻译综述](https://md.research.elamur.ai/free/papers/lv-2024-document-level-neural/) - [↑ back to papers of 2024](https://md.research.elamur.ai/free/papers/2024/page/4/) ## References 5 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. 历史相关四元组分布如下: [Tech Corp,Announce partnership,Research Lab,2023-01-10] [find](https://scholar.google.com/scholar?q=%E5%8E%86%E5%8F%B2%E7%9B%B8%E5%85%B3%E5%9B%9B%E5%85%83%E7%BB%84%E5%88%86%E5%B8%83%E5%A6%82%E4%B8%8B%3A+%5BTech+Corp%2CAnnounce+partnership%2CResearch+Lab%2C2023%2D01%2D10%5D) 2. [ Tech Corp,Host conference,Industry Group,2023-01-20] [find](https://scholar.google.com/scholar?q=%5B+Tech+Corp%2CHost+conference%2CIndustry+Group%2C2023%2D01%2D20%5D) 3. [ Tech Corp,Secure funding,Venture Fund,2023-02-05] [find](https://scholar.google.com/scholar?q=%5B+Tech+Corp%2CSecure+funding%2CVenture+Fund%2C2023%2D02%2D05%5D) 4. [ Tech Corp,Expand market,New Region,2023-03-12] [find](https://scholar.google.com/scholar?q=%5B+Tech+Corp%2CExpand+market%2CNew+Region%2C2023%2D03%2D12%5D) 5. [ Tech Corp,Acquire,Small Startup,2023-04-18] [find](https://scholar.google.com/scholar?q=%5B+Tech+Corp%2CAcquire%2CSmall+Startup%2C2023%2D04%2D18%5D)