{"id":"W4321061955","doi":"10.1109/apsec57359.2022.00088","title":"A new measure to assess the systematicity of the abstracts of reviews self-identifying as systematic reviews","year":2022,"lang":"en","type":"article","venue":"","topic":"Software Engineering Research","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University","funders":"","keywords":"Systematic review; Computer science; Quality (philosophy); Field (mathematics); Data science; Measure (data warehouse); Management science; Engineering ethics; MEDLINE; Epistemology; Political science; Data mining; 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":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3510544,0.004069116,0.007193475,0.06590802,0.00248445,0.008373046,0.003821934,0.00611957,0.005388203],"category_scores_gemma":[0.7717251,0.001901231,0.01563577,0.04531027,0.006279535,0.01722729,0.008098298,0.003867616,0.001371235],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006510852,"about_ca_system_score_gemma":0.0109892,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001602265,"about_ca_topic_score_gemma":0.002639398,"domain_scores_codex":[0.436382,0.2595616,0.1928256,0.01983451,0.08956627,0.001830074],"domain_scores_gemma":[0.104918,0.7267021,0.08577683,0.02830915,0.05269456,0.001599334],"domain_codex":null,"domain_gemma":"reporting","domain_candidate":"reporting","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.006357433,0.0006315893,0.1320906,0.1877839,0.05857655,0.0009317657,0.01108239,0.008777881,0.006707043,0.0367648,0.02437922,0.5259168],"study_design_scores_gemma":[0.009464694,0.009754476,0.2453103,0.07066958,0.109982,0.008767988,0.006622246,0.06968152,0.01677731,0.2769985,0.1725428,0.003428576],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.05793934,0.1776076,0.6620949,0.01005556,0.004571246,0.03164142,0.03233331,0.003448835,0.02030771],"genre_scores_gemma":[0.3716102,0.01891747,0.5518795,0.00278776,0.002326,0.04137562,0.009280882,0.0004667604,0.001355764],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6489456,"threshold_uncertainty_score":0.8002655,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1328783213091661,"score_gpt":0.3400922068514812,"score_spread":0.2072138855423151,"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."}}