{"id":"W7117513518","doi":"10.1109/icprs66293.2025.11302825","title":"Embedding Confidence to Enhance Trust in AI Document Entity Extraction","year":2025,"lang":"","type":"article","venue":"","topic":"Topic Modeling","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Springboard (Canada)","funders":"","keywords":"Flagging; Reliability (semiconductor); Workflow; Embedding; Matching (statistics); Information extraction; Data extraction","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":[0.01322336,0.001056918,0.001160131,0.00454696,0.0009788194,0.00477767,0.001741807,0.001858577,0.002413434],"category_scores_gemma":[0.1548654,0.0006552088,0.0009344438,0.003019845,0.001175769,0.01029726,0.00536916,0.002452208,0.001726814],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001340381,"about_ca_system_score_gemma":0.001991202,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00315348,"about_ca_topic_score_gemma":0.004099855,"domain_scores_codex":[0.9871197,0.005426324,0.0016121,0.002241593,0.003074576,0.000525757],"domain_scores_gemma":[0.8825613,0.07743977,0.01133157,0.01761759,0.01020679,0.0008430373],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0011161,0.0002394884,0.05222215,0.001309528,0.0003209591,0.000516523,0.003826812,0.07681642,0.01665528,0.02508673,0.01258312,0.8093069],"study_design_scores_gemma":[0.0000558801,0.0002301787,0.008060283,0.0002198076,0.0001435826,0.0004738136,0.0006508986,0.8980799,0.02795819,0.05171341,0.01229208,0.0001218783],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1200233,0.002151256,0.8605075,0.001849613,0.0001509726,0.000251518,0.002149491,0.009279287,0.003637009],"genre_scores_gemma":[0.8051624,0.000400916,0.1896352,0.0001962518,0.0001252206,0.0001179562,0.002497791,0.0005271211,0.001337266],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01322336,"threshold_uncertainty_score":0.06993258,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01357400971565047,"score_gpt":0.3583434044234295,"score_spread":0.3447693947077791,"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."}}