{"id":"W6925486270","doi":"10.17615/1tt9-6558","title":"Appendices of the Conference Paper Titiled \"An Empirical Study Evaluating ChatGPT’s Performance in Generating Search Strategies for Systematic Reviews\"","year":2024,"lang":"en","type":"article","venue":"UNC Libraries","topic":"Chemical and Physical Properties in Aqueous Solutions","field":"Chemical Engineering","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Empirical research; Presentation (obstetrics); Systematic review; Association (psychology)","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02005052,0.00167206,0.002841426,0.01308183,0.002530326,0.005568977,0.002900993,0.002480665,0.6524628],"category_scores_gemma":[0.4434857,0.001595227,0.004057042,0.01690506,0.0008500532,0.004588237,0.005090129,0.004311142,0.207267],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004477508,"about_ca_system_score_gemma":0.006639009,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004712312,"about_ca_topic_score_gemma":0.005306008,"domain_scores_codex":[0.9814411,0.006820306,0.003631727,0.0009668182,0.006680465,0.0004594997],"domain_scores_gemma":[0.5793062,0.2936961,0.01324128,0.01324794,0.09516702,0.005341473],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.0001524916,0.00005158319,0.00006757533,0.001818067,0.0000292775,0.00002992287,0.00003121032,0.0002498368,0.00004346744,0.0004275294,0.9799887,0.01711047],"study_design_scores_gemma":[0.001691813,0.0004429132,0.003327663,0.007609653,0.0002156206,0.0002925599,0.0003273552,0.001668108,0.0005847542,0.008419702,0.9752134,0.0002064416],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"dataset","genre_gemma":"empirical","genre_scores_codex":[0.003532284,0.01230506,0.0348113,0.08425359,0.2035832,0.02896675,0.4717253,0.008185373,0.1526371],"genre_scores_gemma":[0.0276923,0.01504846,0.07434539,0.04555262,0.04916665,0.09348708,0.4204025,0.009087376,0.2652177],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9799495,"threshold_uncertainty_score":0.4957197,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1256230589761242,"score_gpt":0.3534634812365293,"score_spread":0.227840422260405,"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."}}