{"id":"W4409166237","doi":"10.1007/978-3-031-88720-8_51","title":"Eval4RAG: Workshop on Evaluation of Retrieval-Augmented Generation Systems","year":2025,"lang":"en","type":"book-chapter","venue":"Lecture notes in computer science","topic":"Topic Modeling","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Waterloo","funders":"","keywords":"Computer science; Information retrieval; Artificial intelligence","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01692415,0.003225888,0.002670664,0.002607381,0.0009810925,0.004576709,0.006885929,0.003550638,0.03924872],"category_scores_gemma":[0.02580345,0.0009275952,0.001611261,0.001956866,0.001222712,0.005013062,0.004044524,0.002346106,0.009421802],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002660233,"about_ca_system_score_gemma":0.002192268,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01206984,"about_ca_topic_score_gemma":0.01005167,"domain_scores_codex":[0.9790847,0.01258292,0.0007430185,0.00173354,0.005089956,0.0007657788],"domain_scores_gemma":[0.9875311,0.006165009,0.0002422391,0.002852242,0.002859091,0.0003502263],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.006632835,0.001222525,0.001993864,0.002016799,0.00075501,0.0002823333,0.0003759854,0.04191178,0.01556437,0.01309644,0.2018331,0.7143151],"study_design_scores_gemma":[0.002651055,0.003378702,0.005405526,0.0007002144,0.0008854275,0.0005970608,0.0005707527,0.6429775,0.07915612,0.03737202,0.2260303,0.0002754279],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"other","genre_scores_codex":[0.1091558,0.02924948,0.6485755,0.004566624,0.006053441,0.002956212,0.03046408,0.0866713,0.08230758],"genre_scores_gemma":[0.3522904,0.003339604,0.5032074,0.001615132,0.0007881462,0.001532582,0.06075885,0.009415038,0.06705274],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.03924872,"threshold_uncertainty_score":0.1313,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06232949037925545,"score_gpt":0.3027192913743151,"score_spread":0.2403898009950596,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}