Sada-RealMetaRFANN Paper-RAG v1.1
Abstract
Sada-RealMetaRFANN Paper-RAG v1.1 is a real-metadata benchmark for range-filtered approximate nearest neighbor retrieval in academic Paper-RAG settings. The dataset is derived from publicly available arXiv metadata and is designed to evaluate filtered vector search under realistic metadata conditions. Each paper record is represented by a 384-dimensional sentence-transformer embedding generated from the paper title and abstract. The benchmark includes real temporal metadata, shuffled same-distribution controls, uniform-random controls, balanced selectivity buckets, RFANN query sets, and exact top-10 filtered ground truth. The motivation is to support reproducible research on metadata-aware vector retrieval. Many filtered ANN and RFANN evaluations rely on synthetic or random numerical attributes when real metadata is unavailable. Sada-RealMetaRFANN Paper-RAG v1.1 provides a controlled alternative by pairing paper embeddings with real publication-time metadata and counterfactual metadata variants. This release is intended for research on filtered vector search, retrieval-augmented generation, vector databases, hybrid retrieval, selectivity-aware retrieval, and memory-conscious retrieval evaluation.
Read the paper
Related studies
- Knowledge Management Perspective of Generative Artificial Intelligence
- Navigating Generative Artificial Intelligence Promises and Perils for Knowledge and Creative Work
- Characterizing the evolving research on enterprise content management
- Large language models and multimodal foundation models for precision oncology
- From Generative AI to Extended Reality: Multidisciplinary Perspectives on the Challenges, Opportunities, and Future of Educational Computing
- Generative Artificial Intelligence: A New Engine for Advancing Environmental Science and Engineering
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.