Cluster-Aware Dual-Level Test Specification Generation for Large-Scale Automotive Software Requirements

2026-06-17 · Source: cs.SE updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

A novel "Cluster-then-Summarize" pipeline has been developed to automate test specification generation for large-scale automotive software requirements, addressing the manual effort and scalability issues in meeting Automotive SPICE SWE.6 and ISO 26262 standards. This three-stage pipeline first embeds requirements using all-MiniLM-L6-v2 sentence transformers, then groups them via UMAP dimensionality reduction and HDBSCAN clustering with an adaptive "min_cluster_size" selection based on Silhouette and Calinski–Harabasz scores. Next, a multi-level map-reduce summarization algorithm, using a batch size of 10 and merge factor of 3, distills each cluster into concise, domain-conformant descriptions, preserving quantitative thresholds and safety integrity levels. Finally, it generates dual-level test specifications—individual requirement verification and cluster-level integration tests—by leveraging cluster topology, nearby-cluster context, and RAG grounded in industry standards. Evaluation across seven automotive datasets demonstrates improved integration test coverage, enhanced summarization fidelity (ROUGE-L 0.3793, BERTScore 0.8908), and higher test quality, with 89.59% overall faithfulness.

Key takeaway

For Machine Learning Engineers tasked with automating ASPICE SWE.6 compliance for large automotive software requirement sets, you should consider implementing a cluster-aware dual-level test generation pipeline. This approach significantly improves integration test coverage and summarization fidelity by leveraging semantic clustering and multi-level summarization. It effectively addresses the scalability challenges of manual processes, ensuring robust and traceable test specifications for safety-critical systems.

Key insights

Clustering requirements before LLM summarization and dual-level test generation significantly improves coverage and fidelity for large-scale automotive software.

Principles

• Cluster topology can drive multi-level test generation.
• Adaptive clustering parameters enhance scalability.
• Map-reduce summarization maintains content fidelity.

Method

Embed requirements (all-MiniLM-L6-v2), reduce dimensions (UMAP), cluster (HDBSCAN with auto "min_cluster_size"), then apply multi-level map-reduce summarization. Generate individual and cluster-level tests with RAG and nearby-cluster context.

In practice

• Use UMAP+HDBSCAN for scalable requirement clustering.
• Implement map-reduce for LLM summarization of large text sets.
• Inject cluster context for improved LLM test generation.

Topics

• Automotive SPICE SWE.6
• Large Language Models
• Requirements Engineering
• HDBSCAN Clustering
• Test Specification Generation
• Retrieval-Augmented Generation

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Cluster-Aware Dual-Level Test Specification Generation for Large-Scale Automotive Software Requirements — A "Cluster-then-Summarize" pipeline automates scalable, context-aware test specification generation for automotive software requirements.
• Clustered Self-Assessment: A Simple yet Effective Method for Uncertainty Quantification in Large Language Models — The Clustered Self-Assessment method quantifies LLM uncertainty by structuring semantically clustered generations into multiple-choice questions for self-evaluation.
• Towards the Readability of LLM-Generated Codes through Multitask Representation Engineering — MRepE uses orthogonal steering vectors to enhance LLM code readability across multiple metrics while bounding correctness impact.
• Open-SWE-Traces: Advancing Dual-Mode Multilingual Distillation for Software Engineering Agents — Dual-mode multilingual distillation with diverse agentic trajectories significantly improves LLM performance in software engineering tasks.
• Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements — LLMs struggle with domain-standard structured generation requiring simultaneous syntax, vocabulary, and external validation.
• VeriMoA: A Mixture-of-Agents Framework for Spec-to-HDL Generation — VeriMoA enhances LLM-based HDL generation via quality-guided caching and multi-path intermediate representations, improving accuracy without training.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.SE updates on arXiv.org.