LEREDD: LLM-Enabled Requirement Dependency Detection

General Information This repository contains the replication package for our paper: LEREDD: LLM-Enabled Requirement Dependency Detection. The paper proposes an LLM-based approach for the automated detection of direct dependencies from natural language requirements. This repository includes the manually annotated requirement datasets used for the evaluation, the source code for the implementations, and a comprehensive table presenting the results of the few-shot experiments without a threshold-based re-annotation. Annotated Datasets The repository includes a dataset of requirement pairs that are manually annotated for dependencies. The annotation covers three distinct automotive systems: ADB - Adaptive Driving Beam TJA - Traffic Jam Assist APA - Automated Parking Assist For each system, we provide the annotated requirement pairs in addition to the software requirement specification document. Implementations This repository includes the implementation of: Zero-shot prompting (RQ1) Few-shot prompting (RQ2-RQ3) Few-shot prompting and RAG (RQ2-RQ3) Baselines (RQ4-RQ5) The LEREDD approach (RQ4-RQ5) Results This repository includes a comprehensive table that presents the results of the few-shot experiments without a threshold-based re-annotation.