{"id":"W4317565429","doi":"10.5281/zenodo.7552977","title":"AML: A Novel Accuracy Metric for Critical Evaluation of Log Parsing Techniques (results)","year":2023,"lang":"en","type":"paratext","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Natural Language Processing Techniques","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Concordia University","funders":"","keywords":"Metric (unit); Computer science; Parsing; Artificial intelligence; Engineering","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":[0.01310205,0.003128045,0.001825195,0.007322494,0.001041328,0.004519792,0.003036264,0.002798251,0.006735582],"category_scores_gemma":[0.05503656,0.0005067726,0.001360204,0.004459859,0.001200896,0.004523776,0.002561838,0.001581621,0.005705136],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00172677,"about_ca_system_score_gemma":0.001861141,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003369053,"about_ca_topic_score_gemma":0.003860001,"domain_scores_codex":[0.9730636,0.007968762,0.003625577,0.004126442,0.0100094,0.001206302],"domain_scores_gemma":[0.9464891,0.02952734,0.002999767,0.00857094,0.01154728,0.000865498],"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":[0.004767545,0.000945189,0.04178086,0.004722902,0.001652542,0.0005404874,0.0007819966,0.06532543,0.05100328,0.00796595,0.2202419,0.6002719],"study_design_scores_gemma":[0.0006837675,0.002821863,0.05298975,0.0004981258,0.001177941,0.001918138,0.001152906,0.5909499,0.2289562,0.02382703,0.09449378,0.0005305377],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1991889,0.01046656,0.5459684,0.00166138,0.002235659,0.001242688,0.06661114,0.1535606,0.01906469],"genre_scores_gemma":[0.5018263,0.0009890078,0.3866288,0.0005263913,0.0004582542,0.001477067,0.08938271,0.01176867,0.006942937],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.01310205,"threshold_uncertainty_score":0.06929111,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1002241049848907,"score_gpt":0.3688968886522967,"score_spread":0.268672783667406,"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."}}