Xuan Liu , John Ahmet Erkoyuncu , Jerry Ying Hsi Fuh , Wen Feng Lu , Bingbing Li
{"title":"Knowledge extraction for additive manufacturing process via named entity recognition with LLMs","authors":"Xuan Liu , John Ahmet Erkoyuncu , Jerry Ying Hsi Fuh , Wen Feng Lu , Bingbing Li","doi":"10.1016/j.rcim.2024.102900","DOIUrl":null,"url":null,"abstract":"<div><div>This paper proposes a novel NER framework, leveraging the advanced capabilities of Large Language Models (LLMs), to address the limitations of manually defined taxonomy. Our framework integrates the expert knowledge internalized in both academic materials and LLMs through retrieval-augmented generation (RAG) to automatically customize taxonomies for specific manufacturing processes and adopts two distinct strategies of using LLMs — In-Context Learning (ICL) and fine-tuning to complete manufacturing NER tasks with minimal training data. We demonstrate the framework efficiency through its superior ability to define precise taxonomies, identify and classify process-level entities related to the most popular additive manufacturing process fused deposition modeling (FDM) as case study, achieving a high F1 score of 0.9192.</div></div>","PeriodicalId":21452,"journal":{"name":"Robotics and Computer-integrated Manufacturing","volume":"93 ","pages":"Article 102900"},"PeriodicalIF":9.1000,"publicationDate":"2024-11-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Robotics and Computer-integrated Manufacturing","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S073658452400187X","RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, INTERDISCIPLINARY APPLICATIONS","Score":null,"Total":0}
引用次数: 0
Abstract
This paper proposes a novel NER framework, leveraging the advanced capabilities of Large Language Models (LLMs), to address the limitations of manually defined taxonomy. Our framework integrates the expert knowledge internalized in both academic materials and LLMs through retrieval-augmented generation (RAG) to automatically customize taxonomies for specific manufacturing processes and adopts two distinct strategies of using LLMs — In-Context Learning (ICL) and fine-tuning to complete manufacturing NER tasks with minimal training data. We demonstrate the framework efficiency through its superior ability to define precise taxonomies, identify and classify process-level entities related to the most popular additive manufacturing process fused deposition modeling (FDM) as case study, achieving a high F1 score of 0.9192.
本文提出了一种新颖的 NER 框架，利用大型语言模型 (LLM) 的先进功能来解决人工定义分类法的局限性。我们的框架通过检索增强生成（RAG）整合了学术材料和 LLM 中内化的专家知识，为特定的制造流程自动定制分类标准，并采用两种不同的 LLM 使用策略--上下文学习（ICL）和微调，以最少的训练数据完成制造 NER 任务。我们以最流行的增材制造工艺熔融沉积建模（FDM）为案例，展示了该框架在定义精确分类标准、识别和分类与该工艺相关的工艺级实体方面的卓越能力，并取得了 0.9192 的高 F1 分数，从而证明了该框架的高效性。
期刊介绍：
The journal, Robotics and Computer-Integrated Manufacturing, focuses on sharing research applications that contribute to the development of new or enhanced robotics, manufacturing technologies, and innovative manufacturing strategies that are relevant to industry. Papers that combine theory and experimental validation are preferred, while review papers on current robotics and manufacturing issues are also considered. However, papers on traditional machining processes, modeling and simulation, supply chain management, and resource optimization are generally not within the scope of the journal, as there are more appropriate journals for these topics. Similarly, papers that are overly theoretical or mathematical will be directed to other suitable journals. The journal welcomes original papers in areas such as industrial robotics, human-robot collaboration in manufacturing, cloud-based manufacturing, cyber-physical production systems, big data analytics in manufacturing, smart mechatronics, machine learning, adaptive and sustainable manufacturing, and other fields involving unique manufacturing technologies.
京公网安备 11010802042870号
京ICP备2023020795号-1
分享
求助内容：
应助结果提醒方式：
扫码关注我们
