RAG Evaluation Dataset and Results

This dataset supports the findings of the study "Knowledge Base Scope Matters: A Comparative Study of Retrieval-Augmented Generation in Academic Library Collection Development". It includes:

Metadata of 525 English-language academic monographs classified under CLC class P (Astronomy and Earth Sciences), including title, author, publisher, year, CLC code, and book description.

Full scoring results from three RAG configurations (NoRAG, RAG-Geo, RAG-RS), each containing a numerical quality score (0–100) and a textual rationale for each monograph.

Python source code for the automated evaluation pipeline, including retrieval, scoring, statistical analysis (correlation, descriptive statistics), and visualization (boxplots, heatmaps, dual-axis plots).

The knowledge base articles were sourced from the Web of Science Core Collection (SCIE). Due to licensing restrictions, only the article DOIs can be provided upon request. The Qwen3-8B model and nomic-embed-text model are open-weight and publicly accessible.