{"id":"W4407830444","doi":"10.1371/journal.pone.0318421","title":"Behind the mask: Random and selective masking in transformer models applied to specialized social science texts","year":2025,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Computational and Text Analysis Methods","field":"Social Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Western University","funders":"","keywords":"Masking (illustration); Computer science; Transformer; Classifier (UML); Artificial intelligence; Random forest; Coronavirus disease 2019 (COVID-19); Machine learning; Natural language processing; Speech recognition; Pattern recognition (psychology); Engineering; Medicine","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001778147,0.0000724066,0.0002115941,0.0002139419,0.0009035022,0.0001253451,0.0002141193,0.00003776944,0.00003364376],"category_scores_gemma":[0.000132845,0.00005789564,0.00003665045,0.001291966,0.0003347616,0.0001374242,0.00003221143,0.0001137479,0.000003233815],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000108925,"about_ca_system_score_gemma":0.000250949,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000308679,"about_ca_topic_score_gemma":0.00151813,"domain_scores_codex":[0.9986894,0.0001514428,0.0001634971,0.0002368632,0.0005184541,0.000240304],"domain_scores_gemma":[0.9994215,0.0003527475,0.00003156963,0.0000506592,0.00009621992,0.00004732683],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0005571936,0.0006986182,0.001717559,0.00002945632,0.0002701663,0.000001643824,0.2239031,0.000810997,0.01354549,0.6253543,0.00008184031,0.1330296],"study_design_scores_gemma":[0.004367248,0.00003486436,0.03097763,0.0001272009,0.0003903718,1.490694e-7,0.01025824,0.01407669,0.009006794,0.929035,0.001216588,0.0005092162],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5171419,0.0001048713,0.04508915,0.008190127,0.00005574275,0.001071181,0.000003472188,0.00004088737,0.4283027],"genre_scores_gemma":[0.9957618,0.00001726538,0.002881785,0.0003336946,0.0001251502,0.0000454804,5.536063e-7,0.0000032341,0.0008311038],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4786199,"threshold_uncertainty_score":0.69491,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05780898429034564,"score_gpt":0.3366705908533048,"score_spread":0.2788616065629591,"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."}}