ReLUNMF+AC — PubMed corpora, topic factors, and LLM-judge prompt library (Anonymous TKDE Submission)

Reproducibility artefacts accompanying an anonymised submission to IEEE Transactions on Knowledge and Data Engineering, titled A Fast, Multilingual-Capable Factorization Pipeline for Embedding-Based Topic Modeling. The archive bundles: Four PubMed SQLite corpora used in the headline evaluation: nutri (~530k abstracts), oncology (~50k), cardio (~100k), and precision (~416k). Title and abstract text are public-domain PubMed records fetched through the NCBI E-utilities. A single tar.gz archive of the literal LLM-judge prompt templates used in the single-language and multilingual evaluation protocols (rating, easy and hard word-intrusion, and Cross-Lingual Concept Alignment). The open-source code repository that consumes these artefacts is hosted privately during the double-anonymous review window and will be made public at deanonymisation. Author identification is also withheld until then.