In agile development, user stories are used to express user requirements, which employ simple natural language sentences. As software scales grow and evolve continuously, managing user stories becomes increasingly challenging, necessitating the modeling of user stories to clarify their associated semantics. Currently, representing user stories as goal models is an effective approach. However, extracting the semantics of the goal models from user stories is a time-consuming and labor-intensive task. In particular, abstracting low-level user stories into high-level goals remains difficult and relies heavily on the expertise of requirements engineers. This paper leverages the knowledge extraction and comprehension capabilities of large language models (LLMs) to divide the goal model extraction task into two categories: explicit concept extraction and implicit concept extraction. This classification is based on direct extraction from the user story field (explicit) and extraction according to implicit semantics (implicit). A multi-strategy prompting method (MSPM) is devised, which combines prompt templates, few-shot learning, ReAct, and chain-of-thought techniques to suit explicit and implicit goal model concept extraction, ultimately generating iStar models. The experimental results show that the proposed method has an average F 1 scores of 97.9% and 92.7% for the nodes and relationships in the extraction of the goal model, and has advantages in the completeness of the extraction of the goal modeling element.

Manually extracting device knowledge from dense hardware manuals for requirements elicitation is a timeconsuming and error-prone bottleneck in embedded systems development. While Large Language Models (LLMs) offer automation potential, their direct application is unreliable for this high-precision task due to factual inaccuracies and sensitivity to prompt engineering. To address this, we propose a novel hybrid method that synergizes parameter-efficient tuning with the reasoning power of LLMs. Instead of directly tuning the LLM, our approach uses soft prompts to efficiently optimize a lightweight sentence encoder, transforming it into a domainaware semantic retriever. This specialized retriever then generates superior, context-rich prompts for a frozen LLM. Our optimization strategy features a multi-task contrastive loss to discern fine-grained semantics and a clustering-based exemplar selection process to ensure prompt diversity and relevance. On a real-world dataset of devices, our method achieves F1-scores of 87.02% for entity extraction and 77.60% for relation extraction, significantly outperforming baseline LLMs and a strong, heavyweight, domain-specific pre-trained model (EquipBERT).

Modern software systems often involve diverse requirements from multiple stakeholders, leading to potential inconsistencies such as terminological ambiguities, contextual contradictions, and logical conflicts. Identifying these inconsistencies is critical for requirements engineering, but remains challenging due to the need for context-aware semantic analysis across diversity specifications. This paper proposes a hybrid framework that combines large-language models (LLMs) and graph-based merging algorithm to address this challenge. Our methodology first transforms textual requirements into graph-based models, explicitly capturing semantic relationships and contextual dependencies. A multi-graph merging algorithm then unifies these stakeholder-specific graphs into a merged representation, enabling systematic inconsistency detection through structural and semantic analysis. Using the contextual understanding capabilities of LLMs, the framework automates two phases: (1) disambiguation of natural language requirements during graph construction and (2) interpreting detected conflicts through explainable reasoning. We validate the framework through 3 case studies. Experimental results show the practical value of combining graph merging with LLM-powered semantic processing. This work advances requirements engineering by providing a scalable solution for inconsistency detection in diversity-intensive environments.

User stories are the main artifacts maintained in agile development and play an important role in describing requirements and making development plans. Writing good user stories becomes a key part of improving agile development. However, effective storytelling requires clear communication, reduced ambiguity, and enhanced testability. How to write good user stories is a challenge for beginners. In order to increase learner immersion and improve user story writing skills, this paper explores a gamification approach. Specifically, we started by creating specifications for beginners to write good user stories. Then, the process of user story writing is broken down, the gamification elements are refined, and the gamification teaching framework Story Explorer is proposed. In a course with 64 students, we assessed learning engagement and story quality. The results showed that the quality of user stories improved by an average of 11.8%, while students demonstrated a high level of engagement. Using Story Explorer significantly increased students' motivation to write high-quality user stories and the quality of their results.