LLM-Based Goal Extraction in Requirements Engineering: Prompting Strategies and Limitations

A new arXiv preprint (2604.22207) proposes a chain of large language models (LLMs) to automate goal-oriented requirements engineering (GORE) from software documentation. The pipeline involves three phases: actor identification, high-level goal extraction, and low-level goal extraction. The approach uses engineered prompts with in-context learning variants and a generation-critic feedback loop between two LLMs. Experiments measured similarity between input data and in-context examples to assess impact. The pipeline achieved 61% accuracy in low-level goal identification, indicating the approach is best suited as a supportive tool rather than a fully automated solution.