{"id":"W1862540101","doi":"10.19173/irrodl.v15i5.1857","title":"Evaluating the validity and applicability of automated essay scoring in two massive open online courses","year":2014,"lang":"en","type":"article","venue":"The International Review of Research in Open and Distributed Learning","topic":"Online Learning and Analytics","field":"Computer Science","cited_by":56,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Bill and Melinda Gates Foundation","keywords":"Rubric; Grading (engineering); Massive open online course; Higher education; Computer science; Mathematics education; Open education; Distance education; Psychology; Multimedia; Medical education; World Wide Web; Engineering; Medicine","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":[0.0531789,0.0003991509,0.0005310235,0.003008997,0.001123859,0.002962667,0.001457143,0.000940179,0.001061448],"category_scores_gemma":[0.2123565,0.0003860561,0.0006396472,0.002560355,0.001352084,0.001686445,0.002482405,0.001077756,0.0009238577],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001703938,"about_ca_system_score_gemma":0.002907951,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006004626,"about_ca_topic_score_gemma":0.00857311,"domain_scores_codex":[0.9553226,0.02240283,0.004199838,0.004341316,0.01285978,0.0008737181],"domain_scores_gemma":[0.7080337,0.165226,0.02999964,0.01427257,0.07828008,0.004187985],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002032022,0.002711686,0.6767088,0.0004292965,0.0002397203,0.0001794448,0.0109319,0.0009196881,0.003382745,0.001020464,0.001661755,0.2997825],"study_design_scores_gemma":[0.0003676304,0.006381809,0.9519693,0.0005789393,0.0002106331,0.0003018545,0.007478297,0.01473303,0.008553705,0.001410948,0.007886767,0.0001269618],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9892022,0.0003603577,0.003902683,0.0002159297,0.0001466781,0.0005067079,0.000121327,0.00006867814,0.005475452],"genre_scores_gemma":[0.9872185,0.0002978586,0.009480501,0.0002015879,0.0000983663,0.0006020982,0.0003146704,0.00003649125,0.001749918],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0531789,"threshold_uncertainty_score":0.2812403,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1986828325979672,"score_gpt":0.5550126832509729,"score_spread":0.3563298506530057,"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."}}