一句话：提出基于 ISO 能力进阶与 AI 生命周期的知识与推理整合框架，减少 LLM 生成软件制品时对用户提示的依赖。
本文为一项博士研究，针对 LLM 辅助软件制品生成的两大局限——缺乏领域专门知识、以及推理调控几乎完全依赖用户提示——提出两个相互关联的框架。其一是概念框架，沿 DIKW（数据、信息、知识、智慧）层次结构化知识表示（KR）与推理策略（RS），并以 ISO/IEC 33020 的能力进阶为引导；其二是方法框架，将 ISO/IEC 5338 的 AI 生命周期细化为知识获取与推理模块化等活动，同时满足 ISO/IEC 25010 的质量要求。其目标是把既有标准专化为 LLM 生成软件制品的采纳路径，使命题化的领域知识与推理调控不再主要依赖提示交互。摘要未报告实证评估结果，当前贡献为框架层面的构建与标准映射。
Large Language Models (LLMs) are capable to assist software artifact construction, but current limitations are attributed to the lack of specialized domain knowledge and the inability to modulate reasoning, which is mostly supplied through user prompt interactions. This doctoral work introduces two interconnected frameworks to improve LLM effectiveness in software artifact generation: (1) a conceptual framework that structures Knowledge Representation (KR) and Reasoning Strategies (RS) along the Data, Information, Knowledge, Wisdom (DIKW) hierarchy, guided by ISO/IEC 33020 capability progression, and (2) a methodological framework that specialize the ISO/IEC 5338 AI lifecycle in activities for acquiring knowledge, module reasonihg, with ISO/IEC 25010 compliance. The contribution is to reduce the reliance on user prompts to articulate explicit domain knowledge and modulate reasoning, specializing established norms to adoption of LLMs to software artifacts generation.
@inproceedings{paperbot11670,
title = {Toward an ISO-Based Epistemic Capability Framework for Specializing LLMs Through Knowledge and Reasoning Integration for Software Artifact Generation},
author = {Helaine Lins and Clayton Rodrigues and Ivaldir de Farias Júnior},
booktitle = {ICSE},
year = {2026},
doi = {10.1145/3774748.3787653}
}