{"id":"W7120522400","doi":"","title":"Evaluation of a chatbot in the educational context : an experience with Metis","year":2018,"lang":"pt","type":"article","venue":"Lume (Universidade Federal do Rio Grande do Sul)","topic":"AI in Service Interactions","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Metis; Chatbot; Context (archaeology); Task (project management); Knowledge base","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00145129,0.0003377907,0.0003451146,0.0004050017,0.000754548,0.0007033349,0.001832163,0.0001576873,0.001894249],"category_scores_gemma":[0.00009606878,0.000290063,0.000120667,0.001491812,0.0005353304,0.002845866,0.0001923734,0.0003943425,0.0001225199],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000382979,"about_ca_system_score_gemma":0.0008149038,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004241403,"about_ca_topic_score_gemma":0.02243908,"domain_scores_codex":[0.9954148,0.001052218,0.0004186318,0.0008084032,0.001809553,0.0004963711],"domain_scores_gemma":[0.996493,0.0002857978,0.0003992768,0.001093091,0.001567154,0.0001617054],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.0006844233,0.001883646,0.02378415,0.00008313412,0.000312298,0.0000382567,0.8433511,0.0004881016,0.00108809,0.05237236,0.002729698,0.07318481],"study_design_scores_gemma":[0.00892426,0.003296394,0.4195624,0.001103063,0.0004819509,0.0004180834,0.3925087,0.1545694,0.0008014627,0.01017429,0.006556961,0.001603066],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9752076,0.0003342577,0.0059734,0.009439276,0.0009285611,0.0009273879,0.00002371946,0.00003378457,0.007132074],"genre_scores_gemma":[0.9973733,0.0000176181,0.0008780652,0.0005960217,0.0002227219,0.00005390493,0.00001839261,0.00002087789,0.0008190546],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4508424,"threshold_uncertainty_score":0.9999552,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04462012755661569,"score_gpt":0.3179005789250567,"score_spread":0.273280451368441,"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."}}