{"id":"W4408612569","doi":"10.2139/ssrn.5184646","title":"Leveraging Open-Source Large Language Models (Llms) to Enhance Data Extraction in Scoping Reviews: A Case Study on Disability and Ai Applications","year":2025,"lang":"en","type":"preprint","venue":"SSRN Electronic Journal","topic":"Topic Modeling","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Université de Montréal","funders":"","keywords":"Open source; Computer science; Data source; Data science; Data extraction; Natural language processing; World Wide Web; Information retrieval; Political science; MEDLINE; Programming language","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07584662,0.001518032,0.00148549,0.0159493,0.001717393,0.007840084,0.002760399,0.002076274,0.004130813],"category_scores_gemma":[0.2855896,0.001078664,0.003887077,0.01373916,0.000882341,0.008197162,0.006214406,0.002802026,0.002626271],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002545128,"about_ca_system_score_gemma":0.01464663,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005545295,"about_ca_topic_score_gemma":0.02355959,"domain_scores_codex":[0.9379632,0.04305607,0.01075397,0.003340547,0.004469229,0.0004170639],"domain_scores_gemma":[0.4843703,0.4593636,0.0170668,0.02071762,0.01721514,0.00126663],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0009122865,0.0004248694,0.01762564,0.05592512,0.002190363,0.001915876,0.02544137,0.01138514,0.01190083,0.022147,0.04308958,0.8070419],"study_design_scores_gemma":[0.0006854617,0.0008344403,0.02156091,0.03755267,0.008478016,0.002463537,0.01777458,0.1394438,0.04101382,0.1774689,0.5518227,0.0009011264],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.04464655,0.01672044,0.851549,0.01475785,0.0006508614,0.00673887,0.03869855,0.01519557,0.01104221],"genre_scores_gemma":[0.1238262,0.005090553,0.8411294,0.001180064,0.0001836211,0.003603847,0.02155143,0.001506708,0.001928203],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9241534,"threshold_uncertainty_score":0.4011201,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06356386152874574,"score_gpt":0.4121062181455009,"score_spread":0.3485423566167551,"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."}}