Query datasets and run results related to scientific articles about Semantic Embeddings and RAG for Sanctions Name Screening Resources
Year: 2026
Type: software
Abstract: Query datasets and run results related to scientific articles about Semantic Embeddings and RAG for Sanctions Name Screening Resources The artefacts: Embeddings related: directories: query-sets/ (14 query files and sdn_xml.zip, the OFAC snapshot); embeddings/: results/ (96 raw result files, 13,932 query records) and deep-k-probes/ (the k = 10 and k = 20 runs behind Table 10 and the probes Shoihu_UKR_{40,80}.json); metrics_recomputed.csv (authoritative point estimates); metrics_entity_ci.csv (Wilson and entity-cluster intervals, Table 3); paired_tests.csv (Tables 3 and 4); auc_ci.csv (Table 6); threshold_roc.csv and operating_points.csv (Table 7) with operating_points_heldout.csv (Table 8); rank_tail.csv (Table 10); latency_percentiles.csv; dataset_audit.csv; and plots/. RAG related: In directory rag/: analysis-rag-gemini-flash-3-6-medium-topk30/ and analysis-rag-gemini-flash-3-7-low-topk30-run-{01,02,03}/ — four configurations × 14 run directories, containing meta (full prompt, model, effort, token usage, timing), ground_truth, results[] of 30 candidates and rag_response. metrics_multirun.csv — per configuration × group: recall decomposition, alert rates, override and multi-match counts, set metrics, latency and token statistics. headline_ci.csv — headline rates with exact Clopper–Pearson and entity-cluster bootstrap intervals. stability_multirun.csv — per-query verdict and returned-set agreement across the three replicates. specificity_multirun.csv — paired 2-part vs 3-part comparison with entity-cluster bootstrap and distinct-entity counts. cases_multirun.csv — every non-trivial case in every configuration, with the adjudicator's rationale. plots/ — the four figures as PNG.
Authors: Pavlenko, Iegor, Hnatushenko, Volodymyr
Institution: Dnipro University of Technology
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204235281