AutoSecQA: RAG-based Automotive Cybersecurity Answers Dataset

This dataset provides a collection of model-generated answers for question answering in the automotive cybersecurity domain, produced using multiple retrieval-augmented generation (RAG) configurations. It is designed to support research on domain-specific question answering and the evaluation of large language models under different retrieval strategies.

The dataset includes answers generated by several language models: GPT-4o, Gemma-2-9B-IT, LLaMA 3.1 8B Instruct, Mistral 7B Instruct v0.3, and Zephyr 7B Beta, across four RAG configurations: ensemble, semantic, syntactic, and two-stage. This structure enables systematic comparison of model behavior under varying retrieval setups.

Each configuration is organized into separate folders, where CSV files correspond to individual models. All files share a common schema consisting of: q_id (question identifier), Question (input query), and Answer (model-generated response).

Further details on the data collection process, RAG configurations, and evaluation methodology will be available in the associated publication.