{"id":"W2056825598","doi":"10.1145/2538862.2538966","title":"Identifying challenging CS1 concepts in a large problem dataset","year":2014,"lang":"en","type":"article","venue":"","topic":"Teaching and Learning Programming","field":"Computer Science","cited_by":54,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Computer science; Code (set theory); Data science; Term (time); Information retrieval; Mathematics education; Programming language; Psychology","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":[0.002535093,0.0009732734,0.0005980831,0.003760544,0.0009774594,0.001264772,0.001144603,0.001790121,0.002482347],"category_scores_gemma":[0.02130316,0.0002075961,0.0006373398,0.004092277,0.0005843861,0.001173231,0.00185007,0.001614523,0.001795266],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006930352,"about_ca_system_score_gemma":0.001112168,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003317997,"about_ca_topic_score_gemma":0.007094427,"domain_scores_codex":[0.9962882,0.00125798,0.000365423,0.0007964385,0.000990076,0.0003019679],"domain_scores_gemma":[0.9789429,0.01269668,0.002151605,0.002243846,0.002445745,0.001519157],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001659881,0.00447216,0.5314105,0.002843082,0.0005397061,0.002314837,0.006301424,0.02010511,0.01289634,0.002510222,0.1485573,0.2663894],"study_design_scores_gemma":[0.0004112287,0.001585139,0.7484324,0.0002764985,0.0001277308,0.00368418,0.009464681,0.07595515,0.01507615,0.004320426,0.1404092,0.0002571794],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9664912,0.0004797963,0.003545519,0.0006115461,0.00009020671,0.0001622445,0.02589596,0.0005638968,0.00215961],"genre_scores_gemma":[0.8551624,0.0002511061,0.01886722,0.0002530234,0.000106393,0.0004477543,0.1218804,0.000139686,0.002891901],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003760544,"threshold_uncertainty_score":0.01340699,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02529105321376424,"score_gpt":0.315626244481743,"score_spread":0.2903351912679787,"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."}}