arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2501.03067cs.CR
收藏

构建医疗物联网安全需求本体的工具设计与实现

Design and implementation of tools to build an ontology of Security Requirements for Internet of Medical Things

Daniel Naro, Jaime Delgado, Silvia Llorente, Amanda Palomo

AI总结:

本文面向欧洲 IoMT 合规场景,设计从规范文本到 XML/XSD 再到本体的半自动构建工具,并评估 LLM 合并相似实例的辅助效果。

摘要:

在为医疗物联网(IoMT)领域开发设备、架构和服务时,制造商或集成商必须了解法律和规范所提出的安全需求。为了提供引导这些需求的工具,并向第三方确保合规性得到正确满足,构建一个梳理相关法律和规范(针对欧洲语境)的本体非常有用。我们在此研究该本体的开发。由于所考虑的规范文件数量和篇幅都非常庞大,我们建立了一套方法论和工具,以简化从自然文本到本体的转换。第一步是在语料库中手动高亮相关概念,随后手动转换为 XML/XSD。我们开发了一个工具,可将这种半结构化数据转换为本体。由于不同规范使用相似但不同的措辞,我们的方法倾向于在本体中创建相似实例。为了通过实例合并改进本体简化,我们考虑使用 LLM。将 LLM 的响应与我们手动定义的正确响应进行比较。自动化系统响应的质量并未证明足以被盲目信任,只应作为人工修正的起点。

英文摘要:

When developing devices, architectures and services for the Internet of Medical Things (IoMT) world, manufacturers or integrators must be aware of the security requirements expressed by both laws and specifications. To provide tools guiding through these requirements and to assure a third party of the correct compliance, an ontology charting the relevant laws and specifications (for the European context) is very useful. We here address the development of this ontology. Due to the very high number and size of the considered specification documents, we have put in place a methodology and tools to simplify the transition from natural text to an ontology. The first step is a manual highlighting of relevant concepts in the corpus, then a manual translation to XML/XSD is operated. We have developed a tool allowing us to convert this semi-structured data into an ontology. Because the different specifications use similar but different wording, our approach favors the creation of similar instances in the ontology. To improve the ontology simplification through instance merging, we consider the use of LLMs. The responses of the LLMs are compared against our manually defined correct responses. The quality of the responses of the automated system does not prove to be good enough to be trusted blindly, and should only be used as a starting point for a manual correction.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑