{"id":"W4200387115","doi":"10.18280/ria.350609","title":"Automatic Short Answer Grading System in Indonesian Language Using BERT Machine Learning","year":2021,"lang":"en","type":"article","venue":"Revue d intelligence artificielle","topic":"Edcuational Technology Systems","field":"Computer Science","cited_by":14,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Computer science; Confusion matrix; Confusion; Grading (engineering); Artificial intelligence; Natural language processing; Indonesian; Machine learning; Engineering; Linguistics","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002694779,0.0006378569,0.0006118477,0.00219034,0.0004264633,0.001074858,0.0009151581,0.0005253947,0.007272304],"category_scores_gemma":[0.01040622,0.0001783917,0.0003673105,0.001013465,0.0002459614,0.001294246,0.0008213221,0.0005663961,0.00337444],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000741743,"about_ca_system_score_gemma":0.001152958,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002335974,"about_ca_topic_score_gemma":0.00224444,"domain_scores_codex":[0.9965985,0.001165452,0.0004418515,0.0005112415,0.001137306,0.0001456073],"domain_scores_gemma":[0.9903923,0.003186999,0.0007840936,0.0007543705,0.004440075,0.000442282],"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":[0.001541404,0.0007457417,0.01791975,0.0008073366,0.00006454053,0.0005109337,0.001122043,0.005718043,0.05517716,0.003039733,0.03275369,0.8805996],"study_design_scores_gemma":[0.0004075653,0.002568363,0.1056378,0.0004928687,0.0002168298,0.001949732,0.002068051,0.5620289,0.1957549,0.009258482,0.1191137,0.000502809],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3506163,0.0005918854,0.5495306,0.0008593062,0.0004881925,0.002027676,0.004559245,0.06705473,0.02427216],"genre_scores_gemma":[0.6265255,0.0002433752,0.3489569,0.0002146096,0.0001210998,0.001008318,0.005107327,0.0005669988,0.01725604],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.007272304,"threshold_uncertainty_score":0.02432829,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03773554912136949,"score_gpt":0.2826640071354186,"score_spread":0.2449284580140491,"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."}}