{"id":"W7117149526","doi":"10.1145/3756681.3756997","title":"Large Language Models for API Classification: An Explorative Study","year":2025,"lang":"","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université Laval; Université du Québec à Rimouski","funders":"","keywords":"Task (project management); Function (biology); Context (archaeology); Reliability (semiconductor); Software; Resilience (materials science)","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":[0.01760576,0.002205586,0.001058212,0.003681617,0.00121223,0.004274899,0.002270268,0.001556867,0.00261375],"category_scores_gemma":[0.07110538,0.0007808607,0.003185943,0.002927184,0.001196029,0.007180093,0.003111678,0.00492459,0.001844732],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002104456,"about_ca_system_score_gemma":0.001864457,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008428107,"about_ca_topic_score_gemma":0.01089055,"domain_scores_codex":[0.9786592,0.01584646,0.001104672,0.00208489,0.001785457,0.0005194232],"domain_scores_gemma":[0.8081231,0.1777624,0.003699906,0.006059986,0.003565665,0.0007890712],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.003014973,0.007612511,0.2495974,0.005666952,0.002180941,0.002820928,0.01788783,0.07552893,0.01105822,0.01397575,0.05599625,0.5546593],"study_design_scores_gemma":[0.0004163556,0.001235334,0.05742536,0.001184718,0.001010681,0.001872111,0.0106485,0.8552017,0.008747089,0.02368993,0.03825413,0.0003141282],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7997595,0.007161064,0.1643952,0.003278605,0.0003442732,0.001290752,0.01050201,0.005246301,0.008022294],"genre_scores_gemma":[0.8524739,0.001066013,0.1227659,0.0006443956,0.0001572611,0.001275714,0.01912162,0.0008401603,0.001655053],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01760576,"threshold_uncertainty_score":0.09310925,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08677487955050701,"score_gpt":0.3737799482582607,"score_spread":0.2870050687077537,"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."}}