A framework for Chinese event semantic constraint predicates and their generation using PEFT LLM and RAG
文献类型：期刊论文
| 作者 | Huang, Qiaojuan1,2; Wang, Shi1; He, Qing1; Cao, Cungen1 |
| 刊名 | INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS
 |
| 出版日期 | 2025-05-22 |
| 页码 | 26 |
| 关键词 | Event semantic analysis Head-tail event semantic constraint predicates Large language models Parameter-efficient fine-tuning Retrieval-augmented generation |
| ISSN号 | 1868-8071 |
| DOI | 10.1007/s13042-025-02658-1 |
| 英文摘要 | Event semantic analysis is a crucial area of research in natural language processing. It focuses on deeply understanding the semantics of events and their components. While existing semantic frameworks and event semantic labelling cover a wealth of semantic information, further research is needed to explore the fine-grained semantic relationships across participants in different events. To address this problem, we construct a head-tail event semantic constraint predicate framework that relates the complex semantics between different event components. However, considering the large-scale data, an automated method for generating constraint predicates is essential. Therefore, we combine parameter-efficient fine-tuning (PEFT) of large language models (LLMs) with retrieval-augmented generation (RAG) technologies. Specifically, this paper makes four significant contributions. Firstly, we construct a detailed framework of head-tail events semantic constraint predicates by combining manual definitions with the GPT-4o model. The framework includes 117 types of constraint predicates between head-tail events. Secondly, based on this framework, we annotate a dataset of head-tail events semantic constraint predicates and apply it to PEFT LLMs. Thirdly, we use PEFT LLMs prediction error data to construct an event knowledge base covering event entities, semantic roles, and entity types. Fourth, using the event knowledge base, the PEFT LLMs integrate the dual RAG techniques of keywords and vectors to improve the accuracy of generating the constraint predicates. Experimental results show that the method outperforms the baselines, achieving a precision of 97.18% and recall of 97.39%. This research fills the gap in semantic relationships between components across events and provides a novel approach for complex semantic generation tasks. |
| 资助项目 | National Key Research and Development Program of China[2022YFC3302300] ; National Key Research and Development Program of China[2022YFC3303302] ; National Key RD Plan[62476263] ; National Key RD Plan[U2436209] ; National Natural Science Foundation of China |
| WOS研究方向 | Computer Science |
| 语种 | 英语 |
| WOS记录号 | WOS:001493829900001 |
| 出版者 | SPRINGER HEIDELBERG |
| 源URL | [http://119.78.100.204/handle/2XEOYT63/42395]  |
| 专题 | 中国科学院计算技术研究所期刊论文_英文 |
| 通讯作者 | Wang, Shi |
| 作者单位 | 1.Chinese Acad Sci, Inst Comp Technol, Key Lab Intelligent Informat Proc, Beijing 100190, Peoples R China 2.Univ Chinese Acad Sci, Beijing 100049, Peoples R China |
| 推荐引用方式 GB/T 7714 | Huang, Qiaojuan,Wang, Shi,He, Qing,et al. A framework for Chinese event semantic constraint predicates and their generation using PEFT LLM and RAG[J]. INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS,2025:26. |
| APA | Huang, Qiaojuan,Wang, Shi,He, Qing,&Cao, Cungen.(2025).A framework for Chinese event semantic constraint predicates and their generation using PEFT LLM and RAG.INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS,26. |
| MLA | Huang, Qiaojuan,et al."A framework for Chinese event semantic constraint predicates and their generation using PEFT LLM and RAG".INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS (2025):26. |
入库方式： OAI收割
来源：计算技术研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。